[ { "path": ".github/ISSUE_TEMPLATE/bug_report.md", "content": "---\nname: Bug report\nabout: Create a report to help us improve\ntitle: ''\nlabels: ''\nassignees: ''\n\n---\n\n**Describe the bug**\nA clear and concise description of what the bug is.\n\n**To Reproduce**\nSteps to reproduce the behavior:\n1. Go to '...'\n2. Click on '....'\n3. Scroll down to '....'\n4. See error\n\n**Expected behavior**\nA clear and concise description of what you expected to happen.\n\n**Screenshots**\nIf applicable, add screenshots to help explain your problem.\n\n**Desktop (please complete the following information):**\n - OS: [e.g. iOS]\n - Browser [e.g. chrome, safari]\n - Version [e.g. 22]\n\n**Smartphone (please complete the following information):**\n - Device: [e.g. iPhone6]\n - OS: [e.g. iOS8.1]\n - Browser [e.g. stock browser, safari]\n - Version [e.g. 22]\n\n**Additional context**\nAdd any other context about the problem here.\n" }, { "path": ".github/ISSUE_TEMPLATE/feature_request.md", "content": "---\nname: Feature request\nabout: Suggest an idea for this project\ntitle: ''\nlabels: ''\nassignees: ''\n\n---\n\n**Is your feature request related to a problem? Please describe.**\nA clear and concise description of what the problem is. Ex. I'm always frustrated when [...]\n\n**Describe the solution you'd like**\nA clear and concise description of what you want to happen.\n\n**Describe alternatives you've considered**\nA clear and concise description of any alternative solutions or features you've considered.\n\n**Additional context**\nAdd any other context or screenshots about the feature request here.\n" }, { "path": ".gitignore", "content": "# Byte-compiled / optimized / DLL files\n__pycache__/\n*.py[cod]\n*$py.class\n\n# C extensions\n*.so\n\n# Distribution / packaging\n.Python\nbuild/\ndevelop-eggs/\ndist/\ndownloads/\neggs/\n.eggs/\nlib/\nlib64/\nparts/\nsdist/\nvar/\nwheels/\nshare/python-wheels/\n*.egg-info/\n.installed.cfg\n*.egg\nMANIFEST\n\n\n# PyInstaller\n# Usually these files are written by a python script from a template\n# before PyInstaller builds the exe, so as to inject date/other infos into it.\n*.manifest\n*.spec\n\n# Installer logs\npip-log.txt\npip-delete-this-directory.txt\n\n# Unit test / coverage reports\nhtmlcov/\n.tox/\n.nox/\n.coverage\n.coverage.*\n.cache\nnosetests.xml\ncoverage.xml\n*.cover\n*.py,cover\n.hypothesis/\n.pytest_cache/\ncover/\n\n# Translations\n*.mo\n*.pot\n\n# Django stuff:\n*.log\nlocal_settings.py\ndb.sqlite3\ndb.sqlite3-journal\n\n# Flask stuff:\ninstance/\n.webassets-cache\n\n# Scrapy stuff:\n.scrapy\n\n# Sphinx documentation\ndocs/_build/\n\n# PyBuilder\n.pybuilder/\ntarget/\n\n# Jupyter Notebook\n.ipynb_checkpoints\n\n# IPython\nprofile_default/\nipython_config.py\n\n# pyenv\n# For a library or package, you might want to ignore these files since the code is\n# intended to run in multiple environments; otherwise, check them in:\n# .python-version\n\n# pipenv\n# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.\n# However, in case of collaboration, if having platform-specific dependencies or dependencies\n# having no cross-platform support, pipenv may install dependencies that don't work, or not\n# install all needed dependencies.\n#Pipfile.lock\n\n# poetry\n# Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.\n# This is especially recommended for binary packages to ensure reproducibility, and is more\n# commonly ignored for libraries.\n# https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control\n#poetry.lock\n\n# pdm\n# Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.\n#pdm.lock\n# pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it\n# in version control.\n# https://pdm.fming.dev/#use-with-ide\n.pdm.toml\n\n# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm\n__pypackages__/\n\n# Celery stuff\ncelerybeat-schedule\ncelerybeat.pid\n\n# SageMath parsed files\n*.sage.py\n\n# Environments\n.env\n.venv\nenv/\nvenv/\nENV/\nenv.bak/\nvenv.bak/\n\n# Spyder project settings\n.spyderproject\n.spyproject\n\n# Rope project settings\n.ropeproject\n\n# mkdocs documentation\n/site\n\n# mypy\n.mypy_cache/\n.dmypy.json\ndmypy.json\n\n# Pyre type checker\n.pyre/\n\n# pytype static type analyzer\n.pytype/\n\n# Cython debug symbols\ncython_debug/\n\n# PyCharm\n# JetBrains specific template is maintained in a separate JetBrains.gitignore that can\n# be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore\n# and can be added to the global gitignore or merged into this file. For a more nuclear\n# option (not recommended) you can uncomment the following to ignore the entire idea folder.\n.idea/\n\nwandb/\njobs/logs/\n*.out\n*ipynb\n.history/\n*.json\n*.sh\n.ipynb_common\nlogs/\nresults/\nprompts/\noutput/\nckpt/\ndivide_vqa.py\njobs/\n\n*.slurm\nslurm*\nsbatch_generate*\neval_data/\ndataset/Evaluation.md\njupyter_notebook.slurm\n" }, { "path": "CODE_OF_CONDUCT.md", "content": "# Contributor Covenant Code of Conduct\n\n## Our Pledge\n\nWe as members, contributors, and leaders pledge to make participation in our\ncommunity a harassment-free experience for everyone, regardless of age, body\nsize, visible or invisible disability, ethnicity, sex characteristics, gender\nidentity and expression, level of experience, education, socio-economic status,\nnationality, personal appearance, race, religion, or sexual identity\nand orientation.\n\nWe pledge to act and interact in ways that contribute to an open, welcoming,\ndiverse, inclusive, and healthy community.\n\n## Our Standards\n\nExamples of behavior that contributes to a positive environment for our\ncommunity include:\n\n* Demonstrating empathy and kindness toward other people\n* Being respectful of differing opinions, viewpoints, and experiences\n* Giving and gracefully accepting constructive feedback\n* Accepting responsibility and apologizing to those affected by our mistakes,\n and learning from the experience\n* Focusing on what is best not just for us as individuals, but for the\n overall community\n\nExamples of unacceptable behavior include:\n\n* The use of sexualized language or imagery, and sexual attention or\n advances of any kind\n* Trolling, insulting or derogatory comments, and personal or political attacks\n* Public or private harassment\n* Publishing others' private information, such as a physical or email\n address, without their explicit permission\n* Other conduct which could reasonably be considered inappropriate in a\n professional setting\n\n## Enforcement Responsibilities\n\nCommunity leaders are responsible for clarifying and enforcing our standards of\nacceptable behavior and will take appropriate and fair corrective action in\nresponse to any behavior that they deem inappropriate, threatening, offensive,\nor harmful.\n\nCommunity leaders have the right and responsibility to remove, edit, or reject\ncomments, commits, code, wiki edits, issues, and other contributions that are\nnot aligned to this Code of Conduct, and will communicate reasons for moderation\ndecisions when appropriate.\n\n## Scope\n\nThis Code of Conduct applies within all community spaces, and also applies when\nan individual is officially representing the community in public spaces.\nExamples of representing our community include using an official e-mail address,\nposting via an official social media account, or acting as an appointed\nrepresentative at an online or offline event.\n\n## Enforcement\n\nInstances of abusive, harassing, or otherwise unacceptable behavior may be\nreported to the community leaders responsible for enforcement at\nhttps://discord.gg/2aNvvYVv.\nAll complaints will be reviewed and investigated promptly and fairly.\n\nAll community leaders are obligated to respect the privacy and security of the\nreporter of any incident.\n\n## Enforcement Guidelines\n\nCommunity leaders will follow these Community Impact Guidelines in determining\nthe consequences for any action they deem in violation of this Code of Conduct:\n\n### 1. Correction\n\n**Community Impact**: Use of inappropriate language or other behavior deemed\nunprofessional or unwelcome in the community.\n\n**Consequence**: A private, written warning from community leaders, providing\nclarity around the nature of the violation and an explanation of why the\nbehavior was inappropriate. A public apology may be requested.\n\n### 2. Warning\n\n**Community Impact**: A violation through a single incident or series\nof actions.\n\n**Consequence**: A warning with consequences for continued behavior. No\ninteraction with the people involved, including unsolicited interaction with\nthose enforcing the Code of Conduct, for a specified period of time. This\nincludes avoiding interactions in community spaces as well as external channels\nlike social media. Violating these terms may lead to a temporary or\npermanent ban.\n\n### 3. Temporary Ban\n\n**Community Impact**: A serious violation of community standards, including\nsustained inappropriate behavior.\n\n**Consequence**: A temporary ban from any sort of interaction or public\ncommunication with the community for a specified period of time. No public or\nprivate interaction with the people involved, including unsolicited interaction\nwith those enforcing the Code of Conduct, is allowed during this period.\nViolating these terms may lead to a permanent ban.\n\n### 4. Permanent Ban\n\n**Community Impact**: Demonstrating a pattern of violation of community\nstandards, including sustained inappropriate behavior, harassment of an\nindividual, or aggression toward or disparagement of classes of individuals.\n\n**Consequence**: A permanent ban from any sort of public interaction within\nthe community.\n\n## Attribution\n\nThis Code of Conduct is adapted from the [Contributor Covenant][homepage],\nversion 2.0, available at\nhttps://www.contributor-covenant.org/version/2/0/code_of_conduct.html.\n\nCommunity Impact Guidelines were inspired by [Mozilla's code of conduct\nenforcement ladder](https://github.com/mozilla/diversity).\n\n[homepage]: https://www.contributor-covenant.org\n\nFor answers to common questions about this code of conduct, see the FAQ at\nhttps://www.contributor-covenant.org/faq. Translations are available at\nhttps://www.contributor-covenant.org/translations.\n" }, { "path": "LICENSE.md", "content": "BSD 3-Clause License\n\nCopyright 2023 Deyao Zhu\nAll rights reserved.\n\nRedistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:\n\n1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.\n\n2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.\n\n3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.\n\nTHIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS \"AS IS\" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\n" }, { "path": "LICENSE_Lavis.md", "content": "BSD 3-Clause License\n\nCopyright (c) 2022 Salesforce, Inc.\nAll rights reserved.\n\nRedistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:\n\n1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.\n\n2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.\n\n3. Neither the name of Salesforce.com nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.\n\nTHIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS \"AS IS\" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\n" }, { "path": "MiniGPT4_Train.md", "content": "## Training of MiniGPT-4\n\nThe training of MiniGPT-4 contains two alignment stages.\n\n**1. First pretraining stage**\n\nIn the first pretrained stage, the model is trained using image-text pairs from Laion and CC datasets\nto align the vision and language model. To download and prepare the datasets, please check \nour [first stage dataset preparation instruction](dataset/README_1_STAGE.md). \nAfter the first stage, the visual features are mapped and can be understood by the language\nmodel.\nTo launch the first stage training, run the following command. In our experiments, we use 4 A100. \nYou can change the save path in the config file \n[train_configs/minigpt4_stage1_pretrain.yaml](train_configs/minigpt4_stage1_pretrain.yaml)\n\n```bash\ntorchrun --nproc-per-node NUM_GPU train.py --cfg-path train_configs/minigpt4_stage1_pretrain.yaml\n```\n\nA MiniGPT-4 checkpoint with only stage one training can be downloaded \n[here (13B)](https://drive.google.com/file/d/1u9FRRBB3VovP1HxCAlpD9Lw4t4P6-Yq8/view?usp=share_link) or [here (7B)](https://drive.google.com/file/d/1HihQtCEXUyBM1i9DQbaK934wW3TZi-h5/view?usp=share_link).\nCompared to the model after stage two, this checkpoint generate incomplete and repeated sentences frequently.\n\n\n**2. Second finetuning stage**\n\nIn the second stage, we use a small high quality image-text pair dataset created by ourselves\nand convert it to a conversation format to further align MiniGPT-4.\nTo download and prepare our second stage dataset, please check our \n[second stage dataset preparation instruction](dataset/README_2_STAGE.md).\nTo launch the second stage alignment, \nfirst specify the path to the checkpoint file trained in stage 1 in \n[train_configs/minigpt4_stage1_pretrain.yaml](train_configs/minigpt4_stage2_finetune.yaml).\nYou can also specify the output path there. \nThen, run the following command. In our experiments, we use 1 A100.\n\n```bash\ntorchrun --nproc-per-node NUM_GPU train.py --cfg-path train_configs/minigpt4_stage2_finetune.yaml\n```\n\nAfter the second stage alignment, MiniGPT-4 is able to talk about the image coherently and user-friendly. \n" }, { "path": "MiniGPTv2_Train.md", "content": "## Finetune of MiniGPT-4\n\n\nYou firstly need to prepare the dataset. you can follow this step to prepare the dataset.\nour [dataset preparation](dataset/README_MINIGPTv2_FINETUNE.md). \n\nIn the train_configs/minigptv2_finetune.yaml, you need to set up the following paths:\n\nllama_model checkpoint path: \"/path/to/llama_checkpoint\"\n\nckpt: \"/path/to/pretrained_checkpoint\"\n\nckpt save path: \"/path/to/save_checkpoint\"\n\nFor ckpt, you may load from our pretrained model checkpoints:\n| MiniGPT-v2 (after stage-2) | MiniGPT-v2 (after stage-3) | MiniGPT-v2 (online developing demo) | \n|------------------------------|------------------------------|------------------------------|\n| [Download](https://drive.google.com/file/d/1Vi_E7ZtZXRAQcyz4f8E6LtLh2UXABCmu/view?usp=sharing) |[Download](https://drive.google.com/file/d/1HkoUUrjzFGn33cSiUkI-KcT-zysCynAz/view?usp=sharing) | [Download](https://drive.google.com/file/d/1aVbfW7nkCSYx99_vCRyP1sOlQiWVSnAl/view?usp=sharing) |\n\n\n```bash\ntorchrun --nproc-per-node NUM_GPU train.py --cfg-path train_configs/minigptv2_finetune.yaml\n```\n\n" }, { "path": "README.md", "content": "# MiniGPT-V\n\n**MiniGPT-v2: Large Language Model as a Unified Interface for Vision-Language Multi-task Learning**\n\nJun Chen, Deyao Zhu, Xiaoqian Shen, Xiang Li, Zechun Liu, Pengchuan Zhang, Raghuraman Krishnamoorthi, Vikas Chandra, Yunyang Xiong☨, Mohamed Elhoseiny☨\n\n☨equal last author\n\n [![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://www.youtube.com/watch?v=atFCwV2hSY4)\n\n\n **MiniGPT-4: Enhancing Vision-language Understanding with Advanced Large Language Models**\n\nDeyao Zhu*, Jun Chen*, Xiaoqian Shen, Xiang Li, Mohamed Elhoseiny\n\n*equal contribution\n\n [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1OK4kYsZphwt5DXchKkzMBjYF6jnkqh4R?usp=sharing) [![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://www.youtube.com/watch?v=__tftoxpBAw&feature=youtu.be)\n\n*King Abdullah University of Science and Technology*\n\n## πŸ’‘ Get help - [Q&A](https://github.com/Vision-CAIR/MiniGPT-4/discussions/categories/q-a) or [Discord πŸ’¬](https://discord.gg/5WdJkjbAeE)\n\n **Example Community Efforts Built on Top of MiniGPT-4 ** \n \n* **InstructionGPT-4**: A 200-Instruction Paradigm for Fine-Tuning MiniGPT-4 Lai Wei, Zihao Jiang, Weiran Huang, Lichao Sun, Arxiv, 2023\n\n* **PatFig**: Generating Short and Long Captions for Patent Figures.\", Aubakirova, Dana, Kim Gerdes, and Lufei Liu, ICCVW, 2023 \n\n\n* **SkinGPT-4**: An Interactive Dermatology Diagnostic System with Visual Large Language Model, Juexiao Zhou and Xiaonan He and Liyuan Sun and Jiannan Xu and Xiuying Chen and Yuetan Chu and Longxi Zhou and Xingyu Liao and Bin Zhang and Xin Gao, Arxiv, 2023 \n\n\n* **ArtGPT-4**: Artistic Vision-Language Understanding with Adapter-enhanced MiniGPT-4.\", Yuan, Zhengqing, Huiwen Xue, Xinyi Wang, Yongming Liu, Zhuanzhe Zhao, and Kun Wang, Arxiv, 2023 \n\n\n\n\n## News\n[Oct.31 2023] We release the evaluation code of our MiniGPT-v2. \n\n[Oct.24 2023] We release the finetuning code of our MiniGPT-v2.\n\n[Oct.13 2023] Breaking! We release the first major update with our MiniGPT-v2\n\n[Aug.28 2023] We now provide a llama 2 version of MiniGPT-4\n\n## Online Demo\n\nClick the image to chat with MiniGPT-v2 around your images\n[![demo](figs/minigpt2_demo.png)](https://minigpt-v2.github.io/)\n\nClick the image to chat with MiniGPT-4 around your images\n[![demo](figs/online_demo.png)](https://minigpt-4.github.io)\n\n\n## MiniGPT-v2 Examples\n\n![MiniGPT-v2 demos](figs/demo.png)\n\n\n\n## MiniGPT-4 Examples\n | | |\n:-------------------------:|:-------------------------:\n![find wild](figs/examples/wop_2.png) | ![write story](figs/examples/ad_2.png)\n![solve problem](figs/examples/fix_1.png) | ![write Poem](figs/examples/rhyme_1.png)\n\nMore examples can be found in the [project page](https://minigpt-4.github.io).\n\n\n\n## Getting Started\n### Installation\n\n**1. Prepare the code and the environment**\n\nGit clone our repository, creating a python environment and activate it via the following command\n\n```bash\ngit clone https://github.com/Vision-CAIR/MiniGPT-4.git\ncd MiniGPT-4\nconda env create -f environment.yml\nconda activate minigptv\n```\n\n\n**2. Prepare the pretrained LLM weights**\n\n**MiniGPT-v2** is based on Llama2 Chat 7B. For **MiniGPT-4**, we have both Vicuna V0 and Llama 2 version.\nDownload the corresponding LLM weights from the following huggingface space via clone the repository using git-lfs.\n\n| Llama 2 Chat 7B | Vicuna V0 13B | Vicuna V0 7B |\n:------------------------------------------------------------------------------------------------:|:----------------------------------------------------------------------------------------------:|:----------------------------------------------------------------------------------------------:\n[Download](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf/tree/main) | [Downlad](https://huggingface.co/Vision-CAIR/vicuna/tree/main) | [Download](https://huggingface.co/Vision-CAIR/vicuna-7b/tree/main) \n\n\nThen, set the variable *llama_model* in the model config file to the LLM weight path.\n\n* For MiniGPT-v2, set the LLM path \n[here](minigpt4/configs/models/minigpt_v2.yaml#L15) at Line 14.\n\n* For MiniGPT-4 (Llama2), set the LLM path \n[here](minigpt4/configs/models/minigpt4_llama2.yaml#L15) at Line 15.\n\n* For MiniGPT-4 (Vicuna), set the LLM path \n[here](minigpt4/configs/models/minigpt4_vicuna0.yaml#L18) at Line 18\n\n**3. Prepare the pretrained model checkpoints**\n\nDownload the pretrained model checkpoints\n\n\n| MiniGPT-v2 (after stage-2) | MiniGPT-v2 (after stage-3) | MiniGPT-v2 (online developing demo)| \n|------------------------------|------------------------------|------------------------------|\n| [Download](https://drive.google.com/file/d/1Vi_E7ZtZXRAQcyz4f8E6LtLh2UXABCmu/view?usp=sharing) |[Download](https://drive.google.com/file/d/1HkoUUrjzFGn33cSiUkI-KcT-zysCynAz/view?usp=sharing) | [Download](https://drive.google.com/file/d/1aVbfW7nkCSYx99_vCRyP1sOlQiWVSnAl/view?usp=sharing) |\n\n\nFor **MiniGPT-v2**, set the path to the pretrained checkpoint in the evaluation config file \nin [eval_configs/minigptv2_eval.yaml](eval_configs/minigptv2_eval.yaml#L10) at Line 8.\n\n\n\n| MiniGPT-4 (Vicuna 13B) | MiniGPT-4 (Vicuna 7B) | MiniGPT-4 (LLaMA-2 Chat 7B) |\n|----------------------------|---------------------------|---------------------------------|\n| [Download](https://drive.google.com/file/d/1a4zLvaiDBr-36pasffmgpvH5P7CKmpze/view?usp=share_link) | [Download](https://drive.google.com/file/d/1RY9jV0dyqLX-o38LrumkKRh6Jtaop58R/view?usp=sharing) | [Download](https://drive.google.com/file/d/11nAPjEok8eAGGEG1N2vXo3kBLCg0WgUk/view?usp=sharing) |\n\nFor **MiniGPT-4**, set the path to the pretrained checkpoint in the evaluation config file \nin [eval_configs/minigpt4_eval.yaml](eval_configs/minigpt4_eval.yaml#L10) at Line 8 for Vicuna version or [eval_configs/minigpt4_llama2_eval.yaml](eval_configs/minigpt4_llama2_eval.yaml#L10) for LLama2 version. \n\n\n\n### Launching Demo Locally\n\nFor MiniGPT-v2, run\n```\npython demo_v2.py --cfg-path eval_configs/minigptv2_eval.yaml --gpu-id 0\n```\n\nFor MiniGPT-4 (Vicuna version), run\n\n```\npython demo.py --cfg-path eval_configs/minigpt4_eval.yaml --gpu-id 0\n```\n\nFor MiniGPT-4 (Llama2 version), run\n\n```\npython demo.py --cfg-path eval_configs/minigpt4_llama2_eval.yaml --gpu-id 0\n```\n\n\nTo save GPU memory, LLMs loads as 8 bit by default, with a beam search width of 1. \nThis configuration requires about 23G GPU memory for 13B LLM and 11.5G GPU memory for 7B LLM. \nFor more powerful GPUs, you can run the model\nin 16 bit by setting `low_resource` to `False` in the relevant config file:\n\n* MiniGPT-v2: [minigptv2_eval.yaml](eval_configs/minigptv2_eval.yaml#6) \n* MiniGPT-4 (Llama2): [minigpt4_llama2_eval.yaml](eval_configs/minigpt4_llama2_eval.yaml#6)\n* MiniGPT-4 (Vicuna): [minigpt4_eval.yaml](eval_configs/minigpt4_eval.yaml#6)\n\nThanks [@WangRongsheng](https://github.com/WangRongsheng), you can also run MiniGPT-4 on [Colab](https://colab.research.google.com/drive/1OK4kYsZphwt5DXchKkzMBjYF6jnkqh4R?usp=sharing)\n\n\n### Training\nFor training details of MiniGPT-4, check [here](MiniGPT4_Train.md).\n\nFor finetuning details of MiniGPT-v2, check [here](MiniGPTv2_Train.md)\n\n\n### Evaluation\nFor finetuning details of MiniGPT-v2, check [here](eval_scripts/EVAL_README.md) \n\n\n## Acknowledgement\n\n+ [BLIP2](https://huggingface.co/docs/transformers/main/model_doc/blip-2) The model architecture of MiniGPT-4 follows BLIP-2. Don't forget to check this great open-source work if you don't know it before!\n+ [Lavis](https://github.com/salesforce/LAVIS) This repository is built upon Lavis!\n+ [Vicuna](https://github.com/lm-sys/FastChat) The fantastic language ability of Vicuna with only 13B parameters is just amazing. And it is open-source!\n+ [LLaMA](https://github.com/facebookresearch/llama) The strong open-sourced LLaMA 2 language model.\n\n\nIf you're using MiniGPT-4/MiniGPT-v2 in your research or applications, please cite using this BibTeX:\n```bibtex\n\n\n@article{chen2023minigptv2,\n title={MiniGPT-v2: large language model as a unified interface for vision-language multi-task learning}, \n author={Chen, Jun and Zhu, Deyao and Shen, Xiaoqian and Li, Xiang and Liu, Zechu and Zhang, Pengchuan and Krishnamoorthi, Raghuraman and Chandra, Vikas and Xiong, Yunyang and Elhoseiny, Mohamed},\n year={2023},\n journal={arXiv preprint arXiv:2310.09478},\n}\n\n@article{zhu2023minigpt,\n title={MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models},\n author={Zhu, Deyao and Chen, Jun and Shen, Xiaoqian and Li, Xiang and Elhoseiny, Mohamed},\n journal={arXiv preprint arXiv:2304.10592},\n year={2023}\n}\n```\n\n\n## License\nThis repository is under [BSD 3-Clause License](LICENSE.md).\nMany codes are based on [Lavis](https://github.com/salesforce/LAVIS) with \nBSD 3-Clause License [here](LICENSE_Lavis.md).\n" }, { "path": "SECURITY.md", "content": "# Security Policy\n\n## Supported Versions\n\nUse this section to tell people about which versions of your project are\ncurrently being supported with security updates.\n\n| Version | Supported |\n| ------- | ------------------ |\n| 5.1.x | :white_check_mark: |\n| 5.0.x | :x: |\n| 4.0.x | :white_check_mark: |\n| < 4.0 | :x: |\n\n## Reporting a Vulnerability\n\nUse this section to tell people how to report a vulnerability.\n\nTell them where to go, how often they can expect to get an update on a\nreported vulnerability, what to expect if the vulnerability is accepted or\ndeclined, etc.\n" }, { "path": "dataset/README_1_STAGE.md", "content": "## Download the filtered Conceptual Captions, SBU, LAION datasets\n\n### Pre-training datasets download:\nWe use the filtered synthetic captions prepared by BLIP. For more details about the dataset, please refer to [BLIP](https://github.com/salesforce/BLIP).\n\nIt requires ~2.3T to store LAION and CC3M+CC12M+SBU datasets\n\nImage source | Filtered synthetic caption by ViT-L\n--- | :---:\nCC3M+CC12M+SBU | Download\nLAION115M | Download\n\nThis will download two json files \n```\nccs_synthetic_filtered_large.json\nlaion_synthetic_filtered_large.json\n```\n\n## prepare the data step-by-step\n\n\n### setup the dataset folder and move the annotation file to the data storage folder\n```\nexport MINIGPT4_DATASET=/YOUR/PATH/FOR/LARGE/DATASET/\nmkdir ${MINIGPT4_DATASET}/cc_sbu\nmkdir ${MINIGPT4_DATASET}/laion\nmv ccs_synthetic_filtered_large.json ${MINIGPT4_DATASET}/cc_sbu\nmv laion_synthetic_filtered_large.json ${MINIGPT4_DATASET}/laion\n```\n\n### Convert the scripts to data storate folder\n```\ncp convert_cc_sbu.py ${MINIGPT4_DATASET}/cc_sbu\ncp download_cc_sbu.sh ${MINIGPT4_DATASET}/cc_sbu\ncp convert_laion.py ${MINIGPT4_DATASET}/laion\ncp download_laion.sh ${MINIGPT4_DATASET}/laion\n```\n\n\n### Convert the laion and cc_sbu annotation file format to be img2dataset format\n```\ncd ${MINIGPT4_DATASET}/cc_sbu\npython convert_cc_sbu.py\n\ncd ${MINIGPT4_DATASET}/laion\npython convert_laion.py\n```\n\n### Download the datasets with img2dataset\n```\ncd ${MINIGPT4_DATASET}/cc_sbu\nsh download_cc_sbu.sh\ncd ${MINIGPT4_DATASET}/laion\nsh download_laion.sh\n```\n\n\nThe final dataset structure\n\n```\n.\nβ”œβ”€β”€ ${MINIGPT4_DATASET}\nβ”‚ β”œβ”€β”€ cc_sbu\nβ”‚ β”œβ”€β”€ convert_cc_sbu.py\nβ”‚ β”œβ”€β”€ download_cc_sbu.sh\nβ”‚ β”œβ”€β”€ ccs_synthetic_filtered_large.json\nβ”‚ β”œβ”€β”€ ccs_synthetic_filtered_large.tsv\nβ”‚ └── cc_sbu_dataset\nβ”‚ β”œβ”€β”€ 00000.tar\nβ”‚ β”œβ”€β”€ 00000.parquet\nβ”‚ ...\nβ”‚ β”œβ”€β”€ laion\nβ”‚ β”œβ”€β”€ convert_laion.py\nβ”‚ β”œβ”€β”€ download_laion.sh\nβ”‚ β”œβ”€β”€ laion_synthetic_filtered_large.json\nβ”‚ β”œβ”€β”€ laion_synthetic_filtered_large.tsv\nβ”‚ └── laion_dataset\nβ”‚ β”œβ”€β”€ 00000.tar\nβ”‚ β”œβ”€β”€ 00000.parquet\nβ”‚ ...\n... \n```\n\n\n## Set up the dataset configuration files\n\nThen, set up the LAION dataset loading path in \n[here](../minigpt4/configs/datasets/laion/defaults.yaml#L5) at Line 5 as \n${MINIGPT4_DATASET}/laion/laion_dataset/{00000..10488}.tar\n\nand the Conceptual Captoin and SBU datasets loading path in \n[here](../minigpt4/configs/datasets/cc_sbu/defaults.yaml#L5) at Line 5 as \n${MINIGPT4_DATASET}/cc_sbu/cc_sbu_dataset/{00000..01255}.tar\n\n\n\n" }, { "path": "dataset/README_2_STAGE.md", "content": "## Second Stage Data Preparation\n\nOur second stage dataset can be downloaded from \n[here](https://drive.google.com/file/d/1nJXhoEcy3KTExr17I7BXqY5Y9Lx_-n-9/view?usp=share_link) \nAfter extraction, you will get a data follder with the following structure:\n\n```\ncc_sbu_align\nβ”œβ”€β”€ filter_cap.json\n└── image\n β”œβ”€β”€ 2.jpg\n β”œβ”€β”€ 3.jpg\n ... \n```\n\nPut the folder to any path you want.\nThen, set up the dataset path in the dataset config file \n[here](../minigpt4/configs/datasets/cc_sbu/align.yaml#L5) at Line 5.\n\n" }, { "path": "dataset/README_MINIGPTv2_FINETUNE.md", "content": "## Download the dataset for finetuning the MiniGPT-v2\n\n\nDownload the dataset\n\nImage source | Download path\n--- | :---:\nCOCO 2014 images | images    captions\nCOCO VQA | vqa train    vqa val\nVisual Genome | images part1    images part2    image meta data \nTextCaps | images    annotations \nRefCOCO | annotations \nRefCOCO+ | annotations \nRefCOCOg | annotations \nOKVQA | annotations \nAOK-VQA | annotations \nOCR-VQA | annotations \nGQA | images    annotations \nFiltered flickr-30k | annotations \nMulti-task conversation | annotations \nFiltered unnatural instruction | annotations \nLLaVA | Compelex reasoning    Detailed description    Conversation \n\n\n\n### COCO captions\nDownload the COCO 2014 images and captions\n\ncoco 2014 images path\n\n```\n${MINIGPTv2_DATASET}\nβ”œβ”€β”€ coco\nβ”‚ β”œβ”€β”€ images\n...\n```\n\n\ncoco caption annotation path\n\n```\n${MINIGPTv2_DATASET}\nβ”œβ”€β”€ coco_captions\nβ”‚ └── annotations\nβ”‚ β”œβ”€β”€ coco_karpathy_train.json\n...\n```\n\nSet **image_path** to the COCO 2014 image folder.\nSimilarly, set **ann_path** to the coco_karpathy_train.json path\n- [minigpt4/configs/datasets/coco/caption.yaml](../minigpt4/configs/datasets/coco/caption.yaml)\n\n### COCO VQA\nDownload the vqa v2 train and validation json files\n\n```\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ vqav2\nβ”‚ β”œβ”€β”€ vqa_train.json\n| β”œβ”€β”€ vqa_val.json\n```\n\nSet **image_path** to the COCO 2014 image folder.\nSimilarly, set **ann_path** to the vqa_train.json and vqa_val.json path\n- [minigpt4/configs/datasets/coco/defaults_vqa.yaml](../minigpt4/configs/datasets/coco/defaults_vqa.yaml)\n\n\n### Visual genome\nDownload visiual genome images and annotation files\n\n```\n${MINIGPTv2_DATASET}\nβ”œβ”€β”€ visual_genome\nβ”‚ β”œβ”€β”€ VG_100K\nβ”‚ β”œβ”€β”€ VG_100K_2\nβ”‚ └── region_descriptions.json\nβ”‚ └── image_data.json\n...\n```\n\nSet **image_path** to visual_genome folder.\nSimilarly, set **ann_path** to the visual_genome folder.\n\n- [minigpt4/configs/datasets/vg/ref.yaml](../minigpt4/configs/datasets/vg/ref.yaml)\n\n\n### TextCaps\nDownload the TextCaps images and annotation files\n\n```\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ textcaps\nβ”‚ β”œβ”€β”€ train_images\nβ”‚ β”œβ”€β”€ TextCaps_0.1_train.json\n```\n\nSet **image_path** to TextCaps train_images folder.\nSimilarly, set **ann_path** to the TextCaps_0.1_train.json path\n\n- [minigpt4/configs/datasets/textcaps/caption.yaml](../minigpt4/configs/datasets/textcaps/caption.yaml)\n\n### RefCOCO, RefCOCO+, RefCOCOg\nDownload the RefCOCO, RefCOCO+, RefCOCOg annotation files\n\n```\n\n${MINIGPTv2_DATASET}\nβ”œβ”€β”€ refcoco_annotations\nβ”‚ β”œβ”€β”€ refcoco\nβ”‚ β”‚ β”œβ”€β”€ instances.json\nβ”‚ β”‚ β”œβ”€β”€ refs(google).p\nβ”‚ β”‚ └── refs(unc).p\nβ”‚ β”œβ”€β”€ refcoco+\nβ”‚ β”‚ β”œβ”€β”€ instances.json\nβ”‚ β”‚ └── refs(unc).p\nβ”‚ └── refcocog\nβ”‚ β”œβ”€β”€ instances.json\nβ”‚ β”œβ”€β”€ refs(google).p\nβ”‚ └─── refs(und).p\n...\n```\n\n\nSet **image_path** to the COCO 2014 image folder.\nSimilarly, set **ann_path** in all the following configs to the above folder *refcoco_annotations* that contains refcoco, refcoco+, and refcocog.\n\n- [minigpt4/configs/datasets/coco_bbox/refcoco.yaml](../minigpt4/configs/datasets/coco_bbox/refcoco.yaml)\n- [minigpt4/configs/datasets/coco_bbox/refcocog.yaml](../minigpt4/configs/datasets/coco_bbox/refcocog.yaml) \n- [minigpt4/configs/datasets/coco_bbox/refcocop.yaml](../minigpt4/configs/datasets/coco_bbox/refcocop.yaml)\n- [minigpt4/configs/datasets/coco_bbox/invrefcoco.yaml](../minigpt4/configs/datasets/coco_bbox/invrefcoco.yaml)\n- [minigpt4/configs/datasets/coco_bbox/invrefcocog.yaml](../minigpt4/configs/datasets/coco_bbox/invrefcocog.yaml) \n- [minigpt4/configs/datasets/coco_bbox/invrefcocop.yaml](../minigpt4/configs/datasets/coco_bbox/invrefcocop.yaml)\n\n\n\n\n### OKVQA\n\n\n```\nLocation_you_like\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ okvqa\nβ”‚ β”œβ”€β”€ okvqa_train.json\n```\n\nSet **image_path** to the COCO 2014 image folder.\nSimilarly, set **ann_path** to the location of the OKVQA dataset\n- [minigpt4/configs/datasets/okvqa/defaults.yaml](../minigpt4/configs/datasets/okvqa/defaults.yaml)\n\n\n### COCO-VQA\n\n- [OK-VQA Input Questions](https://okvqa.allenai.org/static/data/OpenEnded_mscoco_train2014_questions.json.zip)\n- [OK-VQA Annotations](https://okvqa.allenai.org/static/data/mscoco_train2014_annotations.json.zip)\n\n\n### AOK-VQA\nDownload the AOK-VQA annotation dataset\n\n```\nexport AOKVQA_DIR=YOUR_DATASET_PATH\nmkdir -p ${AOKVQA_DIR}\ncurl -fsSL https://prior-datasets.s3.us-east-2.amazonaws.com/aokvqa/aokvqa_v1p0.tar.gz | tar xvz -C ${AOKVQA_DIR}\n```\n\n```\nLocation_you_like\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ aokvqa\nβ”‚ β”œβ”€β”€ aokvqa_v1p0_train.json\n```\n\n\nSet **image_path** to the COCO 2014 image folder.\nSimilarly, set **ann_path** to the location of the AOKVQA dataset\n- [minigpt4/configs/datasets/aokvqa/defaults.yaml](../minigpt4/configs/datasets/aokvqa/defaults.yaml)\n\n\n\n### OCR-VQA\nDownload the OCR-VQA annotation files\ndownload the images with loadDataset.py script\n\n```\nLocation_you_like\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ ocrvqa\nβ”‚ β”œβ”€β”€ images\nβ”‚ β”œβ”€β”€ dataset.json\n```\n\nSet **image_path** as the ocrvqa/images folder.\nSimilarly, set **ann_path** to the dataset.json\n- [minigpt4/configs/datasets/ocrvqa/ocrvqa.yaml](../minigpt4/configs/datasets/ocrvqa/ocrvqa.yaml)\n\n### GQA\nDownload the GQA annotation files and images\n\n```\nLocation_you_like\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ gqa\nβ”‚ β”œβ”€β”€ images\nβ”‚ β”œβ”€β”€ train_balanced_questions.json\n```\n\nSet **image_path** as the gqa/images folder.\nSimilarly, set **ann_path** to the train_balanced_questions.json\n- [minigpt4/configs/datasets/gqa/balanced_val.yaml](../minigpt4/configs/datasets/gqa/balanced_val.yaml)\n\n\n\n### filtered Flickr-30k\nDownload filtered Flickr-30k images (fill this [form](https://forms.illinois.edu/sec/229675) on official website or from [kaggle](https://www.kaggle.com/datasets/hsankesara/flickr-image-dataset/download?datasetVersionNumber=1)) and annotation files\n\n```\n${MINIGPTv2_DATASET}\nβ”œβ”€β”€ filtered_flickr\nβ”‚ β”œβ”€β”€ images\nβ”‚ β”œβ”€β”€ captiontobbox.json\nβ”‚ β”œβ”€β”€ groundedcaption.json\nβ”‚ └── phrasetobbox.json\n...\n```\n\nSet **image_path** as the flickr-30k images foler.\nSimilarly, set **ann_path** to the groundedcaption.json, captiontobbox.json and phrasetobbox.json for the \ngrounded image caption, caption to bbox, and phrase to bbox datasets.\n\n- [minigpt4/configs/datasets/flickr/default.yaml](../minigpt4/configs/datasets/flickr/default.yaml)\n- [minigpt4/configs/datasets/flickr/caption_to_phrase.yaml](../minigpt4/configs/datasets/flickr/caption_to_phrase.yaml)\n- [minigpt4/configs/datasets/flickr/object_to_phrase.yaml](../minigpt4/configs/datasets/flickr/object_to_phrase.yaml)\n\n\n### Multi-task conversation\nDownload the multi-task converstation dataset\n\n```\nLocation_you_like\n${MINIGPTv2_DATASET}\nβ”œβ”€β”€ multitask_conversation\nβ”‚ └── multitask_conversation.json\n...\n```\n\nSet **image_path** as the COCO 2014 images folder.\nSimilarly, set **ann_path** to the multitask_conversation.json file path\n\n- [minigpt4/configs/datasets/multitask_conversation/default.yaml](../minigpt4/configs/datasets/multitask_conversation/default.yaml)\n\n### Unnatural instruction\nDownload the filtered unnatural instruction annotation files (we remove the very long sentences from the original unnatural instruction dataset)\n\n```\nLocation_you_like\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ unnatural_instructions\nβ”‚ β”œβ”€β”€ filtered_unnatural_instruction.json\n```\n\nThere is no image path.\nSimilarly, set **ann_path** to the filtered_unnatural_instruction.json file path\n\n- [minigpt4/configs/datasets/nlp/unnatural_instruction.yaml](../minigpt4/configs/datasets/nlp/unnatural_instruction.yaml)\n\n### LLaVA\n\n```\nLocation_you_like\nβ”œβ”€β”€ ${MINIGPTv2_DATASET}\nβ”‚ β”œβ”€β”€ llava\nβ”‚ β”œβ”€β”€ conversation_58k.json\nβ”‚ β”œβ”€β”€ detail_23k.json\nβ”‚ β”œβ”€β”€ complex_reasoning_77k.json\n```\n\nSet **image_path** to the COCO 2014 image folder.\nSimilarly, set **ann_path** to the location of the previous downloaded conversation_58k.json, \ndetail_23k.json, and complex_reasoning_77k.json in conversation.yaml, detail.yaml, and reason.yaml, respectively.\n\n\n- [minigpt4/configs/datasets/llava/conversation.yaml](../minigpt4/configs/datasets/llava/conversation.yaml)\n- [minigpt4/configs/datasets/llava/detail.yaml](../minigpt4/configs/datasets/llava/detail.yaml) \n- [minigpt4/configs/datasets/llava/reason.yaml](../minigpt4/configs/datasets/llava/reason.yaml)\n" }, { "path": "dataset/convert_cc_sbu.py", "content": "import json\nimport csv\n\n# specify input and output file paths\ninput_file = 'ccs_synthetic_filtered_large.json'\noutput_file = 'ccs_synthetic_filtered_large.tsv'\n\n# load JSON data from input file\nwith open(input_file, 'r') as f:\n data = json.load(f)\n\n# extract header and data from JSON\nheader = data[0].keys()\nrows = [x.values() for x in data]\n\n# write data to TSV file\nwith open(output_file, 'w') as f:\n writer = csv.writer(f, delimiter='\\t')\n writer.writerow(header)\n writer.writerows(rows)\n" }, { "path": "dataset/convert_laion.py", "content": "import json\nimport csv\n\n# specify input and output file paths\ninput_file = 'laion_synthetic_filtered_large.json'\noutput_file = 'laion_synthetic_filtered_large.tsv'\n\n# load JSON data from input file\nwith open(input_file, 'r') as f:\n data = json.load(f)\n\n# extract header and data from JSON\nheader = data[0].keys()\nrows = [x.values() for x in data]\n\n# write data to TSV file\nwith open(output_file, 'w') as f:\n writer = csv.writer(f, delimiter='\\t')\n writer.writerow(header)\n writer.writerows(rows)\n" }, { "path": "demo.py", "content": "import argparse\nimport os\nimport random\n\nimport numpy as np\nimport torch\nimport torch.backends.cudnn as cudnn\nimport gradio as gr\n\nfrom transformers import StoppingCriteriaList\n\nfrom minigpt4.common.config import Config\nfrom minigpt4.common.dist_utils import get_rank\nfrom minigpt4.common.registry import registry\nfrom minigpt4.conversation.conversation import Chat, CONV_VISION_Vicuna0, CONV_VISION_LLama2, StoppingCriteriaSub\n\n# imports modules for registration\nfrom minigpt4.datasets.builders import *\nfrom minigpt4.models import *\nfrom minigpt4.processors import *\nfrom minigpt4.runners import *\nfrom minigpt4.tasks import *\n\n\ndef parse_args():\n parser = argparse.ArgumentParser(description=\"Demo\")\n parser.add_argument(\"--cfg-path\", required=True, help=\"path to configuration file.\")\n parser.add_argument(\"--gpu-id\", type=int, default=0, help=\"specify the gpu to load the model.\")\n parser.add_argument(\n \"--options\",\n nargs=\"+\",\n help=\"override some settings in the used config, the key-value pair \"\n \"in xxx=yyy format will be merged into config file (deprecate), \"\n \"change to --cfg-options instead.\",\n )\n args = parser.parse_args()\n return args\n\n\ndef setup_seeds(config):\n seed = config.run_cfg.seed + get_rank()\n\n random.seed(seed)\n np.random.seed(seed)\n torch.manual_seed(seed)\n\n cudnn.benchmark = False\n cudnn.deterministic = True\n\n\n# ========================================\n# Model Initialization\n# ========================================\n\nconv_dict = {'pretrain_vicuna0': CONV_VISION_Vicuna0,\n 'pretrain_llama2': CONV_VISION_LLama2}\n\nprint('Initializing Chat')\nargs = parse_args()\ncfg = Config(args)\n\nmodel_config = cfg.model_cfg\nmodel_config.device_8bit = args.gpu_id\nmodel_cls = registry.get_model_class(model_config.arch)\nmodel = model_cls.from_config(model_config).to('cuda:{}'.format(args.gpu_id))\n\nCONV_VISION = conv_dict[model_config.model_type]\n\nvis_processor_cfg = cfg.datasets_cfg.cc_sbu_align.vis_processor.train\nvis_processor = registry.get_processor_class(vis_processor_cfg.name).from_config(vis_processor_cfg)\n\nstop_words_ids = [[835], [2277, 29937]]\nstop_words_ids = [torch.tensor(ids).to(device='cuda:{}'.format(args.gpu_id)) for ids in stop_words_ids]\nstopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(stops=stop_words_ids)])\n\nchat = Chat(model, vis_processor, device='cuda:{}'.format(args.gpu_id), stopping_criteria=stopping_criteria)\nprint('Initialization Finished')\n\n\n# ========================================\n# Gradio Setting\n# ========================================\n\n\ndef gradio_reset(chat_state, img_list):\n if chat_state is not None:\n chat_state.messages = []\n if img_list is not None:\n img_list = []\n return None, gr.update(value=None, interactive=True), gr.update(placeholder='Please upload your image first', interactive=False),gr.update(value=\"Upload & Start Chat\", interactive=True), chat_state, img_list\n\n\ndef upload_img(gr_img, text_input, chat_state):\n if gr_img is None:\n return None, None, gr.update(interactive=True), chat_state, None\n chat_state = CONV_VISION.copy()\n img_list = []\n llm_message = chat.upload_img(gr_img, chat_state, img_list)\n chat.encode_img(img_list)\n return gr.update(interactive=False), gr.update(interactive=True, placeholder='Type and press Enter'), gr.update(value=\"Start Chatting\", interactive=False), chat_state, img_list\n\n\ndef gradio_ask(user_message, chatbot, chat_state):\n if len(user_message) == 0:\n return gr.update(interactive=True, placeholder='Input should not be empty!'), chatbot, chat_state\n chat.ask(user_message, chat_state)\n chatbot = chatbot + [[user_message, None]]\n return '', chatbot, chat_state\n\n\ndef gradio_answer(chatbot, chat_state, img_list, num_beams, temperature):\n llm_message = chat.answer(conv=chat_state,\n img_list=img_list,\n num_beams=num_beams,\n temperature=temperature,\n max_new_tokens=300,\n max_length=2000)[0]\n chatbot[-1][1] = llm_message\n return chatbot, chat_state, img_list\n\n\ntitle = \"\"\"

Demo of MiniGPT-4

\"\"\"\ndescription = \"\"\"

This is the demo of MiniGPT-4. Upload your images and start chatting!

\"\"\"\narticle = \"\"\"

\n\"\"\"\n\n#TODO show examples below\n\nwith gr.Blocks() as demo:\n gr.Markdown(title)\n gr.Markdown(description)\n gr.Markdown(article)\n\n with gr.Row():\n with gr.Column(scale=1):\n image = gr.Image(type=\"pil\")\n upload_button = gr.Button(value=\"Upload & Start Chat\", interactive=True, variant=\"primary\")\n clear = gr.Button(\"Restart\")\n \n num_beams = gr.Slider(\n minimum=1,\n maximum=10,\n value=1,\n step=1,\n interactive=True,\n label=\"beam search numbers)\",\n )\n \n temperature = gr.Slider(\n minimum=0.1,\n maximum=2.0,\n value=1.0,\n step=0.1,\n interactive=True,\n label=\"Temperature\",\n )\n\n with gr.Column(scale=2):\n chat_state = gr.State()\n img_list = gr.State()\n chatbot = gr.Chatbot(label='MiniGPT-4')\n text_input = gr.Textbox(label='User', placeholder='Please upload your image first', interactive=False)\n \n upload_button.click(upload_img, [image, text_input, chat_state], [image, text_input, upload_button, chat_state, img_list])\n \n text_input.submit(gradio_ask, [text_input, chatbot, chat_state], [text_input, chatbot, chat_state]).then(\n gradio_answer, [chatbot, chat_state, img_list, num_beams, temperature], [chatbot, chat_state, img_list]\n )\n clear.click(gradio_reset, [chat_state, img_list], [chatbot, image, text_input, upload_button, chat_state, img_list], queue=False)\n\ndemo.launch(share=True, enable_queue=True)\n" }, { "path": "demo_v2.py", "content": "import argparse\nimport os\nimport random\nfrom collections import defaultdict\n\nimport cv2\nimport re\n\nimport numpy as np\nfrom PIL import Image\nimport torch\nimport html\nimport gradio as gr\n\nimport torchvision.transforms as T\nimport torch.backends.cudnn as cudnn\n\nfrom minigpt4.common.config import Config\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.conversation.conversation import Conversation, SeparatorStyle, Chat\n\n# imports modules for registration\nfrom minigpt4.datasets.builders import *\nfrom minigpt4.models import *\nfrom minigpt4.processors import *\nfrom minigpt4.runners import *\nfrom minigpt4.tasks import *\n\n\ndef parse_args():\n parser = argparse.ArgumentParser(description=\"Demo\")\n parser.add_argument(\"--cfg-path\", default='eval_configs/minigptv2_eval.yaml',\n help=\"path to configuration file.\")\n parser.add_argument(\"--gpu-id\", type=int, default=0, help=\"specify the gpu to load the model.\")\n parser.add_argument(\n \"--options\",\n nargs=\"+\",\n help=\"override some settings in the used config, the key-value pair \"\n \"in xxx=yyy format will be merged into config file (deprecate), \"\n \"change to --cfg-options instead.\",\n )\n args = parser.parse_args()\n return args\n\n\nrandom.seed(42)\nnp.random.seed(42)\ntorch.manual_seed(42)\n\ncudnn.benchmark = False\ncudnn.deterministic = True\n\nprint('Initializing Chat')\nargs = parse_args()\ncfg = Config(args)\n\ndevice = 'cuda:{}'.format(args.gpu_id)\n\nmodel_config = cfg.model_cfg\nmodel_config.device_8bit = args.gpu_id\nmodel_cls = registry.get_model_class(model_config.arch)\nmodel = model_cls.from_config(model_config).to(device)\nbounding_box_size = 100\n\nvis_processor_cfg = cfg.datasets_cfg.cc_sbu_align.vis_processor.train\nvis_processor = registry.get_processor_class(vis_processor_cfg.name).from_config(vis_processor_cfg)\n\nmodel = model.eval()\n\nCONV_VISION = Conversation(\n system=\"\",\n roles=(r\"[INST] \", r\" [/INST]\"),\n messages=[],\n offset=2,\n sep_style=SeparatorStyle.SINGLE,\n sep=\"\",\n)\n\n\ndef extract_substrings(string):\n # first check if there is no-finished bracket\n index = string.rfind('}')\n if index != -1:\n string = string[:index + 1]\n\n pattern = r'

(.*?)\\}(?!<)'\n matches = re.findall(pattern, string)\n substrings = [match for match in matches]\n\n return substrings\n\n\ndef is_overlapping(rect1, rect2):\n x1, y1, x2, y2 = rect1\n x3, y3, x4, y4 = rect2\n return not (x2 < x3 or x1 > x4 or y2 < y3 or y1 > y4)\n\n\ndef computeIoU(bbox1, bbox2):\n x1, y1, x2, y2 = bbox1\n x3, y3, x4, y4 = bbox2\n intersection_x1 = max(x1, x3)\n intersection_y1 = max(y1, y3)\n intersection_x2 = min(x2, x4)\n intersection_y2 = min(y2, y4)\n intersection_area = max(0, intersection_x2 - intersection_x1 + 1) * max(0, intersection_y2 - intersection_y1 + 1)\n bbox1_area = (x2 - x1 + 1) * (y2 - y1 + 1)\n bbox2_area = (x4 - x3 + 1) * (y4 - y3 + 1)\n union_area = bbox1_area + bbox2_area - intersection_area\n iou = intersection_area / union_area\n return iou\n\n\ndef save_tmp_img(visual_img):\n file_name = \"\".join([str(random.randint(0, 9)) for _ in range(5)]) + \".jpg\"\n file_path = \"/tmp/gradio\" + file_name\n visual_img.save(file_path)\n return file_path\n\n\ndef mask2bbox(mask):\n if mask is None:\n return ''\n mask = mask.resize([100, 100], resample=Image.NEAREST)\n mask = np.array(mask)[:, :, 0]\n\n rows = np.any(mask, axis=1)\n cols = np.any(mask, axis=0)\n\n if rows.sum():\n # Get the top, bottom, left, and right boundaries\n rmin, rmax = np.where(rows)[0][[0, -1]]\n cmin, cmax = np.where(cols)[0][[0, -1]]\n bbox = '{{<{}><{}><{}><{}>}}'.format(cmin, rmin, cmax, rmax)\n else:\n bbox = ''\n\n return bbox\n\n\ndef escape_markdown(text):\n # List of Markdown special characters that need to be escaped\n md_chars = ['<', '>']\n\n # Escape each special character\n for char in md_chars:\n text = text.replace(char, '\\\\' + char)\n\n return text\n\n\ndef reverse_escape(text):\n md_chars = ['\\\\<', '\\\\>']\n\n for char in md_chars:\n text = text.replace(char, char[1:])\n\n return text\n\n\ncolors = [\n (255, 0, 0),\n (0, 255, 0),\n (0, 0, 255),\n (210, 210, 0),\n (255, 0, 255),\n (0, 255, 255),\n (114, 128, 250),\n (0, 165, 255),\n (0, 128, 0),\n (144, 238, 144),\n (238, 238, 175),\n (255, 191, 0),\n (0, 128, 0),\n (226, 43, 138),\n (255, 0, 255),\n (0, 215, 255),\n]\n\ncolor_map = {\n f\"{color_id}\": f\"#{hex(color[2])[2:].zfill(2)}{hex(color[1])[2:].zfill(2)}{hex(color[0])[2:].zfill(2)}\" for\n color_id, color in enumerate(colors)\n}\n\nused_colors = colors\n\n\ndef visualize_all_bbox_together(image, generation):\n if image is None:\n return None, ''\n\n generation = html.unescape(generation)\n\n image_width, image_height = image.size\n image = image.resize([500, int(500 / image_width * image_height)])\n image_width, image_height = image.size\n\n string_list = extract_substrings(generation)\n if string_list: # it is grounding or detection\n mode = 'all'\n entities = defaultdict(list)\n i = 0\n j = 0\n for string in string_list:\n try:\n obj, string = string.split('

')\n except ValueError:\n print('wrong string: ', string)\n continue\n bbox_list = string.split('')\n flag = False\n for bbox_string in bbox_list:\n integers = re.findall(r'-?\\d+', bbox_string)\n if len(integers) == 4:\n x0, y0, x1, y1 = int(integers[0]), int(integers[1]), int(integers[2]), int(integers[3])\n left = x0 / bounding_box_size * image_width\n bottom = y0 / bounding_box_size * image_height\n right = x1 / bounding_box_size * image_width\n top = y1 / bounding_box_size * image_height\n\n entities[obj].append([left, bottom, right, top])\n\n j += 1\n flag = True\n if flag:\n i += 1\n else:\n integers = re.findall(r'-?\\d+', generation)\n\n if len(integers) == 4: # it is refer\n mode = 'single'\n\n entities = list()\n x0, y0, x1, y1 = int(integers[0]), int(integers[1]), int(integers[2]), int(integers[3])\n left = x0 / bounding_box_size * image_width\n bottom = y0 / bounding_box_size * image_height\n right = x1 / bounding_box_size * image_width\n top = y1 / bounding_box_size * image_height\n entities.append([left, bottom, right, top])\n else:\n # don't detect any valid bbox to visualize\n return None, ''\n\n if len(entities) == 0:\n return None, ''\n\n if isinstance(image, Image.Image):\n image_h = image.height\n image_w = image.width\n image = np.array(image)\n\n elif isinstance(image, str):\n if os.path.exists(image):\n pil_img = Image.open(image).convert(\"RGB\")\n image = np.array(pil_img)[:, :, [2, 1, 0]]\n image_h = pil_img.height\n image_w = pil_img.width\n else:\n raise ValueError(f\"invaild image path, {image}\")\n elif isinstance(image, torch.Tensor):\n\n image_tensor = image.cpu()\n reverse_norm_mean = torch.tensor([0.48145466, 0.4578275, 0.40821073])[:, None, None]\n reverse_norm_std = torch.tensor([0.26862954, 0.26130258, 0.27577711])[:, None, None]\n image_tensor = image_tensor * reverse_norm_std + reverse_norm_mean\n pil_img = T.ToPILImage()(image_tensor)\n image_h = pil_img.height\n image_w = pil_img.width\n image = np.array(pil_img)[:, :, [2, 1, 0]]\n else:\n raise ValueError(f\"invaild image format, {type(image)} for {image}\")\n\n indices = list(range(len(entities)))\n\n new_image = image.copy()\n\n previous_bboxes = []\n # size of text\n text_size = 0.5\n # thickness of text\n text_line = 1 # int(max(1 * min(image_h, image_w) / 512, 1))\n box_line = 2\n (c_width, text_height), _ = cv2.getTextSize(\"F\", cv2.FONT_HERSHEY_COMPLEX, text_size, text_line)\n base_height = int(text_height * 0.675)\n text_offset_original = text_height - base_height\n text_spaces = 2\n\n # num_bboxes = sum(len(x[-1]) for x in entities)\n used_colors = colors # random.sample(colors, k=num_bboxes)\n\n color_id = -1\n for entity_idx, entity_name in enumerate(entities):\n if mode == 'single' or mode == 'identify':\n bboxes = entity_name\n bboxes = [bboxes]\n else:\n bboxes = entities[entity_name]\n color_id += 1\n for bbox_id, (x1_norm, y1_norm, x2_norm, y2_norm) in enumerate(bboxes):\n skip_flag = False\n orig_x1, orig_y1, orig_x2, orig_y2 = int(x1_norm), int(y1_norm), int(x2_norm), int(y2_norm)\n\n color = used_colors[entity_idx % len(used_colors)] # tuple(np.random.randint(0, 255, size=3).tolist())\n new_image = cv2.rectangle(new_image, (orig_x1, orig_y1), (orig_x2, orig_y2), color, box_line)\n\n if mode == 'all':\n l_o, r_o = box_line // 2 + box_line % 2, box_line // 2 + box_line % 2 + 1\n\n x1 = orig_x1 - l_o\n y1 = orig_y1 - l_o\n\n if y1 < text_height + text_offset_original + 2 * text_spaces:\n y1 = orig_y1 + r_o + text_height + text_offset_original + 2 * text_spaces\n x1 = orig_x1 + r_o\n\n # add text background\n (text_width, text_height), _ = cv2.getTextSize(f\" {entity_name}\", cv2.FONT_HERSHEY_COMPLEX, text_size,\n text_line)\n text_bg_x1, text_bg_y1, text_bg_x2, text_bg_y2 = x1, y1 - (\n text_height + text_offset_original + 2 * text_spaces), x1 + text_width, y1\n\n for prev_bbox in previous_bboxes:\n if computeIoU((text_bg_x1, text_bg_y1, text_bg_x2, text_bg_y2), prev_bbox['bbox']) > 0.95 and \\\n prev_bbox['phrase'] == entity_name:\n skip_flag = True\n break\n while is_overlapping((text_bg_x1, text_bg_y1, text_bg_x2, text_bg_y2), prev_bbox['bbox']):\n text_bg_y1 += (text_height + text_offset_original + 2 * text_spaces)\n text_bg_y2 += (text_height + text_offset_original + 2 * text_spaces)\n y1 += (text_height + text_offset_original + 2 * text_spaces)\n\n if text_bg_y2 >= image_h:\n text_bg_y1 = max(0, image_h - (text_height + text_offset_original + 2 * text_spaces))\n text_bg_y2 = image_h\n y1 = image_h\n break\n if not skip_flag:\n alpha = 0.5\n for i in range(text_bg_y1, text_bg_y2):\n for j in range(text_bg_x1, text_bg_x2):\n if i < image_h and j < image_w:\n if j < text_bg_x1 + 1.35 * c_width:\n # original color\n bg_color = color\n else:\n # white\n bg_color = [255, 255, 255]\n new_image[i, j] = (alpha * new_image[i, j] + (1 - alpha) * np.array(bg_color)).astype(\n np.uint8)\n\n cv2.putText(\n new_image, f\" {entity_name}\", (x1, y1 - text_offset_original - 1 * text_spaces),\n cv2.FONT_HERSHEY_COMPLEX, text_size, (0, 0, 0), text_line, cv2.LINE_AA\n )\n\n previous_bboxes.append(\n {'bbox': (text_bg_x1, text_bg_y1, text_bg_x2, text_bg_y2), 'phrase': entity_name})\n\n if mode == 'all':\n def color_iterator(colors):\n while True:\n for color in colors:\n yield color\n\n color_gen = color_iterator(colors)\n\n # Add colors to phrases and remove

\n def colored_phrases(match):\n phrase = match.group(1)\n color = next(color_gen)\n return f'{phrase}'\n\n generation = re.sub(r'{<\\d+><\\d+><\\d+><\\d+>}|', '', generation)\n generation_colored = re.sub(r'

(.*?)

', colored_phrases, generation)\n else:\n generation_colored = ''\n\n pil_image = Image.fromarray(new_image)\n return pil_image, generation_colored\n\n\ndef gradio_reset(chat_state, img_list):\n if chat_state is not None:\n chat_state.messages = []\n if img_list is not None:\n img_list = []\n return None, gr.update(value=None, interactive=True), gr.update(placeholder='Upload your image and chat',\n interactive=True), chat_state, img_list\n\n\ndef image_upload_trigger(upload_flag, replace_flag, img_list):\n # set the upload flag to true when receive a new image.\n # if there is an old image (and old conversation), set the replace flag to true to reset the conv later.\n upload_flag = 1\n if img_list:\n replace_flag = 1\n return upload_flag, replace_flag\n\n\ndef example_trigger(text_input, image, upload_flag, replace_flag, img_list):\n # set the upload flag to true when receive a new image.\n # if there is an old image (and old conversation), set the replace flag to true to reset the conv later.\n upload_flag = 1\n if img_list or replace_flag == 1:\n replace_flag = 1\n\n return upload_flag, replace_flag\n\n\ndef gradio_ask(user_message, chatbot, chat_state, gr_img, img_list, upload_flag, replace_flag):\n if len(user_message) == 0:\n text_box_show = 'Input should not be empty!'\n else:\n text_box_show = ''\n\n if isinstance(gr_img, dict):\n gr_img, mask = gr_img['image'], gr_img['mask']\n else:\n mask = None\n\n if '[identify]' in user_message:\n # check if user provide bbox in the text input\n integers = re.findall(r'-?\\d+', user_message)\n if len(integers) != 4: # no bbox in text\n bbox = mask2bbox(mask)\n user_message = user_message + bbox\n\n if chat_state is None:\n chat_state = CONV_VISION.copy()\n\n if upload_flag:\n if replace_flag:\n chat_state = CONV_VISION.copy() # new image, reset everything\n replace_flag = 0\n chatbot = []\n img_list = []\n llm_message = chat.upload_img(gr_img, chat_state, img_list)\n upload_flag = 0\n\n chat.ask(user_message, chat_state)\n\n chatbot = chatbot + [[user_message, None]]\n\n if '[identify]' in user_message:\n visual_img, _ = visualize_all_bbox_together(gr_img, user_message)\n if visual_img is not None:\n file_path = save_tmp_img(visual_img)\n chatbot = chatbot + [[(file_path,), None]]\n\n return text_box_show, chatbot, chat_state, img_list, upload_flag, replace_flag\n\n\ndef gradio_answer(chatbot, chat_state, img_list, temperature):\n llm_message = chat.answer(conv=chat_state,\n img_list=img_list,\n temperature=temperature,\n max_new_tokens=500,\n max_length=2000)[0]\n chatbot[-1][1] = llm_message\n return chatbot, chat_state\n\n\ndef gradio_stream_answer(chatbot, chat_state, img_list, temperature):\n if len(img_list) > 0:\n if not isinstance(img_list[0], torch.Tensor):\n chat.encode_img(img_list)\n streamer = chat.stream_answer(conv=chat_state,\n img_list=img_list,\n temperature=temperature,\n max_new_tokens=500,\n max_length=2000)\n output = ''\n for new_output in streamer:\n escapped = escape_markdown(new_output)\n output += escapped\n chatbot[-1][1] = output\n yield chatbot, chat_state\n chat_state.messages[-1][1] = ''\n return chatbot, chat_state\n\n\ndef gradio_visualize(chatbot, gr_img):\n if isinstance(gr_img, dict):\n gr_img, mask = gr_img['image'], gr_img['mask']\n\n unescaped = reverse_escape(chatbot[-1][1])\n visual_img, generation_color = visualize_all_bbox_together(gr_img, unescaped)\n if visual_img is not None:\n if len(generation_color):\n chatbot[-1][1] = generation_color\n file_path = save_tmp_img(visual_img)\n chatbot = chatbot + [[None, (file_path,)]]\n\n return chatbot\n\n\ndef gradio_taskselect(idx):\n prompt_list = [\n '',\n '[grounding] describe this image in detail',\n '[refer] ',\n '[detection] ',\n '[identify] what is this ',\n '[vqa] '\n ]\n instruct_list = [\n '**Hint:** Type in whatever you want',\n '**Hint:** Send the command to generate a grounded image description',\n '**Hint:** Type in a phrase about an object in the image and send the command',\n '**Hint:** Type in a caption or phrase, and see object locations in the image',\n '**Hint:** Draw a bounding box on the uploaded image then send the command. Click the \"clear\" botton on the top right of the image before redraw',\n '**Hint:** Send a question to get a short answer',\n ]\n return prompt_list[idx], instruct_list[idx]\n\n\n\n\nchat = Chat(model, vis_processor, device=device)\n\ntitle = \"\"\"

MiniGPT-v2 Demo

\"\"\"\ndescription = 'Welcome to Our MiniGPT-v2 Chatbot Demo!'\n# article = \"\"\"

\"\"\"\narticle = \"\"\"

\"\"\"\n\nintroduction = '''\nFor Abilities Involving Visual Grounding:\n1. Grounding: CLICK **Send** to generate a grounded image description.\n2. Refer: Input a referring object and CLICK **Send**.\n3. Detection: Write a caption or phrase, and CLICK **Send**.\n4. Identify: Draw the bounding box on the uploaded image window and CLICK **Send** to generate the bounding box. (CLICK \"clear\" button before re-drawing next time).\n5. VQA: Input a visual question and CLICK **Send**.\n6. No Tag: Input whatever you want and CLICK **Send** without any tagging\n\nYou can also simply chat in free form!\n'''\n\ntext_input = gr.Textbox(placeholder='Upload your image and chat', interactive=True, show_label=False, container=False,\n scale=8)\nwith gr.Blocks() as demo:\n gr.Markdown(title)\n # gr.Markdown(description)\n gr.Markdown(article)\n\n with gr.Row():\n with gr.Column(scale=0.5):\n image = gr.Image(type=\"pil\", tool='sketch', brush_radius=20)\n\n temperature = gr.Slider(\n minimum=0.1,\n maximum=1.5,\n value=0.6,\n step=0.1,\n interactive=True,\n label=\"Temperature\",\n )\n\n clear = gr.Button(\"Restart\")\n\n gr.Markdown(introduction)\n\n with gr.Column():\n chat_state = gr.State(value=None)\n img_list = gr.State(value=[])\n chatbot = gr.Chatbot(label='MiniGPT-v2')\n\n dataset = gr.Dataset(\n components=[gr.Textbox(visible=False)],\n samples=[['No Tag'], ['Grounding'], ['Refer'], ['Detection'], ['Identify'], ['VQA']],\n type=\"index\",\n label='Task Shortcuts',\n )\n task_inst = gr.Markdown('**Hint:** Upload your image and chat')\n with gr.Row():\n text_input.render()\n send = gr.Button(\"Send\", variant='primary', size='sm', scale=1)\n\n upload_flag = gr.State(value=0)\n replace_flag = gr.State(value=0)\n image.upload(image_upload_trigger, [upload_flag, replace_flag, img_list], [upload_flag, replace_flag])\n\n with gr.Row():\n with gr.Column():\n gr.Examples(examples=[\n [\"examples_v2/office.jpg\", \"[grounding] describe this image in detail\", upload_flag, replace_flag,\n img_list],\n [\"examples_v2/sofa.jpg\", \"[detection] sofas\", upload_flag, replace_flag, img_list],\n [\"examples_v2/2000x1372_wmkn_0012149409555.jpg\", \"[refer] the world cup\", upload_flag, replace_flag,\n img_list],\n [\"examples_v2/KFC-20-for-20-Nuggets.jpg\", \"[identify] what is this {<4><50><30><65>}\", upload_flag,\n replace_flag, img_list],\n ], inputs=[image, text_input, upload_flag, replace_flag, img_list], fn=example_trigger,\n outputs=[upload_flag, replace_flag])\n with gr.Column():\n gr.Examples(examples=[\n [\"examples_v2/glip_test.jpg\", \"[vqa] where should I hide in this room when playing hide and seek\",\n upload_flag, replace_flag, img_list],\n [\"examples_v2/float.png\", \"Please write a poem about the image\", upload_flag, replace_flag, img_list],\n [\"examples_v2/thief.png\", \"Is the weapon fateful\", upload_flag, replace_flag, img_list],\n [\"examples_v2/cockdial.png\", \"What might happen in this image in the next second\", upload_flag,\n replace_flag, img_list],\n ], inputs=[image, text_input, upload_flag, replace_flag, img_list], fn=example_trigger,\n outputs=[upload_flag, replace_flag])\n\n dataset.click(\n gradio_taskselect,\n inputs=[dataset],\n outputs=[text_input, task_inst],\n show_progress=\"hidden\",\n postprocess=False,\n queue=False,\n )\n\n text_input.submit(\n gradio_ask,\n [text_input, chatbot, chat_state, image, img_list, upload_flag, replace_flag],\n [text_input, chatbot, chat_state, img_list, upload_flag, replace_flag], queue=False\n ).success(\n gradio_stream_answer,\n [chatbot, chat_state, img_list, temperature],\n [chatbot, chat_state]\n ).success(\n gradio_visualize,\n [chatbot, image],\n [chatbot],\n queue=False,\n )\n\n send.click(\n gradio_ask,\n [text_input, chatbot, chat_state, image, img_list, upload_flag, replace_flag],\n [text_input, chatbot, chat_state, img_list, upload_flag, replace_flag], queue=False\n ).success(\n gradio_stream_answer,\n [chatbot, chat_state, img_list, temperature],\n [chatbot, chat_state]\n ).success(\n gradio_visualize,\n [chatbot, image],\n [chatbot],\n queue=False,\n )\n\n clear.click(gradio_reset, [chat_state, img_list], [chatbot, image, text_input, chat_state, img_list], queue=False)\n\ndemo.launch(share=True, enable_queue=True)\n" }, { "path": "environment.yml", "content": "name: minigptv\nchannels:\n - pytorch\n - defaults\n - anaconda\ndependencies:\n - python=3.9\n - cudatoolkit\n - pip\n - pip:\n - torch==2.0.0\n - torchaudio\n - torchvision\n - huggingface-hub==0.18.0\n - matplotlib==3.7.0\n - psutil==5.9.4\n - iopath\n - pyyaml==6.0\n - regex==2022.10.31\n - tokenizers==0.13.2\n - tqdm==4.64.1\n - transformers==4.30.0\n - timm==0.6.13\n - webdataset==0.2.48\n - omegaconf==2.3.0\n - opencv-python==4.7.0.72\n - decord==0.6.0\n - peft==0.2.0\n - sentence-transformers\n - gradio==3.47.1\n - accelerate==0.20.3\n - bitsandbytes==0.37.0\n - scikit-image\n - visual-genome\n - wandb\n" }, { "path": "eval_configs/minigpt4_eval.yaml", "content": "model:\n arch: minigpt4\n model_type: pretrain_vicuna0\n max_txt_len: 160\n end_sym: \"###\"\n low_resource: True\n prompt_template: '###Human: {} ###Assistant: '\n ckpt: 'please set this value to the path of pretrained checkpoint'\n\n\ndatasets:\n cc_sbu_align:\n vis_processor:\n train:\n name: \"blip2_image_eval\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n\nrun:\n task: image_text_pretrain\n" }, { "path": "eval_configs/minigpt4_llama2_eval.yaml", "content": "model:\n arch: minigpt4\n model_type: pretrain_llama2\n max_txt_len: 160\n end_sym: \"\"\n low_resource: True\n prompt_template: '[INST] {} [/INST] '\n ckpt: 'please set this value to the path of pretrained checkpoint'\n\n\ndatasets:\n cc_sbu_align:\n vis_processor:\n train:\n name: \"blip2_image_eval\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n\nrun:\n task: image_text_pretrain\n" }, { "path": "eval_configs/minigptv2_benchmark_evaluation.yaml", "content": "model:\n arch: minigpt_v2\n model_type: pretrain\n max_txt_len: 500\n end_sym: \"\"\n low_resource: False\n prompt_template: '[INST] {} [/INST]'\n llama_model: \"\"\n ckpt: \"\"\n lora_r: 64\n lora_alpha: 16\n\n\ndatasets:\n cc_sbu_align:\n vis_processor:\n train:\n name: \"blip2_image_eval\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n\nevaluation_datasets:\n refcoco:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n refcocog:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n refcoco+:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n gqa:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n okvqa:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n vizwiz:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n iconvqa:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n vsr:\n eval_file_path: cambridgeltl/vsr_zeroshot \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 10\n hm:\n eval_file_path: /path/to/eval/annotation/path \n img_path: /path/to/eval/image/path \n max_new_tokens: 20\n batch_size: 100\n\nrun:\n task: image_text_pretrain\n name: minigptv2_evaluation\n save_path: /path/to/save/folder_path\n\n \n\n \n\n" }, { "path": "eval_configs/minigptv2_eval.yaml", "content": "model:\n arch: minigpt_v2\n model_type: pretrain\n max_txt_len: 500\n end_sym: \"\"\n low_resource: True\n prompt_template: '[INST] {} [/INST]'\n ckpt: \"please set this value to the path of pretrained checkpoint\"\n lora_r: 64\n lora_alpha: 16\n\n\ndatasets:\n cc_sbu_align:\n vis_processor:\n train:\n name: \"blip2_image_eval\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n\nrun:\n task: image_text_pretrain\n" }, { "path": "eval_scripts/EVAL_README.md", "content": "## Evaluation Instruction for MiniGPT-v2\n\n### Data preparation\nImages download\nImage source | Download path\n--- | :---:\nOKVQA| annotations    images\ngqa | annotations    images \nhateful meme | images and annotations \niconqa | images and annotation\nvizwiz | images and annotation\nRefCOCO | annotations \nRefCOCO+ | annotations \nRefCOCOg | annotations \n\n### Evaluation dataset structure\n\n```\n${MINIGPTv2_EVALUATION_DATASET}\nβ”œβ”€β”€ gqa\nβ”‚ └── test_balanced_questions.json\nβ”‚ β”œβ”€β”€ testdev_balanced_questions.json\nβ”‚ β”œβ”€β”€ gqa_images\nβ”œβ”€β”€ hateful_meme\nβ”‚ └── hm_images\nβ”‚ β”œβ”€β”€ dev.jsonl\nβ”œβ”€β”€ iconvqa\nβ”‚ └── iconvqa_images\nβ”‚ β”œβ”€β”€ choose_text_val.json\nβ”œβ”€β”€ vizwiz\nβ”‚ └── vizwiz_images\nβ”‚ β”œβ”€β”€ val.json\nβ”œβ”€β”€ vsr\nβ”‚ └── vsr_images\nβ”œβ”€β”€ okvqa\nβ”‚ β”œβ”€β”€ okvqa_test_split.json\nβ”‚ β”œβ”€β”€ mscoco_val2014_annotations_clean.json\nβ”‚ β”œβ”€β”€ OpenEnded_mscoco_val2014_questions_clean.json\nβ”œβ”€β”€ refcoco\nβ”‚ └── instances.json\nβ”‚ β”œβ”€β”€ refs(google).p\nβ”‚ β”œβ”€β”€ refs(unc).p\nβ”œβ”€β”€ refcoco+\nβ”‚ └── instances.json\nβ”‚ β”œβ”€β”€ refs(unc).p\nβ”œβ”€β”€ refercocog\nβ”‚ └── instances.json\nβ”‚ β”œβ”€β”€ refs(google).p\nβ”‚ β”œβ”€β”€ refs(und).p\n...\n```\n\n\n### environment setup\n\n```\nexport PYTHONPATH=$PYTHONPATH:/path/to/directory/of/MiniGPT-4\n```\n\n### config file setup\n\nSet **llama_model** to the path of LLaMA model. \nSet **ckpt** to the path of our pretrained model. \nSet **eval_file_path** to the path of the annotation files for each evaluation data. \nSet **img_path** to the img_path for each evaluation dataset. \nSet **save_path** to the save_path for each evaluation dataset. \n\nin [eval_configs/minigptv2_benchmark_evaluation.yaml](../eval_configs/minigptv2_benchmark_evaluation.yaml) \n\n\n\n\n### start evalauting RefCOCO, RefCOCO+, RefCOCOg\nport=port_number \ncfg_path=/path/to/eval_configs/minigptv2_benchmark_evaluation.yaml \n\ndataset names: \n| refcoco | refcoco+ | refcocog |\n| ------- | -------- | -------- |\n\n```\ntorchrun --master-port ${port} --nproc_per_node 1 eval_ref.py \\\n --cfg-path ${cfg_path} --dataset refcoco,refcoco+,refcocog --resample\n```\n\n\n### start evaluating visual question answering\n\nport=port_number \ncfg_path=/path/to/eval_configs/minigptv2_benchmark_evaluation.yaml \n\ndataset names: \n| okvqa | vizwiz | iconvqa | gqa | vsr | hm |\n| ------- | -------- | -------- |-------- | -------- | -------- |\n\n\n```\ntorchrun --master-port ${port} --nproc_per_node 1 eval_vqa.py \\\n --cfg-path ${cfg_path} --dataset okvqa,vizwiz,iconvqa,gqa,vsr,hm\n```\n\n\n\n\n" }, { "path": "eval_scripts/eval_ref.py", "content": "import os\nimport re\nimport json\nimport argparse\nfrom collections import defaultdict\nimport random\nimport numpy as np\nfrom PIL import Image\nfrom tqdm import tqdm\nimport torch\nfrom torch.utils.data import DataLoader\nfrom minigpt4.common.config import Config\nfrom minigpt4.common.eval_utils import prepare_texts, init_model, eval_parser, computeIoU\nfrom minigpt4.conversation.conversation import CONV_VISION_minigptv2\n\nfrom minigpt4.datasets.datasets.coco_caption import RefCOCOEvalData\n\ndef list_of_str(arg):\n return list(map(str, arg.split(',')))\n\nparser = eval_parser()\nparser.add_argument(\"--dataset\", type=list_of_str, default='refcoco', help=\"dataset to evaluate\")\nparser.add_argument(\"--res\", type=float, default=100.0, help=\"resolution used in refcoco\")\nparser.add_argument(\"--resample\", action='store_true', help=\"resolution used in refcoco\")\nargs = parser.parse_args()\n\ncfg = Config(args)\n\neval_dict = {'refcoco': ['val','testA','testB'], \n 'refcoco+': ['val','testA','testB'],\n 'refcocog': ['val','test']}\n\n\nmodel, vis_processor = init_model(args)\nmodel.eval()\nCONV_VISION = CONV_VISION_minigptv2\nconv_temp = CONV_VISION.copy()\nconv_temp.system = \"\"\n\n# \nmodel.eval()\nsave_path = cfg.run_cfg.save_path\n\n\n\nfor dataset in args.dataset:\n for split in eval_dict[dataset]:\n\n eval_file_path = cfg.evaluation_datasets_cfg[dataset][\"eval_file_path\"]\n img_path = cfg.evaluation_datasets_cfg[dataset][\"img_path\"]\n batch_size = cfg.evaluation_datasets_cfg[dataset][\"batch_size\"]\n max_new_tokens = cfg.evaluation_datasets_cfg[dataset][\"max_new_tokens\"]\n\n with open(os.path.join(eval_file_path,f\"{dataset}/{dataset}_{split}.json\"), 'r') as f:\n refcoco = json.load(f)\n\n data = RefCOCOEvalData(refcoco, vis_processor, img_path)\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n minigpt4_predict = defaultdict(list)\n resamples = []\n\n for images, questions, img_ids in tqdm(eval_dataloader):\n texts = prepare_texts(questions, conv_temp) # warp the texts with conversation template\n answers = model.generate(images, texts, max_new_tokens=max_new_tokens, do_sample=False)\n for answer, img_id, question in zip(answers, img_ids, questions):\n answer = answer.replace(\"\",\"\").replace(\" \",\"\").strip()\n pattern = r'\\{<\\d{1,3}><\\d{1,3}><\\d{1,3}><\\d{1,3}>\\}'\n if re.match(pattern, answer):\n minigpt4_predict[img_id].append(answer)\n else:\n resamples.append({'img_id': img_id, 'sents': [question.replace('[refer] give me the location of','').strip()]})\n if args.resample:\n for i in range(20):\n data = RefCOCOEvalData(resamples, vis_processor, img_path)\n resamples = []\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n for images, questions, img_ids in tqdm(eval_dataloader):\n texts = prepare_texts(questions, conv_temp) # warp the texts with conversation template\n answers = model.generate(images, texts, max_new_tokens=max_new_tokens, do_sample=False)\n for answer, img_id, question in zip(answers, img_ids, questions):\n answer = answer.replace(\"\",\"\").replace(\" \",\"\").strip()\n pattern = r'\\{<\\d{1,3}><\\d{1,3}><\\d{1,3}><\\d{1,3}>\\}'\n if re.match(pattern, answer) or i == 4:\n minigpt4_predict[img_id].append(answer)\n else:\n resamples.append({'img_id': img_id, 'sents': [question.replace('[refer] give me the location of','').strip()]})\n \n if len(resamples) == 0:\n break\n \n file_save_path = os.path.join(save_path,f\"{args.dataset}_{split}.json\")\n with open(file_save_path,'w') as f:\n json.dump(minigpt4_predict, f)\n\n count=0\n total=len(refcoco)\n res=args.res\n refcoco_dict = defaultdict()\n for item in refcoco:\n refcoco_dict[item['img_id']] = item\n for img_id in refcoco_dict:\n item = refcoco_dict[img_id]\n bbox = item['bbox']\n outputs = minigpt4_predict[img_id]\n for output in outputs:\n try:\n integers = re.findall(r'\\d+', output)\n pred_bbox = [int(num) for num in integers]\n height = item['height']\n width = item['width']\n pred_bbox[0] = pred_bbox[0] / res * width\n pred_bbox[1] = pred_bbox[1] / res * height\n pred_bbox[2] = pred_bbox[2] / res * width\n pred_bbox[3] = pred_bbox[3] / res * height\n\n gt_bbox = [0,0,0,0]\n gt_bbox[0] = bbox[0]\n gt_bbox[1] = bbox[1]\n gt_bbox[2] = bbox[0] + bbox[2]\n gt_bbox[3] = bbox[1] + bbox[3]\n\n iou_score = computeIoU(pred_bbox, gt_bbox)\n if iou_score > 0.5:\n count+=1\n except:\n continue\n \n print(f'{dataset} {split}:', count / total * 100, flush=True)\n" }, { "path": "eval_scripts/eval_vqa.py", "content": "import os\nimport re\nimport json\nimport argparse\nfrom collections import defaultdict\n\nimport numpy as np\nfrom PIL import Image\nfrom tqdm import tqdm\nimport torch\nfrom torch.utils.data import DataLoader\nfrom datasets import load_dataset\n\n\nfrom minigpt4.datasets.datasets.vqa_datasets import OKVQAEvalData,VizWizEvalData,IconQAEvalData,GQAEvalData,VSREvalData,HMEvalData\nfrom minigpt4.common.vqa_tools.VQA.PythonHelperTools.vqaTools.vqa import VQA\nfrom minigpt4.common.vqa_tools.VQA.PythonEvaluationTools.vqaEvaluation.vqaEval import VQAEval\n\nfrom minigpt4.common.eval_utils import prepare_texts, init_model, eval_parser\nfrom minigpt4.conversation.conversation import CONV_VISION_minigptv2\nfrom minigpt4.common.config import Config\n\n\ndef list_of_str(arg):\n return list(map(str, arg.split(',')))\n\nparser = eval_parser()\nparser.add_argument(\"--dataset\", type=list_of_str, default='refcoco', help=\"dataset to evaluate\")\nargs = parser.parse_args()\ncfg = Config(args)\n\n\n\nmodel, vis_processor = init_model(args)\nconv_temp = CONV_VISION_minigptv2.copy()\nconv_temp.system = \"\"\nmodel.eval()\nsave_path = cfg.run_cfg.save_path\n\n\nif 'okvqa' in args.dataset:\n\n eval_file_path = cfg.evaluation_datasets_cfg[\"okvqa\"][\"eval_file_path\"]\n img_path = cfg.evaluation_datasets_cfg[\"okvqa\"][\"img_path\"]\n batch_size = cfg.evaluation_datasets_cfg[\"okvqa\"][\"batch_size\"]\n max_new_tokens = cfg.evaluation_datasets_cfg[\"okvqa\"][\"max_new_tokens\"]\n \n\n evaluation_annntation_path = os.path.join(eval_file_path, \"okvqa_test_split.json\")\n with open(evaluation_annntation_path) as f:\n ok_vqa_test_split = json.load(f)\n\n data = OKVQAEvalData(ok_vqa_test_split, vis_processor, img_path)\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n minigpt4_predict = []\n\n for images, questions, question_ids, img_ids in eval_dataloader:\n texts = prepare_texts(questions, conv_temp) # warp the texts with conversation template\n answers = model.generate(images, texts, max_new_tokens=max_new_tokens, do_sample=False)\n\n for answer, question_id, question, img_id in zip(answers, question_ids, questions, img_ids):\n result = dict()\n answer = answer.lower().replace('','').strip()\n result['answer'] = answer\n result['question_id'] = int(question_id)\n minigpt4_predict.append(result)\n\n file_save_path= os.path.join(save_path,\"okvqa.json\")\n with open(file_save_path,'w') as f:\n json.dump(minigpt4_predict, f)\n\n annFile = os.path.join(eval_file_path,\"mscoco_val2014_annotations_clean.json\")\n quesFile = os.path.join(eval_file_path,\"OpenEnded_mscoco_val2014_questions_clean.json\" )\n\n vqa = VQA(annFile, quesFile)\n vqaRes = vqa.loadRes(file_save_path, quesFile)\n\n vqaEval = VQAEval(vqa, vqaRes, n=2)\n vqaEval.evaluate()\n print (\"Overall OKVQA Accuracy is: %.02f\\n\" %(vqaEval.accuracy['overall']), flush=True)\n\nif 'vizwiz' in args.dataset:\n\n eval_file_path = cfg.evaluation_datasets_cfg[\"vizwiz\"][\"eval_file_path\"]\n img_path = cfg.evaluation_datasets_cfg[\"vizwiz\"][\"img_path\"]\n batch_size = cfg.evaluation_datasets_cfg[\"vizwiz\"][\"batch_size\"]\n max_new_tokens = cfg.evaluation_datasets_cfg[\"vizwiz\"][\"max_new_tokens\"]\n\n vizwiz = json.load(open(eval_file_path, 'r'))\n\n data = VizWizEvalData(vizwiz, vis_processor, img_path)\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n minigpt4_predict = []\n total_acc = []\n for images, texts, gt_answers in tqdm(eval_dataloader):\n texts = prepare_texts(texts, conv_temp) # warp the texts with conversation template\n with torch.no_grad():\n answers = model.generate(images, texts, max_new_tokens=max_new_tokens, do_sample=False,repetition_penalty=1.0)\n\n for answer, gt_answer in zip(answers, gt_answers):\n result = dict()\n result['answer'] = answer.replace('','').strip()\n minigpt4_predict.append(result)\n count=0\n gt_answer = gt_answer.split('_')\n for gt in gt_answer:\n if gt.lower() == answer.lower():\n count += 1\n acc = min(count/3.0, 1.0)\n total_acc.append(acc)\n \n file_save_path = os.path.join(save_path, \"vizwiz.json\")\n with open(file_save_path,'w') as f:\n json.dump(minigpt4_predict, f)\n print('vizwiz Acc: ', np.average(total_acc)* 100.0, flush=True)\n\n\nif 'iconvqa' in args.dataset:\n\n eval_file_path = cfg.evaluation_datasets_cfg[\"iconvqa\"][\"eval_file_path\"]\n img_path = cfg.evaluation_datasets_cfg[\"iconvqa\"][\"img_path\"]\n batch_size = cfg.evaluation_datasets_cfg[\"iconvqa\"][\"batch_size\"]\n max_new_tokens = cfg.evaluation_datasets_cfg[\"iconvqa\"][\"max_new_tokens\"]\n\n iconqa_text_val = json.load(open(eval_file_path,\"r\"))\n\n data = IconQAEvalData(iconqa_text_val, vis_processor, img_path)\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n\n count = 0\n for images, texts, candidates, answers in tqdm(eval_dataloader):\n candidates = [candidate.split('_') for candidate in candidates]\n num_cand = [len(candidate) for candidate in candidates]\n for candidate in candidates:\n candidate.extend(['none'] * (max(num_cand) - len(candidate)))\n candidates = [list(x) for x in zip(*candidates)]\n instructions = [\"[INST] {} [/INST]\".format(text) for text in texts]\n answer_ranks = model.multi_select(images, instructions, candidates, num_cand=num_cand)\n for idx, answer in enumerate(answers):\n if answer_ranks[idx][0] == answer:\n count += 1\n\n print('iconqa Acc: ', count / len(iconqa_text_val) * 100.0, flush=True)\n\n\nif 'gqa' in args.dataset:\n\n eval_file_path = cfg.evaluation_datasets_cfg[\"gqa\"][\"eval_file_path\"]\n img_path = cfg.evaluation_datasets_cfg[\"gqa\"][\"img_path\"]\n batch_size = cfg.evaluation_datasets_cfg[\"gqa\"][\"batch_size\"]\n max_new_tokens = cfg.evaluation_datasets_cfg[\"gqa\"][\"max_new_tokens\"]\n\n gqa = json.load(open(eval_file_path))\n data = GQAEvalData(gqa, vis_processor, img_path)\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n count=0\n total=0\n minigpt4_predict = []\n for images, texts, labels in tqdm(eval_dataloader):\n texts = prepare_texts(texts, conv_temp) # warp the texts with conversation template\n answers = model.generate(images, texts, max_new_tokens=max_new_tokens, do_sample=False)\n\n for answer, label in zip(answers, labels):\n result = dict()\n result['pred'] = answer.lower().replace('','').strip()\n result['gt'] = label\n minigpt4_predict.append(result)\n if answer.lower() == label:\n count+=1\n total+=1\n print('gqa val:', count / total * 100, flush=True)\n\n file_save_path = os.path.join(save_path, \"gqa.json\")\n with open(file_save_path,'w') as f:\n json.dump(minigpt4_predict, f)\n\nif 'vsr' in args.dataset:\n\n img_path = cfg.evaluation_datasets_cfg[\"vsr\"][\"img_path\"]\n batch_size = cfg.evaluation_datasets_cfg[\"vsr\"][\"batch_size\"]\n max_new_tokens = cfg.evaluation_datasets_cfg[\"vsr\"][\"max_new_tokens\"]\n\n annotation = load_dataset(\"cambridgeltl/vsr_zeroshot\", split='test')\n data = VSREvalData(annotation, vis_processor, img_path)\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n count=0\n total=0\n\n minigpt4_predict = []\n\n for images, texts, labels in tqdm(eval_dataloader):\n texts = prepare_texts(texts, conv_temp) # warp the texts with conversation template\n answers = model.generate(images, texts, max_new_tokens=max_new_tokens, do_sample=False)\n\n for answer, label in zip(answers, labels):\n result = dict()\n result['pred'] = answer.replace('','').strip()\n result['gt'] = label\n minigpt4_predict.append(result)\n if answer.lower() == label.lower():\n count+=1\n total+=1\n print('vsr test:', count / total * 100, flush=True)\n file_save_path = os.path.join(save_path,\"vsr.json\")\n with open(file_save_path,'w') as f:\n json.dump(minigpt4_predict, f)\n\nif 'hm' in args.dataset:\n\n eval_file_path = cfg.evaluation_datasets_cfg[\"hm\"][\"eval_file_path\"]\n img_path = cfg.evaluation_datasets_cfg[\"hm\"][\"img_path\"]\n batch_size = cfg.evaluation_datasets_cfg[\"hm\"][\"batch_size\"]\n max_new_tokens = cfg.evaluation_datasets_cfg[\"hm\"][\"max_new_tokens\"]\n\n annotation = []\n with open(eval_file_path, 'r') as jsonl_file:\n for line in jsonl_file:\n json_obj = json.loads(line)\n annotation.append(json_obj)\n\n data = HMEvalData(annotation, vis_processor, img_path)\n eval_dataloader = DataLoader(data, batch_size=batch_size, shuffle=False)\n count=0\n total=0\n\n minigpt4_predict = []\n\n for images, texts, labels in tqdm(eval_dataloader):\n texts = prepare_texts(texts, conv_temp) # warp the texts with conversation template\n \n answers = model.generate(images, texts, max_new_tokens=max_new_tokens, do_sample=False)\n\n for answer, label in zip(answers, labels):\n result = dict()\n if answer.lower().strip() ==\"yes\":\n answer=1\n elif answer.lower().strip()==\"no\":\n answer=0\n else:\n print(\"non-matching answer\",answer)\n\n result['pred'] = answer\n result['gt'] = int(label)\n minigpt4_predict.append(result)\n if answer == label:\n count+=1\n total+=1\n\n print('hm val:', count / total * 100, flush=True)\n file_save_path = os.path.join(save_path, \"hm.json\")\n with open(file_save_path,'w') as f:\n json.dump(minigpt4_predict, f)\n" }, { "path": "minigpt4/__init__.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport os\nimport sys\n\nfrom omegaconf import OmegaConf\n\nfrom minigpt4.common.registry import registry\n\nfrom minigpt4.datasets.builders import *\nfrom minigpt4.models import *\nfrom minigpt4.processors import *\nfrom minigpt4.tasks import *\n\n\nroot_dir = os.path.dirname(os.path.abspath(__file__))\ndefault_cfg = OmegaConf.load(os.path.join(root_dir, \"configs/default.yaml\"))\n\nregistry.register_path(\"library_root\", root_dir)\nrepo_root = os.path.join(root_dir, \"..\")\nregistry.register_path(\"repo_root\", repo_root)\ncache_root = os.path.join(repo_root, default_cfg.env.cache_root)\nregistry.register_path(\"cache_root\", cache_root)\n\nregistry.register(\"MAX_INT\", sys.maxsize)\nregistry.register(\"SPLIT_NAMES\", [\"train\", \"val\", \"test\"])\n" }, { "path": "minigpt4/common/__init__.py", "content": "" }, { "path": "minigpt4/common/config.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport logging\nimport json\nfrom typing import Dict\n\nfrom omegaconf import OmegaConf\nfrom minigpt4.common.registry import registry\n\n\nclass Config:\n def __init__(self, args):\n self.config = {}\n\n self.args = args\n\n # Register the config and configuration for setup\n registry.register(\"configuration\", self)\n\n user_config = self._build_opt_list(self.args.options)\n\n config = OmegaConf.load(self.args.cfg_path)\n\n runner_config = self.build_runner_config(config)\n model_config = self.build_model_config(config, **user_config)\n dataset_config = self.build_dataset_config(config)\n evaluation_dataset_config = self.build_evaluation_dataset_config(config)\n\n # Validate the user-provided runner configuration\n # model and dataset configuration are supposed to be validated by the respective classes\n # [TODO] validate the model/dataset configuration\n # self._validate_runner_config(runner_config)\n\n # Override the default configuration with user options.\n self.config = OmegaConf.merge(\n runner_config, model_config, dataset_config,evaluation_dataset_config, user_config\n )\n\n def _validate_runner_config(self, runner_config):\n \"\"\"\n This method validates the configuration, such that\n 1) all the user specified options are valid;\n 2) no type mismatches between the user specified options and the config.\n \"\"\"\n runner_config_validator = create_runner_config_validator()\n runner_config_validator.validate(runner_config)\n\n def _build_opt_list(self, opts):\n opts_dot_list = self._convert_to_dot_list(opts)\n return OmegaConf.from_dotlist(opts_dot_list)\n\n @staticmethod\n def build_model_config(config, **kwargs):\n model = config.get(\"model\", None)\n assert model is not None, \"Missing model configuration file.\"\n\n model_cls = registry.get_model_class(model.arch)\n assert model_cls is not None, f\"Model '{model.arch}' has not been registered.\"\n\n model_type = kwargs.get(\"model.model_type\", None)\n if not model_type:\n model_type = model.get(\"model_type\", None)\n # else use the model type selected by user.\n\n assert model_type is not None, \"Missing model_type.\"\n\n model_config_path = model_cls.default_config_path(model_type=model_type)\n\n model_config = OmegaConf.create()\n # hierarchy override, customized config > default config\n model_config = OmegaConf.merge(\n model_config,\n OmegaConf.load(model_config_path),\n {\"model\": config[\"model\"]},\n )\n\n return model_config\n\n @staticmethod\n def build_runner_config(config):\n return {\"run\": config.run}\n\n @staticmethod\n def build_dataset_config(config):\n datasets = config.get(\"datasets\", None)\n if datasets is None:\n raise KeyError(\n \"Expecting 'datasets' as the root key for dataset configuration.\"\n )\n\n dataset_config = OmegaConf.create()\n\n for dataset_name in datasets:\n builder_cls = registry.get_builder_class(dataset_name)\n\n dataset_config_type = datasets[dataset_name].get(\"type\", \"default\")\n dataset_config_path = builder_cls.default_config_path(\n type=dataset_config_type\n )\n\n # hierarchy override, customized config > default config\n dataset_config = OmegaConf.merge(\n dataset_config,\n OmegaConf.load(dataset_config_path),\n {\"datasets\": {dataset_name: config[\"datasets\"][dataset_name]}},\n )\n\n return dataset_config\n\n\n @staticmethod\n def build_evaluation_dataset_config(config):\n datasets = config.get(\"evaluation_datasets\", None)\n # if datasets is None:\n # raise KeyError(\n # \"Expecting 'datasets' as the root key for dataset configuration.\"\n # )\n\n dataset_config = OmegaConf.create()\n\n if datasets is not None:\n for dataset_name in datasets:\n builder_cls = registry.get_builder_class(dataset_name)\n\n # hierarchy override, customized config > default config\n dataset_config = OmegaConf.merge(\n dataset_config,\n {\"evaluation_datasets\": {dataset_name: config[\"evaluation_datasets\"][dataset_name]}},\n )\n\n return dataset_config\n\n def _convert_to_dot_list(self, opts):\n if opts is None:\n opts = []\n\n if len(opts) == 0:\n return opts\n\n has_equal = opts[0].find(\"=\") != -1\n\n if has_equal:\n return opts\n\n return [(opt + \"=\" + value) for opt, value in zip(opts[0::2], opts[1::2])]\n\n def get_config(self):\n return self.config\n\n @property\n def run_cfg(self):\n return self.config.run\n\n @property\n def datasets_cfg(self):\n return self.config.datasets\n\n @property\n def evaluation_datasets_cfg(self):\n return self.config.evaluation_datasets\n\n @property\n def model_cfg(self):\n return self.config.model\n\n def pretty_print(self):\n logging.info(\"\\n===== Running Parameters =====\")\n logging.info(self._convert_node_to_json(self.config.run))\n\n logging.info(\"\\n====== Dataset Attributes ======\")\n datasets = self.config.datasets\n\n for dataset in datasets:\n if dataset in self.config.datasets:\n logging.info(f\"\\n======== {dataset} =======\")\n dataset_config = self.config.datasets[dataset]\n logging.info(self._convert_node_to_json(dataset_config))\n else:\n logging.warning(f\"No dataset named '{dataset}' in config. Skipping\")\n\n logging.info(f\"\\n====== Model Attributes ======\")\n logging.info(self._convert_node_to_json(self.config.model))\n\n def _convert_node_to_json(self, node):\n container = OmegaConf.to_container(node, resolve=True)\n return json.dumps(container, indent=4, sort_keys=True)\n\n def to_dict(self):\n return OmegaConf.to_container(self.config)\n\n\ndef node_to_dict(node):\n return OmegaConf.to_container(node)\n\n\nclass ConfigValidator:\n \"\"\"\n This is a preliminary implementation to centralize and validate the configuration.\n May be altered in the future.\n\n A helper class to validate configurations from yaml file.\n\n This serves the following purposes:\n 1. Ensure all the options in the yaml are defined, raise error if not.\n 2. when type mismatches are found, the validator will raise an error.\n 3. a central place to store and display helpful messages for supported configurations.\n\n \"\"\"\n\n class _Argument:\n def __init__(self, name, choices=None, type=None, help=None):\n self.name = name\n self.val = None\n self.choices = choices\n self.type = type\n self.help = help\n\n def __str__(self):\n s = f\"{self.name}={self.val}\"\n if self.type is not None:\n s += f\", ({self.type})\"\n if self.choices is not None:\n s += f\", choices: {self.choices}\"\n if self.help is not None:\n s += f\", ({self.help})\"\n return s\n\n def __init__(self, description):\n self.description = description\n\n self.arguments = dict()\n\n self.parsed_args = None\n\n def __getitem__(self, key):\n assert self.parsed_args is not None, \"No arguments parsed yet.\"\n\n return self.parsed_args[key]\n\n def __str__(self) -> str:\n return self.format_help()\n\n def add_argument(self, *args, **kwargs):\n \"\"\"\n Assume the first argument is the name of the argument.\n \"\"\"\n self.arguments[args[0]] = self._Argument(*args, **kwargs)\n\n def validate(self, config=None):\n \"\"\"\n Convert yaml config (dict-like) to list, required by argparse.\n \"\"\"\n for k, v in config.items():\n assert (\n k in self.arguments\n ), f\"\"\"{k} is not a valid argument. Support arguments are {self.format_arguments()}.\"\"\"\n\n if self.arguments[k].type is not None:\n try:\n self.arguments[k].val = self.arguments[k].type(v)\n except ValueError:\n raise ValueError(f\"{k} is not a valid {self.arguments[k].type}.\")\n\n if self.arguments[k].choices is not None:\n assert (\n v in self.arguments[k].choices\n ), f\"\"\"{k} must be one of {self.arguments[k].choices}.\"\"\"\n\n return config\n\n def format_arguments(self):\n return str([f\"{k}\" for k in sorted(self.arguments.keys())])\n\n def format_help(self):\n # description + key-value pair string for each argument\n help_msg = str(self.description)\n return help_msg + \", available arguments: \" + self.format_arguments()\n\n def print_help(self):\n # display help message\n print(self.format_help())\n\n\ndef create_runner_config_validator():\n validator = ConfigValidator(description=\"Runner configurations\")\n\n validator.add_argument(\n \"runner\",\n type=str,\n choices=[\"runner_base\", \"runner_iter\"],\n help=\"\"\"Runner to use. The \"runner_base\" uses epoch-based training while iter-based\n runner runs based on iters. Default: runner_base\"\"\",\n )\n # add argumetns for training dataset ratios\n validator.add_argument(\n \"train_dataset_ratios\",\n type=Dict[str, float],\n help=\"\"\"Ratios of training dataset. This is used in iteration-based runner.\n Do not support for epoch-based runner because how to define an epoch becomes tricky.\n Default: None\"\"\",\n )\n validator.add_argument(\n \"max_iters\",\n type=float,\n help=\"Maximum number of iterations to run.\",\n )\n validator.add_argument(\n \"max_epoch\",\n type=int,\n help=\"Maximum number of epochs to run.\",\n )\n # add arguments for iters_per_inner_epoch\n validator.add_argument(\n \"iters_per_inner_epoch\",\n type=float,\n help=\"Number of iterations per inner epoch. This is required when runner is runner_iter.\",\n )\n lr_scheds_choices = registry.list_lr_schedulers()\n validator.add_argument(\n \"lr_sched\",\n type=str,\n choices=lr_scheds_choices,\n help=\"Learning rate scheduler to use, from {}\".format(lr_scheds_choices),\n )\n task_choices = registry.list_tasks()\n validator.add_argument(\n \"task\",\n type=str,\n choices=task_choices,\n help=\"Task to use, from {}\".format(task_choices),\n )\n # add arguments for init_lr\n validator.add_argument(\n \"init_lr\",\n type=float,\n help=\"Initial learning rate. This will be the learning rate after warmup and before decay.\",\n )\n # add arguments for min_lr\n validator.add_argument(\n \"min_lr\",\n type=float,\n help=\"Minimum learning rate (after decay).\",\n )\n # add arguments for warmup_lr\n validator.add_argument(\n \"warmup_lr\",\n type=float,\n help=\"Starting learning rate for warmup.\",\n )\n # add arguments for learning rate decay rate\n validator.add_argument(\n \"lr_decay_rate\",\n type=float,\n help=\"Learning rate decay rate. Required if using a decaying learning rate scheduler.\",\n )\n # add arguments for weight decay\n validator.add_argument(\n \"weight_decay\",\n type=float,\n help=\"Weight decay rate.\",\n )\n # add arguments for training batch size\n validator.add_argument(\n \"batch_size_train\",\n type=int,\n help=\"Training batch size.\",\n )\n # add arguments for evaluation batch size\n validator.add_argument(\n \"batch_size_eval\",\n type=int,\n help=\"Evaluation batch size, including validation and testing.\",\n )\n # add arguments for number of workers for data loading\n validator.add_argument(\n \"num_workers\",\n help=\"Number of workers for data loading.\",\n )\n # add arguments for warm up steps\n validator.add_argument(\n \"warmup_steps\",\n type=int,\n help=\"Number of warmup steps. Required if a warmup schedule is used.\",\n )\n # add arguments for random seed\n validator.add_argument(\n \"seed\",\n type=int,\n help=\"Random seed.\",\n )\n # add arguments for output directory\n validator.add_argument(\n \"output_dir\",\n type=str,\n help=\"Output directory to save checkpoints and logs.\",\n )\n # add arguments for whether only use evaluation\n validator.add_argument(\n \"evaluate\",\n help=\"Whether to only evaluate the model. If true, training will not be performed.\",\n )\n # add arguments for splits used for training, e.g. [\"train\", \"val\"]\n validator.add_argument(\n \"train_splits\",\n type=list,\n help=\"Splits to use for training.\",\n )\n # add arguments for splits used for validation, e.g. [\"val\"]\n validator.add_argument(\n \"valid_splits\",\n type=list,\n help=\"Splits to use for validation. If not provided, will skip the validation.\",\n )\n # add arguments for splits used for testing, e.g. [\"test\"]\n validator.add_argument(\n \"test_splits\",\n type=list,\n help=\"Splits to use for testing. If not provided, will skip the testing.\",\n )\n # add arguments for accumulating gradient for iterations\n validator.add_argument(\n \"accum_grad_iters\",\n type=int,\n help=\"Number of iterations to accumulate gradient for.\",\n )\n\n # ====== distributed training ======\n validator.add_argument(\n \"device\",\n type=str,\n choices=[\"cpu\", \"cuda\"],\n help=\"Device to use. Support 'cuda' or 'cpu' as for now.\",\n )\n validator.add_argument(\n \"world_size\",\n type=int,\n help=\"Number of processes participating in the job.\",\n )\n validator.add_argument(\"dist_url\", type=str)\n validator.add_argument(\"distributed\", type=bool)\n # add arguments to opt using distributed sampler during evaluation or not\n validator.add_argument(\n \"use_dist_eval_sampler\",\n type=bool,\n help=\"Whether to use distributed sampler during evaluation or not.\",\n )\n\n # ====== task specific ======\n # generation task specific arguments\n # add arguments for maximal length of text output\n validator.add_argument(\n \"max_len\",\n type=int,\n help=\"Maximal length of text output.\",\n )\n # add arguments for minimal length of text output\n validator.add_argument(\n \"min_len\",\n type=int,\n help=\"Minimal length of text output.\",\n )\n # add arguments number of beams\n validator.add_argument(\n \"num_beams\",\n type=int,\n help=\"Number of beams used for beam search.\",\n )\n\n # vqa task specific arguments\n # add arguments for number of answer candidates\n validator.add_argument(\n \"num_ans_candidates\",\n type=int,\n help=\"\"\"For ALBEF and BLIP, these models first rank answers according to likelihood to select answer candidates.\"\"\",\n )\n # add arguments for inference method\n validator.add_argument(\n \"inference_method\",\n type=str,\n choices=[\"genearte\", \"rank\"],\n help=\"\"\"Inference method to use for question answering. If rank, requires a answer list.\"\"\",\n )\n\n # ====== model specific ======\n validator.add_argument(\n \"k_test\",\n type=int,\n help=\"Number of top k most similar samples from ITC/VTC selection to be tested.\",\n )\n\n return validator\n" }, { "path": "minigpt4/common/dist_utils.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport datetime\nimport functools\nimport os\n\nimport torch\nimport torch.distributed as dist\nimport timm.models.hub as timm_hub\n\n\ndef setup_for_distributed(is_master):\n \"\"\"\n This function disables printing when not in master process\n \"\"\"\n import builtins as __builtin__\n\n builtin_print = __builtin__.print\n\n def print(*args, **kwargs):\n force = kwargs.pop(\"force\", False)\n if is_master or force:\n builtin_print(*args, **kwargs)\n\n __builtin__.print = print\n\n\ndef is_dist_avail_and_initialized():\n if not dist.is_available():\n return False\n if not dist.is_initialized():\n return False\n return True\n\n\ndef get_world_size():\n if not is_dist_avail_and_initialized():\n return 1\n return dist.get_world_size()\n\n\ndef get_rank():\n if not is_dist_avail_and_initialized():\n return 0\n return dist.get_rank()\n\n\ndef is_main_process():\n return get_rank() == 0\n\n\ndef init_distributed_mode(args):\n if args.distributed is False:\n print(\"Not using distributed mode\")\n return\n elif \"RANK\" in os.environ and \"WORLD_SIZE\" in os.environ:\n args.rank = int(os.environ[\"RANK\"])\n args.world_size = int(os.environ[\"WORLD_SIZE\"])\n args.gpu = int(os.environ[\"LOCAL_RANK\"])\n elif \"SLURM_PROCID\" in os.environ:\n args.rank = int(os.environ[\"SLURM_PROCID\"])\n args.gpu = args.rank % torch.cuda.device_count()\n else:\n print(\"Not using distributed mode\")\n args.distributed = False\n return\n\n args.distributed = True\n\n torch.cuda.set_device(args.gpu)\n args.dist_backend = \"nccl\"\n print(\n \"| distributed init (rank {}, world {}): {}\".format(\n args.rank, args.world_size, args.dist_url\n ),\n flush=True,\n )\n torch.distributed.init_process_group(\n backend=args.dist_backend,\n init_method=args.dist_url,\n world_size=args.world_size,\n rank=args.rank,\n timeout=datetime.timedelta(\n days=365\n ), # allow auto-downloading and de-compressing\n )\n torch.distributed.barrier()\n setup_for_distributed(args.rank == 0)\n\n\ndef get_dist_info():\n if torch.__version__ < \"1.0\":\n initialized = dist._initialized\n else:\n initialized = dist.is_initialized()\n if initialized:\n rank = dist.get_rank()\n world_size = dist.get_world_size()\n else: # non-distributed training\n rank = 0\n world_size = 1\n return rank, world_size\n\n\ndef main_process(func):\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n rank, _ = get_dist_info()\n if rank == 0:\n return func(*args, **kwargs)\n\n return wrapper\n\n\ndef download_cached_file(url, check_hash=True, progress=False):\n \"\"\"\n Download a file from a URL and cache it locally. If the file already exists, it is not downloaded again.\n If distributed, only the main process downloads the file, and the other processes wait for the file to be downloaded.\n \"\"\"\n\n def get_cached_file_path():\n # a hack to sync the file path across processes\n parts = torch.hub.urlparse(url)\n filename = os.path.basename(parts.path)\n cached_file = os.path.join(timm_hub.get_cache_dir(), filename)\n\n return cached_file\n\n if is_main_process():\n timm_hub.download_cached_file(url, check_hash, progress)\n\n if is_dist_avail_and_initialized():\n dist.barrier()\n\n return get_cached_file_path()\n" }, { "path": "minigpt4/common/eval_utils.py", "content": "import argparse\nimport numpy as np\nfrom nltk.translate.bleu_score import sentence_bleu\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.common.config import Config\n\n# imports modules for registration\nfrom minigpt4.datasets.builders import *\nfrom minigpt4.models import *\nfrom minigpt4.processors import *\nfrom minigpt4.runners import *\nfrom minigpt4.tasks import *\n\n\n\ndef eval_parser():\n parser = argparse.ArgumentParser(description=\"Demo\")\n parser.add_argument(\"--cfg-path\", required=True, help=\"path to configuration file.\")\n parser.add_argument(\"--name\", type=str, default='A2', help=\"evaluation name\")\n parser.add_argument(\"--ckpt\", type=str, help=\"path to configuration file.\")\n parser.add_argument(\"--eval_opt\", type=str, default='all', help=\"path to configuration file.\")\n parser.add_argument(\"--max_new_tokens\", type=int, default=10, help=\"max number of generated tokens\")\n parser.add_argument(\"--batch_size\", type=int, default=32)\n parser.add_argument(\"--lora_r\", type=int, default=64, help=\"lora rank of the model\")\n parser.add_argument(\"--lora_alpha\", type=int, default=16, help=\"lora alpha\")\n parser.add_argument(\n \"--options\",\n nargs=\"+\",\n help=\"override some settings in the used config, the key-value pair \"\n \"in xxx=yyy format will be merged into config file (deprecate), \"\n \"change to --cfg-options instead.\",\n )\n return parser\n\n\ndef prepare_texts(texts, conv_temp):\n convs = [conv_temp.copy() for _ in range(len(texts))]\n [conv.append_message(\n conv.roles[0], ' {}'.format(text)) for conv, text in zip(convs, texts)]\n [conv.append_message(conv.roles[1], None) for conv in convs]\n texts = [conv.get_prompt() for conv in convs]\n return texts\n\n\ndef init_model(args):\n print('Initialization Model')\n cfg = Config(args)\n # cfg.model_cfg.ckpt = args.ckpt\n # cfg.model_cfg.lora_r = args.lora_r\n # cfg.model_cfg.lora_alpha = args.lora_alpha\n\n model_config = cfg.model_cfg\n model_cls = registry.get_model_class(model_config.arch)\n model = model_cls.from_config(model_config).to('cuda:0')\n\n# import pudb; pudb.set_trace()\n key = list(cfg.datasets_cfg.keys())[0]\n vis_processor_cfg = cfg.datasets_cfg.get(key).vis_processor.train\n vis_processor = registry.get_processor_class(vis_processor_cfg.name).from_config(vis_processor_cfg)\n print('Initialization Finished')\n return model, vis_processor\n\ndef computeIoU(bbox1, bbox2):\n x1, y1, x2, y2 = bbox1\n x3, y3, x4, y4 = bbox2\n intersection_x1 = max(x1, x3)\n intersection_y1 = max(y1, y3)\n intersection_x2 = min(x2, x4)\n intersection_y2 = min(y2, y4)\n intersection_area = max(0, intersection_x2 - intersection_x1 + 1) * max(0, intersection_y2 - intersection_y1 + 1)\n bbox1_area = (x2 - x1 + 1) * (y2 - y1 + 1)\n bbox2_area = (x4 - x3 + 1) * (y4 - y3 + 1)\n union_area = bbox1_area + bbox2_area - intersection_area\n iou = intersection_area / union_area\n return iou\n" }, { "path": "minigpt4/common/gradcam.py", "content": "import numpy as np\nfrom matplotlib import pyplot as plt\nfrom scipy.ndimage import filters\nfrom skimage import transform as skimage_transform\n\n\ndef getAttMap(img, attMap, blur=True, overlap=True):\n attMap -= attMap.min()\n if attMap.max() > 0:\n attMap /= attMap.max()\n attMap = skimage_transform.resize(attMap, (img.shape[:2]), order=3, mode=\"constant\")\n if blur:\n attMap = filters.gaussian_filter(attMap, 0.02 * max(img.shape[:2]))\n attMap -= attMap.min()\n attMap /= attMap.max()\n cmap = plt.get_cmap(\"jet\")\n attMapV = cmap(attMap)\n attMapV = np.delete(attMapV, 3, 2)\n if overlap:\n attMap = (\n 1 * (1 - attMap**0.7).reshape(attMap.shape + (1,)) * img\n + (attMap**0.7).reshape(attMap.shape + (1,)) * attMapV\n )\n return attMap\n" }, { "path": "minigpt4/common/logger.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport datetime\nimport logging\nimport time\nfrom collections import defaultdict, deque\n\nimport torch\nimport torch.distributed as dist\n\nfrom minigpt4.common import dist_utils\n\n\nclass SmoothedValue(object):\n \"\"\"Track a series of values and provide access to smoothed values over a\n window or the global series average.\n \"\"\"\n\n def __init__(self, window_size=20, fmt=None):\n if fmt is None:\n fmt = \"{median:.4f} ({global_avg:.4f})\"\n self.deque = deque(maxlen=window_size)\n self.total = 0.0\n self.count = 0\n self.fmt = fmt\n\n def update(self, value, n=1):\n self.deque.append(value)\n self.count += n\n self.total += value * n\n\n def synchronize_between_processes(self):\n \"\"\"\n Warning: does not synchronize the deque!\n \"\"\"\n if not dist_utils.is_dist_avail_and_initialized():\n return\n t = torch.tensor([self.count, self.total], dtype=torch.float64, device=\"cuda\")\n dist.barrier()\n dist.all_reduce(t)\n t = t.tolist()\n self.count = int(t[0])\n self.total = t[1]\n\n @property\n def median(self):\n d = torch.tensor(list(self.deque))\n return d.median().item()\n\n @property\n def avg(self):\n d = torch.tensor(list(self.deque), dtype=torch.float32)\n return d.mean().item()\n\n @property\n def global_avg(self):\n return self.total / self.count\n\n @property\n def max(self):\n return max(self.deque)\n\n @property\n def value(self):\n return self.deque[-1]\n\n def __str__(self):\n return self.fmt.format(\n median=self.median,\n avg=self.avg,\n global_avg=self.global_avg,\n max=self.max,\n value=self.value,\n )\n\n\nclass MetricLogger(object):\n def __init__(self, delimiter=\"\\t\"):\n self.meters = defaultdict(SmoothedValue)\n self.delimiter = delimiter\n\n def update(self, **kwargs):\n for k, v in kwargs.items():\n if isinstance(v, torch.Tensor):\n v = v.item()\n assert isinstance(v, (float, int))\n self.meters[k].update(v)\n\n def __getattr__(self, attr):\n if attr in self.meters:\n return self.meters[attr]\n if attr in self.__dict__:\n return self.__dict__[attr]\n raise AttributeError(\n \"'{}' object has no attribute '{}'\".format(type(self).__name__, attr)\n )\n\n def __str__(self):\n loss_str = []\n for name, meter in self.meters.items():\n loss_str.append(\"{}: {}\".format(name, str(meter)))\n return self.delimiter.join(loss_str)\n\n def global_avg(self):\n loss_str = []\n for name, meter in self.meters.items():\n loss_str.append(\"{}: {:.4f}\".format(name, meter.global_avg))\n return self.delimiter.join(loss_str)\n\n def synchronize_between_processes(self):\n for meter in self.meters.values():\n meter.synchronize_between_processes()\n\n def add_meter(self, name, meter):\n self.meters[name] = meter\n\n def log_every(self, iterable, print_freq, header=None):\n i = 0\n if not header:\n header = \"\"\n start_time = time.time()\n end = time.time()\n iter_time = SmoothedValue(fmt=\"{avg:.4f}\")\n data_time = SmoothedValue(fmt=\"{avg:.4f}\")\n space_fmt = \":\" + str(len(str(len(iterable)))) + \"d\"\n log_msg = [\n header,\n \"[{0\" + space_fmt + \"}/{1}]\",\n \"eta: {eta}\",\n \"{meters}\",\n \"time: {time}\",\n \"data: {data}\",\n ]\n if torch.cuda.is_available():\n log_msg.append(\"max mem: {memory:.0f}\")\n log_msg = self.delimiter.join(log_msg)\n MB = 1024.0 * 1024.0\n for obj in iterable:\n data_time.update(time.time() - end)\n yield obj\n iter_time.update(time.time() - end)\n if i % print_freq == 0 or i == len(iterable) - 1:\n eta_seconds = iter_time.global_avg * (len(iterable) - i)\n eta_string = str(datetime.timedelta(seconds=int(eta_seconds)))\n if torch.cuda.is_available():\n print(\n log_msg.format(\n i,\n len(iterable),\n eta=eta_string,\n meters=str(self),\n time=str(iter_time),\n data=str(data_time),\n memory=torch.cuda.max_memory_allocated() / MB,\n )\n )\n else:\n print(\n log_msg.format(\n i,\n len(iterable),\n eta=eta_string,\n meters=str(self),\n time=str(iter_time),\n data=str(data_time),\n )\n )\n i += 1\n end = time.time()\n total_time = time.time() - start_time\n total_time_str = str(datetime.timedelta(seconds=int(total_time)))\n print(\n \"{} Total time: {} ({:.4f} s / it)\".format(\n header, total_time_str, total_time / len(iterable)\n )\n )\n\n\nclass AttrDict(dict):\n def __init__(self, *args, **kwargs):\n super(AttrDict, self).__init__(*args, **kwargs)\n self.__dict__ = self\n\n\ndef setup_logger():\n logging.basicConfig(\n level=logging.INFO if dist_utils.is_main_process() else logging.WARN,\n format=\"%(asctime)s [%(levelname)s] %(message)s\",\n handlers=[logging.StreamHandler()],\n )\n" }, { "path": "minigpt4/common/optims.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport math\n\nfrom minigpt4.common.registry import registry\n\n\n@registry.register_lr_scheduler(\"linear_warmup_step_lr\")\nclass LinearWarmupStepLRScheduler:\n def __init__(\n self,\n optimizer,\n max_epoch,\n min_lr,\n init_lr,\n decay_rate=1,\n warmup_start_lr=-1,\n warmup_steps=0,\n **kwargs\n ):\n self.optimizer = optimizer\n\n self.max_epoch = max_epoch\n self.min_lr = min_lr\n\n self.decay_rate = decay_rate\n\n self.init_lr = init_lr\n self.warmup_steps = warmup_steps\n self.warmup_start_lr = warmup_start_lr if warmup_start_lr >= 0 else init_lr\n\n def step(self, cur_epoch, cur_step):\n if cur_epoch == 0:\n warmup_lr_schedule(\n step=cur_step,\n optimizer=self.optimizer,\n max_step=self.warmup_steps,\n init_lr=self.warmup_start_lr,\n max_lr=self.init_lr,\n )\n else:\n step_lr_schedule(\n epoch=cur_epoch,\n optimizer=self.optimizer,\n init_lr=self.init_lr,\n min_lr=self.min_lr,\n decay_rate=self.decay_rate,\n )\n\n\n@registry.register_lr_scheduler(\"linear_warmup_cosine_lr\")\nclass LinearWarmupCosineLRScheduler:\n def __init__(\n self,\n optimizer,\n max_epoch,\n iters_per_epoch,\n min_lr,\n init_lr,\n warmup_steps=0,\n warmup_start_lr=-1,\n **kwargs\n ):\n self.optimizer = optimizer\n\n self.max_epoch = max_epoch\n self.iters_per_epoch = iters_per_epoch\n self.min_lr = min_lr\n\n self.init_lr = init_lr\n self.warmup_steps = warmup_steps\n self.warmup_start_lr = warmup_start_lr if warmup_start_lr >= 0 else init_lr\n\n def step(self, cur_epoch, cur_step):\n total_cur_step = cur_epoch * self.iters_per_epoch + cur_step\n if total_cur_step < self.warmup_steps:\n warmup_lr_schedule(\n step=cur_step,\n optimizer=self.optimizer,\n max_step=self.warmup_steps,\n init_lr=self.warmup_start_lr,\n max_lr=self.init_lr,\n )\n else:\n cosine_lr_schedule(\n epoch=total_cur_step,\n optimizer=self.optimizer,\n max_epoch=self.max_epoch * self.iters_per_epoch,\n init_lr=self.init_lr,\n min_lr=self.min_lr,\n )\n\n\ndef cosine_lr_schedule(optimizer, epoch, max_epoch, init_lr, min_lr):\n \"\"\"Decay the learning rate\"\"\"\n lr = (init_lr - min_lr) * 0.5 * (\n 1.0 + math.cos(math.pi * epoch / max_epoch)\n ) + min_lr\n for param_group in optimizer.param_groups:\n param_group[\"lr\"] = lr\n\n\ndef warmup_lr_schedule(optimizer, step, max_step, init_lr, max_lr):\n \"\"\"Warmup the learning rate\"\"\"\n lr = min(max_lr, init_lr + (max_lr - init_lr) * step / max(max_step, 1))\n for param_group in optimizer.param_groups:\n param_group[\"lr\"] = lr\n\n\ndef step_lr_schedule(optimizer, epoch, init_lr, min_lr, decay_rate):\n \"\"\"Decay the learning rate\"\"\"\n lr = max(min_lr, init_lr * (decay_rate**epoch))\n for param_group in optimizer.param_groups:\n param_group[\"lr\"] = lr\n" }, { "path": "minigpt4/common/registry.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\n\nclass Registry:\n mapping = {\n \"builder_name_mapping\": {},\n \"task_name_mapping\": {},\n \"processor_name_mapping\": {},\n \"model_name_mapping\": {},\n \"lr_scheduler_name_mapping\": {},\n \"runner_name_mapping\": {},\n \"state\": {},\n \"paths\": {},\n }\n\n @classmethod\n def register_builder(cls, name):\n r\"\"\"Register a dataset builder to registry with key 'name'\n\n Args:\n name: Key with which the builder will be registered.\n\n Usage:\n\n from minigpt4.common.registry import registry\n from minigpt4.datasets.base_dataset_builder import BaseDatasetBuilder\n \"\"\"\n\n def wrap(builder_cls):\n from minigpt4.datasets.builders.base_dataset_builder import BaseDatasetBuilder\n\n assert issubclass(\n builder_cls, BaseDatasetBuilder\n ), \"All builders must inherit BaseDatasetBuilder class, found {}\".format(\n builder_cls\n )\n if name in cls.mapping[\"builder_name_mapping\"]:\n raise KeyError(\n \"Name '{}' already registered for {}.\".format(\n name, cls.mapping[\"builder_name_mapping\"][name]\n )\n )\n cls.mapping[\"builder_name_mapping\"][name] = builder_cls\n return builder_cls\n\n return wrap\n\n @classmethod\n def register_task(cls, name):\n r\"\"\"Register a task to registry with key 'name'\n\n Args:\n name: Key with which the task will be registered.\n\n Usage:\n\n from minigpt4.common.registry import registry\n \"\"\"\n\n def wrap(task_cls):\n from minigpt4.tasks.base_task import BaseTask\n\n assert issubclass(\n task_cls, BaseTask\n ), \"All tasks must inherit BaseTask class\"\n if name in cls.mapping[\"task_name_mapping\"]:\n raise KeyError(\n \"Name '{}' already registered for {}.\".format(\n name, cls.mapping[\"task_name_mapping\"][name]\n )\n )\n cls.mapping[\"task_name_mapping\"][name] = task_cls\n return task_cls\n\n return wrap\n\n @classmethod\n def register_model(cls, name):\n r\"\"\"Register a task to registry with key 'name'\n\n Args:\n name: Key with which the task will be registered.\n\n Usage:\n\n from minigpt4.common.registry import registry\n \"\"\"\n\n def wrap(model_cls):\n from minigpt4.models import BaseModel\n\n assert issubclass(\n model_cls, BaseModel\n ), \"All models must inherit BaseModel class\"\n if name in cls.mapping[\"model_name_mapping\"]:\n raise KeyError(\n \"Name '{}' already registered for {}.\".format(\n name, cls.mapping[\"model_name_mapping\"][name]\n )\n )\n cls.mapping[\"model_name_mapping\"][name] = model_cls\n return model_cls\n\n return wrap\n\n @classmethod\n def register_processor(cls, name):\n r\"\"\"Register a processor to registry with key 'name'\n\n Args:\n name: Key with which the task will be registered.\n\n Usage:\n\n from minigpt4.common.registry import registry\n \"\"\"\n\n def wrap(processor_cls):\n from minigpt4.processors import BaseProcessor\n\n assert issubclass(\n processor_cls, BaseProcessor\n ), \"All processors must inherit BaseProcessor class\"\n if name in cls.mapping[\"processor_name_mapping\"]:\n raise KeyError(\n \"Name '{}' already registered for {}.\".format(\n name, cls.mapping[\"processor_name_mapping\"][name]\n )\n )\n cls.mapping[\"processor_name_mapping\"][name] = processor_cls\n return processor_cls\n\n return wrap\n\n @classmethod\n def register_lr_scheduler(cls, name):\n r\"\"\"Register a model to registry with key 'name'\n\n Args:\n name: Key with which the task will be registered.\n\n Usage:\n\n from minigpt4.common.registry import registry\n \"\"\"\n\n def wrap(lr_sched_cls):\n if name in cls.mapping[\"lr_scheduler_name_mapping\"]:\n raise KeyError(\n \"Name '{}' already registered for {}.\".format(\n name, cls.mapping[\"lr_scheduler_name_mapping\"][name]\n )\n )\n cls.mapping[\"lr_scheduler_name_mapping\"][name] = lr_sched_cls\n return lr_sched_cls\n\n return wrap\n\n @classmethod\n def register_runner(cls, name):\n r\"\"\"Register a model to registry with key 'name'\n\n Args:\n name: Key with which the task will be registered.\n\n Usage:\n\n from minigpt4.common.registry import registry\n \"\"\"\n\n def wrap(runner_cls):\n if name in cls.mapping[\"runner_name_mapping\"]:\n raise KeyError(\n \"Name '{}' already registered for {}.\".format(\n name, cls.mapping[\"runner_name_mapping\"][name]\n )\n )\n cls.mapping[\"runner_name_mapping\"][name] = runner_cls\n return runner_cls\n\n return wrap\n\n @classmethod\n def register_path(cls, name, path):\n r\"\"\"Register a path to registry with key 'name'\n\n Args:\n name: Key with which the path will be registered.\n\n Usage:\n\n from minigpt4.common.registry import registry\n \"\"\"\n assert isinstance(path, str), \"All path must be str.\"\n if name in cls.mapping[\"paths\"]:\n raise KeyError(\"Name '{}' already registered.\".format(name))\n cls.mapping[\"paths\"][name] = path\n\n @classmethod\n def register(cls, name, obj):\n r\"\"\"Register an item to registry with key 'name'\n\n Args:\n name: Key with which the item will be registered.\n\n Usage::\n\n from minigpt4.common.registry import registry\n\n registry.register(\"config\", {})\n \"\"\"\n path = name.split(\".\")\n current = cls.mapping[\"state\"]\n\n for part in path[:-1]:\n if part not in current:\n current[part] = {}\n current = current[part]\n\n current[path[-1]] = obj\n\n # @classmethod\n # def get_trainer_class(cls, name):\n # return cls.mapping[\"trainer_name_mapping\"].get(name, None)\n\n @classmethod\n def get_builder_class(cls, name):\n return cls.mapping[\"builder_name_mapping\"].get(name, None)\n\n @classmethod\n def get_model_class(cls, name):\n return cls.mapping[\"model_name_mapping\"].get(name, None)\n\n @classmethod\n def get_task_class(cls, name):\n return cls.mapping[\"task_name_mapping\"].get(name, None)\n\n @classmethod\n def get_processor_class(cls, name):\n return cls.mapping[\"processor_name_mapping\"].get(name, None)\n\n @classmethod\n def get_lr_scheduler_class(cls, name):\n return cls.mapping[\"lr_scheduler_name_mapping\"].get(name, None)\n\n @classmethod\n def get_runner_class(cls, name):\n return cls.mapping[\"runner_name_mapping\"].get(name, None)\n\n @classmethod\n def list_runners(cls):\n return sorted(cls.mapping[\"runner_name_mapping\"].keys())\n\n @classmethod\n def list_models(cls):\n return sorted(cls.mapping[\"model_name_mapping\"].keys())\n\n @classmethod\n def list_tasks(cls):\n return sorted(cls.mapping[\"task_name_mapping\"].keys())\n\n @classmethod\n def list_processors(cls):\n return sorted(cls.mapping[\"processor_name_mapping\"].keys())\n\n @classmethod\n def list_lr_schedulers(cls):\n return sorted(cls.mapping[\"lr_scheduler_name_mapping\"].keys())\n\n @classmethod\n def list_datasets(cls):\n return sorted(cls.mapping[\"builder_name_mapping\"].keys())\n\n @classmethod\n def get_path(cls, name):\n return cls.mapping[\"paths\"].get(name, None)\n\n @classmethod\n def get(cls, name, default=None, no_warning=False):\n r\"\"\"Get an item from registry with key 'name'\n\n Args:\n name (string): Key whose value needs to be retrieved.\n default: If passed and key is not in registry, default value will\n be returned with a warning. Default: None\n no_warning (bool): If passed as True, warning when key doesn't exist\n will not be generated. Useful for MMF's\n internal operations. Default: False\n \"\"\"\n original_name = name\n name = name.split(\".\")\n value = cls.mapping[\"state\"]\n for subname in name:\n value = value.get(subname, default)\n if value is default:\n break\n\n if (\n \"writer\" in cls.mapping[\"state\"]\n and value == default\n and no_warning is False\n ):\n cls.mapping[\"state\"][\"writer\"].warning(\n \"Key {} is not present in registry, returning default value \"\n \"of {}\".format(original_name, default)\n )\n return value\n\n @classmethod\n def unregister(cls, name):\n r\"\"\"Remove an item from registry with key 'name'\n\n Args:\n name: Key which needs to be removed.\n Usage::\n\n from mmf.common.registry import registry\n\n config = registry.unregister(\"config\")\n \"\"\"\n return cls.mapping[\"state\"].pop(name, None)\n\n\nregistry = Registry()\n" }, { "path": "minigpt4/common/utils.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport io\nimport json\nimport logging\nimport os\nimport pickle\nimport re\nimport shutil\nimport urllib\nimport urllib.error\nimport urllib.request\nfrom typing import Optional\nfrom urllib.parse import urlparse\n\nimport numpy as np\nimport pandas as pd\nimport yaml\nfrom iopath.common.download import download\nfrom iopath.common.file_io import file_lock, g_pathmgr\nfrom minigpt4.common.registry import registry\nfrom torch.utils.model_zoo import tqdm\nfrom torchvision.datasets.utils import (\n check_integrity,\n download_file_from_google_drive,\n extract_archive,\n)\n\n\ndef now():\n from datetime import datetime\n\n return datetime.now().strftime(\"%Y%m%d%H%M\")[:-1]\n\n\ndef is_url(url_or_filename):\n parsed = urlparse(url_or_filename)\n return parsed.scheme in (\"http\", \"https\")\n\n\ndef get_cache_path(rel_path):\n return os.path.expanduser(os.path.join(registry.get_path(\"cache_root\"), rel_path))\n\n\ndef get_abs_path(rel_path):\n return os.path.join(registry.get_path(\"library_root\"), rel_path)\n\n\ndef load_json(filename):\n with open(filename, \"r\") as f:\n return json.load(f)\n\n\n# The following are adapted from torchvision and vissl\n# torchvision: https://github.com/pytorch/vision\n# vissl: https://github.com/facebookresearch/vissl/blob/main/vissl/utils/download.py\n\n\ndef makedir(dir_path):\n \"\"\"\n Create the directory if it does not exist.\n \"\"\"\n is_success = False\n try:\n if not g_pathmgr.exists(dir_path):\n g_pathmgr.mkdirs(dir_path)\n is_success = True\n except BaseException:\n print(f\"Error creating directory: {dir_path}\")\n return is_success\n\n\ndef get_redirected_url(url: str):\n \"\"\"\n Given a URL, returns the URL it redirects to or the\n original URL in case of no indirection\n \"\"\"\n import requests\n\n with requests.Session() as session:\n with session.get(url, stream=True, allow_redirects=True) as response:\n if response.history:\n return response.url\n else:\n return url\n\n\ndef to_google_drive_download_url(view_url: str) -> str:\n \"\"\"\n Utility function to transform a view URL of google drive\n to a download URL for google drive\n Example input:\n https://drive.google.com/file/d/137RyRjvTBkBiIfeYBNZBtViDHQ6_Ewsp/view\n Example output:\n https://drive.google.com/uc?export=download&id=137RyRjvTBkBiIfeYBNZBtViDHQ6_Ewsp\n \"\"\"\n splits = view_url.split(\"/\")\n assert splits[-1] == \"view\"\n file_id = splits[-2]\n return f\"https://drive.google.com/uc?export=download&id={file_id}\"\n\n\ndef download_google_drive_url(url: str, output_path: str, output_file_name: str):\n \"\"\"\n Download a file from google drive\n Downloading an URL from google drive requires confirmation when\n the file of the size is too big (google drive notifies that\n anti-viral checks cannot be performed on such files)\n \"\"\"\n import requests\n\n with requests.Session() as session:\n\n # First get the confirmation token and append it to the URL\n with session.get(url, stream=True, allow_redirects=True) as response:\n for k, v in response.cookies.items():\n if k.startswith(\"download_warning\"):\n url = url + \"&confirm=\" + v\n\n # Then download the content of the file\n with session.get(url, stream=True, verify=True) as response:\n makedir(output_path)\n path = os.path.join(output_path, output_file_name)\n total_size = int(response.headers.get(\"Content-length\", 0))\n with open(path, \"wb\") as file:\n from tqdm import tqdm\n\n with tqdm(total=total_size) as progress_bar:\n for block in response.iter_content(\n chunk_size=io.DEFAULT_BUFFER_SIZE\n ):\n file.write(block)\n progress_bar.update(len(block))\n\n\ndef _get_google_drive_file_id(url: str) -> Optional[str]:\n parts = urlparse(url)\n\n if re.match(r\"(drive|docs)[.]google[.]com\", parts.netloc) is None:\n return None\n\n match = re.match(r\"/file/d/(?P[^/]*)\", parts.path)\n if match is None:\n return None\n\n return match.group(\"id\")\n\n\ndef _urlretrieve(url: str, filename: str, chunk_size: int = 1024) -> None:\n with open(filename, \"wb\") as fh:\n with urllib.request.urlopen(\n urllib.request.Request(url, headers={\"User-Agent\": \"vissl\"})\n ) as response:\n with tqdm(total=response.length) as pbar:\n for chunk in iter(lambda: response.read(chunk_size), \"\"):\n if not chunk:\n break\n pbar.update(chunk_size)\n fh.write(chunk)\n\n\ndef download_url(\n url: str,\n root: str,\n filename: Optional[str] = None,\n md5: Optional[str] = None,\n) -> None:\n \"\"\"Download a file from a url and place it in root.\n Args:\n url (str): URL to download file from\n root (str): Directory to place downloaded file in\n filename (str, optional): Name to save the file under.\n If None, use the basename of the URL.\n md5 (str, optional): MD5 checksum of the download. If None, do not check\n \"\"\"\n root = os.path.expanduser(root)\n if not filename:\n filename = os.path.basename(url)\n fpath = os.path.join(root, filename)\n\n makedir(root)\n\n # check if file is already present locally\n if check_integrity(fpath, md5):\n print(\"Using downloaded and verified file: \" + fpath)\n return\n\n # expand redirect chain if needed\n url = get_redirected_url(url)\n\n # check if file is located on Google Drive\n file_id = _get_google_drive_file_id(url)\n if file_id is not None:\n return download_file_from_google_drive(file_id, root, filename, md5)\n\n # download the file\n try:\n print(\"Downloading \" + url + \" to \" + fpath)\n _urlretrieve(url, fpath)\n except (urllib.error.URLError, IOError) as e: # type: ignore[attr-defined]\n if url[:5] == \"https\":\n url = url.replace(\"https:\", \"http:\")\n print(\n \"Failed download. Trying https -> http instead.\"\n \" Downloading \" + url + \" to \" + fpath\n )\n _urlretrieve(url, fpath)\n else:\n raise e\n\n # check integrity of downloaded file\n if not check_integrity(fpath, md5):\n raise RuntimeError(\"File not found or corrupted.\")\n\n\ndef download_and_extract_archive(\n url: str,\n download_root: str,\n extract_root: Optional[str] = None,\n filename: Optional[str] = None,\n md5: Optional[str] = None,\n remove_finished: bool = False,\n) -> None:\n download_root = os.path.expanduser(download_root)\n if extract_root is None:\n extract_root = download_root\n if not filename:\n filename = os.path.basename(url)\n\n download_url(url, download_root, filename, md5)\n\n archive = os.path.join(download_root, filename)\n print(\"Extracting {} to {}\".format(archive, extract_root))\n extract_archive(archive, extract_root, remove_finished)\n\n\ndef cache_url(url: str, cache_dir: str) -> str:\n \"\"\"\n This implementation downloads the remote resource and caches it locally.\n The resource will only be downloaded if not previously requested.\n \"\"\"\n parsed_url = urlparse(url)\n dirname = os.path.join(cache_dir, os.path.dirname(parsed_url.path.lstrip(\"/\")))\n makedir(dirname)\n filename = url.split(\"/\")[-1]\n cached = os.path.join(dirname, filename)\n with file_lock(cached):\n if not os.path.isfile(cached):\n logging.info(f\"Downloading {url} to {cached} ...\")\n cached = download(url, dirname, filename=filename)\n logging.info(f\"URL {url} cached in {cached}\")\n return cached\n\n\n# TODO (prigoyal): convert this into RAII-style API\ndef create_file_symlink(file1, file2):\n \"\"\"\n Simply create the symlinks for a given file1 to file2.\n Useful during model checkpointing to symlinks to the\n latest successful checkpoint.\n \"\"\"\n try:\n if g_pathmgr.exists(file2):\n g_pathmgr.rm(file2)\n g_pathmgr.symlink(file1, file2)\n except Exception as e:\n logging.info(f\"Could NOT create symlink. Error: {e}\")\n\n\ndef save_file(data, filename, append_to_json=True, verbose=True):\n \"\"\"\n Common i/o utility to handle saving data to various file formats.\n Supported:\n .pkl, .pickle, .npy, .json\n Specifically for .json, users have the option to either append (default)\n or rewrite by passing in Boolean value to append_to_json.\n \"\"\"\n if verbose:\n logging.info(f\"Saving data to file: {filename}\")\n file_ext = os.path.splitext(filename)[1]\n if file_ext in [\".pkl\", \".pickle\"]:\n with g_pathmgr.open(filename, \"wb\") as fopen:\n pickle.dump(data, fopen, pickle.HIGHEST_PROTOCOL)\n elif file_ext == \".npy\":\n with g_pathmgr.open(filename, \"wb\") as fopen:\n np.save(fopen, data)\n elif file_ext == \".json\":\n if append_to_json:\n with g_pathmgr.open(filename, \"a\") as fopen:\n fopen.write(json.dumps(data, sort_keys=True) + \"\\n\")\n fopen.flush()\n else:\n with g_pathmgr.open(filename, \"w\") as fopen:\n fopen.write(json.dumps(data, sort_keys=True) + \"\\n\")\n fopen.flush()\n elif file_ext == \".yaml\":\n with g_pathmgr.open(filename, \"w\") as fopen:\n dump = yaml.dump(data)\n fopen.write(dump)\n fopen.flush()\n else:\n raise Exception(f\"Saving {file_ext} is not supported yet\")\n\n if verbose:\n logging.info(f\"Saved data to file: {filename}\")\n\n\ndef load_file(filename, mmap_mode=None, verbose=True, allow_pickle=False):\n \"\"\"\n Common i/o utility to handle loading data from various file formats.\n Supported:\n .pkl, .pickle, .npy, .json\n For the npy files, we support reading the files in mmap_mode.\n If the mmap_mode of reading is not successful, we load data without the\n mmap_mode.\n \"\"\"\n if verbose:\n logging.info(f\"Loading data from file: {filename}\")\n\n file_ext = os.path.splitext(filename)[1]\n if file_ext == \".txt\":\n with g_pathmgr.open(filename, \"r\") as fopen:\n data = fopen.readlines()\n elif file_ext in [\".pkl\", \".pickle\"]:\n with g_pathmgr.open(filename, \"rb\") as fopen:\n data = pickle.load(fopen, encoding=\"latin1\")\n elif file_ext == \".npy\":\n if mmap_mode:\n try:\n with g_pathmgr.open(filename, \"rb\") as fopen:\n data = np.load(\n fopen,\n allow_pickle=allow_pickle,\n encoding=\"latin1\",\n mmap_mode=mmap_mode,\n )\n except ValueError as e:\n logging.info(\n f\"Could not mmap {filename}: {e}. Trying without g_pathmgr\"\n )\n data = np.load(\n filename,\n allow_pickle=allow_pickle,\n encoding=\"latin1\",\n mmap_mode=mmap_mode,\n )\n logging.info(\"Successfully loaded without g_pathmgr\")\n except Exception:\n logging.info(\"Could not mmap without g_pathmgr. Trying without mmap\")\n with g_pathmgr.open(filename, \"rb\") as fopen:\n data = np.load(fopen, allow_pickle=allow_pickle, encoding=\"latin1\")\n else:\n with g_pathmgr.open(filename, \"rb\") as fopen:\n data = np.load(fopen, allow_pickle=allow_pickle, encoding=\"latin1\")\n elif file_ext == \".json\":\n with g_pathmgr.open(filename, \"r\") as fopen:\n data = json.load(fopen)\n elif file_ext == \".yaml\":\n with g_pathmgr.open(filename, \"r\") as fopen:\n data = yaml.load(fopen, Loader=yaml.FullLoader)\n elif file_ext == \".csv\":\n with g_pathmgr.open(filename, \"r\") as fopen:\n data = pd.read_csv(fopen)\n else:\n raise Exception(f\"Reading from {file_ext} is not supported yet\")\n return data\n\n\ndef abspath(resource_path: str):\n \"\"\"\n Make a path absolute, but take into account prefixes like\n \"http://\" or \"manifold://\"\n \"\"\"\n regex = re.compile(r\"^\\w+://\")\n if regex.match(resource_path) is None:\n return os.path.abspath(resource_path)\n else:\n return resource_path\n\n\ndef makedir(dir_path):\n \"\"\"\n Create the directory if it does not exist.\n \"\"\"\n is_success = False\n try:\n if not g_pathmgr.exists(dir_path):\n g_pathmgr.mkdirs(dir_path)\n is_success = True\n except BaseException:\n logging.info(f\"Error creating directory: {dir_path}\")\n return is_success\n\n\ndef is_url(input_url):\n \"\"\"\n Check if an input string is a url. look for http(s):// and ignoring the case\n \"\"\"\n is_url = re.match(r\"^(?:http)s?://\", input_url, re.IGNORECASE) is not None\n return is_url\n\n\ndef cleanup_dir(dir):\n \"\"\"\n Utility for deleting a directory. Useful for cleaning the storage space\n that contains various training artifacts like checkpoints, data etc.\n \"\"\"\n if os.path.exists(dir):\n logging.info(f\"Deleting directory: {dir}\")\n shutil.rmtree(dir)\n logging.info(f\"Deleted contents of directory: {dir}\")\n\n\ndef get_file_size(filename):\n \"\"\"\n Given a file, get the size of file in MB\n \"\"\"\n size_in_mb = os.path.getsize(filename) / float(1024**2)\n return size_in_mb\n" }, { "path": "minigpt4/common/vqa_tools/VQA/PythonEvaluationTools/vqaEvalDemo.py", "content": "# coding: utf-8\n\nimport sys\ndataDir = '../../VQA'\nsys.path.insert(0, '%s/PythonHelperTools/vqaTools' %(dataDir))\nfrom vqa import VQA\nfrom vqaEvaluation.vqaEval import VQAEval\nimport matplotlib.pyplot as plt\nimport skimage.io as io\nimport json\nimport random\nimport os\n\n# set up file names and paths\nversionType ='v2_' # this should be '' when using VQA v2.0 dataset\ntaskType ='OpenEnded' # 'OpenEnded' only for v2.0. 'OpenEnded' or 'MultipleChoice' for v1.0\ndataType ='mscoco' # 'mscoco' only for v1.0. 'mscoco' for real and 'abstract_v002' for abstract for v1.0. \ndataSubType ='train2014'\nannFile ='%s/Annotations/%s%s_%s_annotations.json'%(dataDir, versionType, dataType, dataSubType)\nquesFile ='%s/Questions/%s%s_%s_%s_questions.json'%(dataDir, versionType, taskType, dataType, dataSubType)\nimgDir ='%s/Images/%s/%s/' %(dataDir, dataType, dataSubType)\nresultType ='fake'\nfileTypes = ['results', 'accuracy', 'evalQA', 'evalQuesType', 'evalAnsType'] \n\n# An example result json file has been provided in './Results' folder. \n\n[resFile, accuracyFile, evalQAFile, evalQuesTypeFile, evalAnsTypeFile] = ['%s/Results/%s%s_%s_%s_%s_%s.json'%(dataDir, versionType, taskType, dataType, dataSubType, \\\nresultType, fileType) for fileType in fileTypes] \n\n# create vqa object and vqaRes object\nvqa = VQA(annFile, quesFile)\nvqaRes = vqa.loadRes(resFile, quesFile)\n\n# create vqaEval object by taking vqa and vqaRes\nvqaEval = VQAEval(vqa, vqaRes, n=2) #n is precision of accuracy (number of places after decimal), default is 2\n\n# evaluate results\n\"\"\"\nIf you have a list of question ids on which you would like to evaluate your results, pass it as a list to below function\nBy default it uses all the question ids in annotation file\n\"\"\"\nvqaEval.evaluate() \n\n# print accuracies\nprint \"\\n\"\nprint \"Overall Accuracy is: %.02f\\n\" %(vqaEval.accuracy['overall'])\nprint \"Per Question Type Accuracy is the following:\"\nfor quesType in vqaEval.accuracy['perQuestionType']:\n\tprint \"%s : %.02f\" %(quesType, vqaEval.accuracy['perQuestionType'][quesType])\nprint \"\\n\"\nprint \"Per Answer Type Accuracy is the following:\"\nfor ansType in vqaEval.accuracy['perAnswerType']:\n\tprint \"%s : %.02f\" %(ansType, vqaEval.accuracy['perAnswerType'][ansType])\nprint \"\\n\"\n# demo how to use evalQA to retrieve low score result\nevals = [quesId for quesId in vqaEval.evalQA if vqaEval.evalQA[quesId]<35] #35 is per question percentage accuracy\nif len(evals) > 0:\n\tprint 'ground truth answers'\n\trandomEval = random.choice(evals)\n\trandomAnn = vqa.loadQA(randomEval)\n\tvqa.showQA(randomAnn)\n\n\tprint '\\n'\n\tprint 'generated answer (accuracy %.02f)'%(vqaEval.evalQA[randomEval])\n\tann = vqaRes.loadQA(randomEval)[0]\n\tprint \"Answer: %s\\n\" %(ann['answer'])\n\n\timgId = randomAnn[0]['image_id']\n\timgFilename = 'COCO_' + dataSubType + '_'+ str(imgId).zfill(12) + '.jpg'\n\tif os.path.isfile(imgDir + imgFilename):\n\t\tI = io.imread(imgDir + imgFilename)\n\t\tplt.imshow(I)\n\t\tplt.axis('off')\n\t\tplt.show()\n\n# plot accuracy for various question types\nplt.bar(range(len(vqaEval.accuracy['perQuestionType'])), vqaEval.accuracy['perQuestionType'].values(), align='center')\nplt.xticks(range(len(vqaEval.accuracy['perQuestionType'])), vqaEval.accuracy['perQuestionType'].keys(), rotation='0',fontsize=10)\nplt.title('Per Question Type Accuracy', fontsize=10)\nplt.xlabel('Question Types', fontsize=10)\nplt.ylabel('Accuracy', fontsize=10)\nplt.show()\n\n# save evaluation results to ./Results folder\njson.dump(vqaEval.accuracy, open(accuracyFile, 'w'))\njson.dump(vqaEval.evalQA, open(evalQAFile, 'w'))\njson.dump(vqaEval.evalQuesType, open(evalQuesTypeFile, 'w'))\njson.dump(vqaEval.evalAnsType, open(evalAnsTypeFile, 'w'))\n\n" }, { "path": "minigpt4/common/vqa_tools/VQA/PythonEvaluationTools/vqaEvaluation/__init__.py", "content": "author='aagrawal'\n" }, { "path": "minigpt4/common/vqa_tools/VQA/PythonEvaluationTools/vqaEvaluation/vqaEval.py", "content": "# coding=utf-8\n\n__author__='aagrawal'\n\nimport re\n# This code is based on the code written by Tsung-Yi Lin for MSCOCO Python API available at the following link:\n# (https://github.com/tylin/coco-caption/blob/master/pycocoevalcap/eval.py).\nimport sys\n\n\nclass VQAEval:\n\tdef __init__(self, vqa, vqaRes, n=2):\n\t\tself.n \t\t\t = n\n\t\tself.accuracy = {}\n\t\tself.evalQA = {}\n\t\tself.evalQuesType = {}\n\t\tself.evalAnsType = {}\n\t\tself.vqa \t\t = vqa\n\t\tself.vqaRes = vqaRes\n\t\tself.params\t\t = {'question_id': vqa.getQuesIds()}\n\t\tself.contractions = {\"aint\": \"ain't\", \"arent\": \"aren't\", \"cant\": \"can't\", \"couldve\": \"could've\", \"couldnt\": \"couldn't\", \\\n\t\t\t\t\t\t\t \"couldn'tve\": \"couldn't've\", \"couldnt've\": \"couldn't've\", \"didnt\": \"didn't\", \"doesnt\": \"doesn't\", \"dont\": \"don't\", \"hadnt\": \"hadn't\", \\\n\t\t\t\t\t\t\t \"hadnt've\": \"hadn't've\", \"hadn'tve\": \"hadn't've\", \"hasnt\": \"hasn't\", \"havent\": \"haven't\", \"hed\": \"he'd\", \"hed've\": \"he'd've\", \\\n\t\t\t\t\t\t\t \"he'dve\": \"he'd've\", \"hes\": \"he's\", \"howd\": \"how'd\", \"howll\": \"how'll\", \"hows\": \"how's\", \"Id've\": \"I'd've\", \"I'dve\": \"I'd've\", \\\n\t\t\t\t\t\t\t \"Im\": \"I'm\", \"Ive\": \"I've\", \"isnt\": \"isn't\", \"itd\": \"it'd\", \"itd've\": \"it'd've\", \"it'dve\": \"it'd've\", \"itll\": \"it'll\", \"let's\": \"let's\", \\\n\t\t\t\t\t\t\t \"maam\": \"ma'am\", \"mightnt\": \"mightn't\", \"mightnt've\": \"mightn't've\", \"mightn'tve\": \"mightn't've\", \"mightve\": \"might've\", \\\n\t\t\t\t\t\t\t \"mustnt\": \"mustn't\", \"mustve\": \"must've\", \"neednt\": \"needn't\", \"notve\": \"not've\", \"oclock\": \"o'clock\", \"oughtnt\": \"oughtn't\", \\\n\t\t\t\t\t\t\t \"ow's'at\": \"'ow's'at\", \"'ows'at\": \"'ow's'at\", \"'ow'sat\": \"'ow's'at\", \"shant\": \"shan't\", \"shed've\": \"she'd've\", \"she'dve\": \"she'd've\", \\\n\t\t\t\t\t\t\t \"she's\": \"she's\", \"shouldve\": \"should've\", \"shouldnt\": \"shouldn't\", \"shouldnt've\": \"shouldn't've\", \"shouldn'tve\": \"shouldn't've\", \\\n\t\t\t\t\t\t\t \"somebody'd\": \"somebodyd\", \"somebodyd've\": \"somebody'd've\", \"somebody'dve\": \"somebody'd've\", \"somebodyll\": \"somebody'll\", \\\n\t\t\t\t\t\t\t \"somebodys\": \"somebody's\", \"someoned\": \"someone'd\", \"someoned've\": \"someone'd've\", \"someone'dve\": \"someone'd've\", \\\n\t\t\t\t\t\t\t \"someonell\": \"someone'll\", \"someones\": \"someone's\", \"somethingd\": \"something'd\", \"somethingd've\": \"something'd've\", \\\n\t\t\t\t\t\t\t \"something'dve\": \"something'd've\", \"somethingll\": \"something'll\", \"thats\": \"that's\", \"thered\": \"there'd\", \"thered've\": \"there'd've\", \\\n\t\t\t\t\t\t\t \"there'dve\": \"there'd've\", \"therere\": \"there're\", \"theres\": \"there's\", \"theyd\": \"they'd\", \"theyd've\": \"they'd've\", \\\n\t\t\t\t\t\t\t \"they'dve\": \"they'd've\", \"theyll\": \"they'll\", \"theyre\": \"they're\", \"theyve\": \"they've\", \"twas\": \"'twas\", \"wasnt\": \"wasn't\", \\\n\t\t\t\t\t\t\t \"wed've\": \"we'd've\", \"we'dve\": \"we'd've\", \"weve\": \"we've\", \"werent\": \"weren't\", \"whatll\": \"what'll\", \"whatre\": \"what're\", \\\n\t\t\t\t\t\t\t \"whats\": \"what's\", \"whatve\": \"what've\", \"whens\": \"when's\", \"whered\": \"where'd\", \"wheres\": \"where's\", \"whereve\": \"where've\", \\\n\t\t\t\t\t\t\t \"whod\": \"who'd\", \"whod've\": \"who'd've\", \"who'dve\": \"who'd've\", \"wholl\": \"who'll\", \"whos\": \"who's\", \"whove\": \"who've\", \"whyll\": \"why'll\", \\\n\t\t\t\t\t\t\t \"whyre\": \"why're\", \"whys\": \"why's\", \"wont\": \"won't\", \"wouldve\": \"would've\", \"wouldnt\": \"wouldn't\", \"wouldnt've\": \"wouldn't've\", \\\n\t\t\t\t\t\t\t \"wouldn'tve\": \"wouldn't've\", \"yall\": \"y'all\", \"yall'll\": \"y'all'll\", \"y'allll\": \"y'all'll\", \"yall'd've\": \"y'all'd've\", \\\n\t\t\t\t\t\t\t \"y'alld've\": \"y'all'd've\", \"y'all'dve\": \"y'all'd've\", \"youd\": \"you'd\", \"youd've\": \"you'd've\", \"you'dve\": \"you'd've\", \\\n\t\t\t\t\t\t\t \"youll\": \"you'll\", \"youre\": \"you're\", \"youve\": \"you've\"}\n\t\tself.manualMap = { 'none': '0',\n\t\t\t\t\t\t\t 'zero': '0',\n\t\t\t\t\t\t\t 'one': '1',\n\t\t\t\t\t\t\t 'two': '2',\n\t\t\t\t\t\t\t 'three': '3',\n\t\t\t\t\t\t\t 'four': '4',\n\t\t\t\t\t\t\t 'five': '5',\n\t\t\t\t\t\t\t 'six': '6',\n\t\t\t\t\t\t\t 'seven': '7',\n\t\t\t\t\t\t\t 'eight': '8',\n\t\t\t\t\t\t\t 'nine': '9',\n\t\t\t\t\t\t\t 'ten': '10'\n\t\t\t\t\t\t\t}\n\t\tself.articles = ['a',\n\t\t\t\t\t\t\t 'an',\n\t\t\t\t\t\t\t 'the'\n\t\t\t\t\t\t\t]\n\n\n\t\tself.periodStrip = re.compile(\"(?!<=\\d)(\\.)(?!\\d)\")\n\t\tself.commaStrip = re.compile(\"(\\d)(\\,)(\\d)\")\n\t\tself.punct = [';', r\"/\", '[', ']', '\"', '{', '}',\n\t\t\t\t\t\t\t '(', ')', '=', '+', '\\\\', '_', '-',\n\t\t\t\t\t\t\t '>', '<', '@', '`', ',', '?', '!']\n\n\n\tdef evaluate(self, quesIds=None):\n\t\tif quesIds == None:\n\t\t\tquesIds = [quesId for quesId in self.params['question_id']]\n\t\tgts = {}\n\t\tres = {}\n\t\tfor quesId in quesIds:\n\t\t\tgts[quesId] = self.vqa.qa[quesId]\n\t\t\tres[quesId] = self.vqaRes.qa[quesId]\n\n\t\t# =================================================\n\t\t# Compute accuracy\n\t\t# =================================================\n\t\taccQA = []\n\t\taccQuesType = {}\n\t\taccAnsType = {}\n\t\t# print \"computing accuracy\"\n\t\tstep = 0\n\t\tfor quesId in quesIds:\n\t\t\tfor ansDic in gts[quesId]['answers']:\n\t\t\t\tansDic['answer'] = ansDic['answer'].replace('\\n', ' ')\n\t\t\t\tansDic['answer'] = ansDic['answer'].replace('\\t', ' ')\n\t\t\t\tansDic['answer'] = ansDic['answer'].strip()\n\t\t\tresAns = res[quesId]['answer']\n\t\t\tresAns = resAns.replace('\\n', ' ')\n\t\t\tresAns = resAns.replace('\\t', ' ')\n\t\t\tresAns = resAns.strip()\n\t\t\tgtAcc = []\n\t\t\tgtAnswers = [ans['answer'] for ans in gts[quesId]['answers']]\n\n\t\t\tif len(set(gtAnswers)) > 1:\n\t\t\t\tfor ansDic in gts[quesId]['answers']:\n\t\t\t\t\tansDic['answer'] = self.processPunctuation(ansDic['answer'])\n\t\t\t\t\tansDic['answer'] = self.processDigitArticle(ansDic['answer'])\n\t\t\t\tresAns = self.processPunctuation(resAns)\n\t\t\t\tresAns = self.processDigitArticle(resAns)\n\n\t\t\tfor gtAnsDatum in gts[quesId]['answers']:\n\t\t\t\totherGTAns = [item for item in gts[quesId]['answers'] if item!=gtAnsDatum]\n\t\t\t\tmatchingAns = [item for item in otherGTAns if item['answer'].lower()==resAns.lower()]\n\t\t\t\tacc = min(1, float(len(matchingAns))/3)\n\t\t\t\tgtAcc.append(acc)\n\t\t\tquesType = gts[quesId]['question_type']\n\t\t\tansType = gts[quesId]['answer_type']\n\t\t\tavgGTAcc = float(sum(gtAcc))/len(gtAcc)\n\t\t\taccQA.append(avgGTAcc)\n\t\t\tif quesType not in accQuesType:\n\t\t\t\taccQuesType[quesType] = []\n\t\t\taccQuesType[quesType].append(avgGTAcc)\n\t\t\tif ansType not in accAnsType:\n\t\t\t\taccAnsType[ansType] = []\n\t\t\taccAnsType[ansType].append(avgGTAcc)\n\t\t\tself.setEvalQA(quesId, avgGTAcc)\n\t\t\tself.setEvalQuesType(quesId, quesType, avgGTAcc)\n\t\t\tself.setEvalAnsType(quesId, ansType, avgGTAcc)\n\t\t\tif step%100 == 0:\n\t\t\t\tself.updateProgress(step/float(len(quesIds)))\n\t\t\tstep = step + 1\n\n\t\tself.setAccuracy(accQA, accQuesType, accAnsType)\n\t\t# print \"Done computing accuracy\"\n\n\tdef processPunctuation(self, inText):\n\t\toutText = inText\n\t\tfor p in self.punct:\n\t\t\tif (p + ' ' in inText or ' ' + p in inText) or (re.search(self.commaStrip, inText) != None):\n\t\t\t\toutText = outText.replace(p, '')\n\t\t\telse:\n\t\t\t\toutText = outText.replace(p, ' ')\n\t\toutText = self.periodStrip.sub(\"\",\n\t\t\t\t\t\t\t\t\t outText,\n\t\t\t\t\t\t\t\t\t re.UNICODE)\n\t\treturn outText\n\n\tdef processDigitArticle(self, inText):\n\t\toutText = []\n\t\ttempText = inText.lower().split()\n\t\tfor word in tempText:\n\t\t\tword = self.manualMap.setdefault(word, word)\n\t\t\tif word not in self.articles:\n\t\t\t\toutText.append(word)\n\t\t\telse:\n\t\t\t\tpass\n\t\tfor wordId, word in enumerate(outText):\n\t\t\tif word in self.contractions:\n\t\t\t\toutText[wordId] = self.contractions[word]\n\t\toutText = ' '.join(outText)\n\t\treturn outText\n\n\tdef setAccuracy(self, accQA, accQuesType, accAnsType):\n\t\tself.accuracy['overall'] = round(100*float(sum(accQA))/len(accQA), self.n)\n\t\tself.accuracy['perQuestionType'] = {quesType: round(100*float(sum(accQuesType[quesType]))/len(accQuesType[quesType]), self.n) for quesType in accQuesType}\n\t\tself.accuracy['perAnswerType'] = {ansType: round(100*float(sum(accAnsType[ansType]))/len(accAnsType[ansType]), self.n) for ansType in accAnsType}\n\n\tdef setEvalQA(self, quesId, acc):\n\t\tself.evalQA[quesId] = round(100*acc, self.n)\n\n\tdef setEvalQuesType(self, quesId, quesType, acc):\n\t\tif quesType not in self.evalQuesType:\n\t\t\tself.evalQuesType[quesType] = {}\n\t\tself.evalQuesType[quesType][quesId] = round(100*acc, self.n)\n\n\tdef setEvalAnsType(self, quesId, ansType, acc):\n\t\tif ansType not in self.evalAnsType:\n\t\t\tself.evalAnsType[ansType] = {}\n\t\tself.evalAnsType[ansType][quesId] = round(100*acc, self.n)\n\n\tdef updateProgress(self, progress):\n\t\tbarLength = 20\n\t\tstatus = \"\"\n\t\tif isinstance(progress, int):\n\t\t\tprogress = float(progress)\n\t\tif not isinstance(progress, float):\n\t\t\tprogress = 0\n\t\t\tstatus = \"error: progress var must be float\\r\\n\"\n\t\tif progress < 0:\n\t\t\tprogress = 0\n\t\t\tstatus = \"Halt...\\r\\n\"\n\t\tif progress >= 1:\n\t\t\tprogress = 1\n\t\t\tstatus = \"Done...\\r\\n\"\n\t\tblock = int(round(barLength*progress))\n\t\ttext = \"\\rFinshed Percent: [{0}] {1}% {2}\".format( \"#\"*block + \"-\"*(barLength-block), int(progress*100), status)\n\t\tsys.stdout.write(text)\n\t\tsys.stdout.flush()\n" }, { "path": "minigpt4/common/vqa_tools/VQA/PythonHelperTools/vqaDemo.py", "content": "# coding: utf-8\n\nfrom vqaTools.vqa import VQA\nimport random\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nimport os\n\ndataDir\t\t='../../VQA'\nversionType ='v2_' # this should be '' when using VQA v2.0 dataset\ntaskType ='OpenEnded' # 'OpenEnded' only for v2.0. 'OpenEnded' or 'MultipleChoice' for v1.0\ndataType ='mscoco' # 'mscoco' only for v1.0. 'mscoco' for real and 'abstract_v002' for abstract for v1.0.\ndataSubType ='train2014'\nannFile ='%s/Annotations/%s%s_%s_annotations.json'%(dataDir, versionType, dataType, dataSubType)\nquesFile ='%s/Questions/%s%s_%s_%s_questions.json'%(dataDir, versionType, taskType, dataType, dataSubType)\nimgDir \t\t= '%s/Images/%s/%s/' %(dataDir, dataType, dataSubType)\n\n# initialize VQA api for QA annotations\nvqa=VQA(annFile, quesFile)\n\n# load and display QA annotations for given question types\n\"\"\"\nAll possible quesTypes for abstract and mscoco has been provided in respective text files in ../QuestionTypes/ folder.\n\"\"\"\nannIds = vqa.getQuesIds(quesTypes='how many'); \nanns = vqa.loadQA(annIds)\nrandomAnn = random.choice(anns)\nvqa.showQA([randomAnn])\nimgId = randomAnn['image_id']\nimgFilename = 'COCO_' + dataSubType + '_'+ str(imgId).zfill(12) + '.jpg'\nif os.path.isfile(imgDir + imgFilename):\n\tI = io.imread(imgDir + imgFilename)\n\tplt.imshow(I)\n\tplt.axis('off')\n\tplt.show()\n\n# load and display QA annotations for given answer types\n\"\"\"\nansTypes can be one of the following\nyes/no\nnumber\nother\n\"\"\"\nannIds = vqa.getQuesIds(ansTypes='yes/no'); \nanns = vqa.loadQA(annIds)\nrandomAnn = random.choice(anns)\nvqa.showQA([randomAnn])\nimgId = randomAnn['image_id']\nimgFilename = 'COCO_' + dataSubType + '_'+ str(imgId).zfill(12) + '.jpg'\nif os.path.isfile(imgDir + imgFilename):\n\tI = io.imread(imgDir + imgFilename)\n\tplt.imshow(I)\n\tplt.axis('off')\n\tplt.show()\n\n# load and display QA annotations for given images\n\"\"\"\nUsage: vqa.getImgIds(quesIds=[], quesTypes=[], ansTypes=[])\nAbove method can be used to retrieve imageIds for given question Ids or given question types or given answer types.\n\"\"\"\nids = vqa.getImgIds()\nannIds = vqa.getQuesIds(imgIds=random.sample(ids,5)); \nanns = vqa.loadQA(annIds)\nrandomAnn = random.choice(anns)\nvqa.showQA([randomAnn]) \nimgId = randomAnn['image_id']\nimgFilename = 'COCO_' + dataSubType + '_'+ str(imgId).zfill(12) + '.jpg'\nif os.path.isfile(imgDir + imgFilename):\n\tI = io.imread(imgDir + imgFilename)\n\tplt.imshow(I)\n\tplt.axis('off')\n\tplt.show()\n\n" }, { "path": "minigpt4/common/vqa_tools/VQA/PythonHelperTools/vqaTools/__init__.py", "content": "__author__ = 'aagrawal'\n" }, { "path": "minigpt4/common/vqa_tools/VQA/PythonHelperTools/vqaTools/vqa.py", "content": "__author__ = 'aagrawal'\n__version__ = '0.9'\n\n# Interface for accessing the VQA dataset.\n\n# This code is based on the code written by Tsung-Yi Lin for MSCOCO Python API available at the following link: \n# (https://github.com/pdollar/coco/blob/master/PythonAPI/pycocotools/coco.py).\n\n# The following functions are defined:\n# VQA - VQA class that loads VQA annotation file and prepares data structures.\n# getQuesIds - Get question ids that satisfy given filter conditions.\n# getImgIds - Get image ids that satisfy given filter conditions.\n# loadQA - Load questions and answers with the specified question ids.\n# showQA - Display the specified questions and answers.\n# loadRes - Load result file and create result object.\n\n# Help on each function can be accessed by: \"help(COCO.function)\"\n\nimport json\nimport datetime\nimport copy\n\n\nclass VQA:\n def __init__(self, annotation_file=None, question_file=None):\n \"\"\"\n Constructor of VQA helper class for reading and visualizing questions and answers.\n :param annotation_file (str): location of VQA annotation file\n :return:\n \"\"\"\n # load dataset\n self.dataset = {}\n self.questions = {}\n self.qa = {}\n self.qqa = {}\n self.imgToQA = {}\n if not annotation_file == None and not question_file == None:\n # print 'loading VQA annotations and questions into memory...'\n time_t = datetime.datetime.utcnow()\n dataset = json.load(open(annotation_file, 'r'))\n questions = json.load(open(question_file, 'r'))\n # print datetime.datetime.utcnow() - time_t\n self.dataset = dataset\n self.questions = questions\n self.createIndex()\n\n def createIndex(self):\n imgToQA = {ann['image_id']: [] for ann in self.dataset['annotations']}\n qa = {ann['question_id']: [] for ann in self.dataset['annotations']}\n qqa = {ann['question_id']: [] for ann in self.dataset['annotations']}\n for ann in self.dataset['annotations']:\n imgToQA[ann['image_id']] += [ann]\n qa[ann['question_id']] = ann\n for ques in self.questions['questions']:\n qqa[ques['question_id']] = ques\n # print 'index created!'\n\n # create class members\n self.qa = qa\n self.qqa = qqa\n self.imgToQA = imgToQA\n\n def info(self):\n \"\"\"\n Print information about the VQA annotation file.\n :return:\n \"\"\"\n\n # for key, value in self.datset['info'].items():\n # \tprint '%s: %s'%(key, value)\n\n def getQuesIds(self, imgIds=[], quesTypes=[], ansTypes=[]):\n \"\"\"\n Get question ids that satisfy given filter conditions. default skips that filter\n :param \timgIds (int array) : get question ids for given imgs\n quesTypes (str array) : get question ids for given question types\n ansTypes (str array) : get question ids for given answer types\n :return: ids (int array) : integer array of question ids\n \"\"\"\n imgIds = imgIds if type(imgIds) == list else [imgIds]\n quesTypes = quesTypes if type(quesTypes) == list else [quesTypes]\n ansTypes = ansTypes if type(ansTypes) == list else [ansTypes]\n\n if len(imgIds) == len(quesTypes) == len(ansTypes) == 0:\n anns = self.dataset['annotations']\n else:\n if not len(imgIds) == 0:\n anns = sum([self.imgToQA[imgId] for imgId in imgIds if imgId in self.imgToQA], [])\n else:\n anns = self.dataset['annotations']\n anns = anns if len(quesTypes) == 0 else [ann for ann in anns if ann['question_type'] in quesTypes]\n anns = anns if len(ansTypes) == 0 else [ann for ann in anns if ann['answer_type'] in ansTypes]\n ids = [ann['question_id'] for ann in anns]\n return ids\n\n def getImgIds(self, quesIds=[], quesTypes=[], ansTypes=[]):\n \"\"\"\n Get image ids that satisfy given filter conditions. default skips that filter\n :param quesIds (int array) : get image ids for given question ids\n quesTypes (str array) : get image ids for given question types\n ansTypes (str array) : get image ids for given answer types\n :return: ids (int array) : integer array of image ids\n \"\"\"\n quesIds = quesIds if type(quesIds) == list else [quesIds]\n quesTypes = quesTypes if type(quesTypes) == list else [quesTypes]\n ansTypes = ansTypes if type(ansTypes) == list else [ansTypes]\n\n if len(quesIds) == len(quesTypes) == len(ansTypes) == 0:\n anns = self.dataset['annotations']\n else:\n if not len(quesIds) == 0:\n anns = sum([self.qa[quesId] for quesId in quesIds if quesId in self.qa], [])\n else:\n anns = self.dataset['annotations']\n anns = anns if len(quesTypes) == 0 else [ann for ann in anns if ann['question_type'] in quesTypes]\n anns = anns if len(ansTypes) == 0 else [ann for ann in anns if ann['answer_type'] in ansTypes]\n ids = [ann['image_id'] for ann in anns]\n return ids\n\n def loadQA(self, ids=[]):\n \"\"\"\n Load questions and answers with the specified question ids.\n :param ids (int array) : integer ids specifying question ids\n :return: qa (object array) : loaded qa objects\n \"\"\"\n if type(ids) == list:\n return [self.qa[id] for id in ids]\n elif type(ids) == int:\n return [self.qa[ids]]\n\n def showQA(self, anns):\n \"\"\"\n Display the specified annotations.\n :param anns (array of object): annotations to display\n :return: None\n \"\"\"\n if len(anns) == 0:\n return 0\n for ann in anns:\n quesId = ann['question_id']\n print(\"Question: %s\" % (self.qqa[quesId]['question']))\n for ans in ann['answers']:\n print(\"Answer %d: %s\" % (ans['answer_id'], ans['answer']))\n\n def loadRes(self, resFile, quesFile):\n \"\"\"\n Load result file and return a result object.\n :param resFile (str) : file name of result file\n :return: res (obj) : result api object\n \"\"\"\n res = VQA()\n res.questions = json.load(open(quesFile))\n res.dataset['info'] = copy.deepcopy(self.questions['info'])\n res.dataset['task_type'] = copy.deepcopy(self.questions['task_type'])\n res.dataset['data_type'] = copy.deepcopy(self.questions['data_type'])\n res.dataset['data_subtype'] = copy.deepcopy(self.questions['data_subtype'])\n res.dataset['license'] = copy.deepcopy(self.questions['license'])\n\n # print 'Loading and preparing results... '\n time_t = datetime.datetime.utcnow()\n anns = json.load(open(resFile))\n assert type(anns) == list, 'results is not an array of objects'\n annsQuesIds = [ann['question_id'] for ann in anns]\n assert set(annsQuesIds) == set(self.getQuesIds()), \\\n 'Results do not correspond to current VQA set. Either the results do not have predictions for all question ids in annotation file or there is atleast one question id that does not belong to the question ids in the annotation file.'\n for ann in anns:\n quesId = ann['question_id']\n if res.dataset['task_type'] == 'Multiple Choice':\n assert ann['answer'] in self.qqa[quesId][\n 'multiple_choices'], 'predicted answer is not one of the multiple choices'\n qaAnn = self.qa[quesId]\n ann['image_id'] = qaAnn['image_id']\n ann['question_type'] = qaAnn['question_type']\n ann['answer_type'] = qaAnn['answer_type']\n # print 'DONE (t=%0.2fs)'%((datetime.datetime.utcnow() - time_t).total_seconds())\n\n res.dataset['annotations'] = anns\n res.createIndex()\n return res\n" }, { "path": "minigpt4/common/vqa_tools/VQA/QuestionTypes/abstract_v002_question_types.txt", "content": "how many\nwhat color is the\nis the\nwhere is the\nwhat\nwhat is\nare the\nwhat is the\nis there a\ndoes the\nis the woman\nis the man\nwhat is on the\nis it\nis the girl\nis the boy\nis the dog\nare they\nwho is\nwhat kind of\nwhat color are the\nwhat is in the\nwhat is the man\nis there\nwhat is the woman\nwhat are the\nwhat is the boy\nare there\nwhat is the girl\nis this\nhow\nwhich\nhow many people are\nis the cat\nwhy is the\nare\nwill the\nwhat type of\nwhat is the dog\ndo\nis she\ndoes\ndo the\nis\nis the baby\nare there any\nis the lady\ncan\nwhat animal is\nwhere are the\nis the sun\nwhat are they\ndid the\nwhat is the cat\nwhat is the lady\nhow many clouds are\nis that\nis the little girl\nis he\nare these\nhow many trees are\nhow many pillows\nare the people\nwhy\nis the young\nhow many windows are\nis this a\nwhat is the little\nis the tv\nhow many animals are\nwho\nhow many pictures\nhow many plants are\nhow many birds are\nwhat color is\nwhat is the baby\nis anyone\nwhat color\nhow many bushes\nis the old man\nnone of the above\n" }, { "path": "minigpt4/common/vqa_tools/VQA/QuestionTypes/mscoco_question_types.txt", "content": "how many\nis the\nwhat\nwhat color is the\nwhat is the\nis this\nis this a\nwhat is\nare the\nwhat kind of\nis there a\nwhat type of\nis it\nwhat are the\nwhere is the\nis there\ndoes the\nwhat color are the\nare these\nare there\nwhich\nis\nwhat is the man\nis the man\nare\nhow\ndoes this\nwhat is on the\nwhat does the\nhow many people are\nwhat is in the\nwhat is this\ndo\nwhat are\nare they\nwhat time\nwhat sport is\nare there any\nis he\nwhat color is\nwhy\nwhere are the\nwhat color\nwho is\nwhat animal is\nis the woman\nis this an\ndo you\nhow many people are in\nwhat room is\nhas\nis this person\nwhat is the woman\ncan you\nwhy is the\nis the person\nwhat is the color of the\nwhat is the person\ncould\nwas\nis that a\nwhat number is\nwhat is the name\nwhat brand\nnone of the above\n" }, { "path": "minigpt4/common/vqa_tools/VQA/README.md", "content": "Python API and Evaluation Code for v2.0 and v1.0 releases of the VQA dataset.\n===================\n## VQA v2.0 release ##\nThis release consists of\n- Real \n\t- 82,783 MS COCO training images, 40,504 MS COCO validation images and 81,434 MS COCO testing images (images are obtained from [MS COCO website] (http://mscoco.org/dataset/#download))\n\t- 443,757 questions for training, 214,354 questions for validation and 447,793 questions for testing\n\t- 4,437,570 answers for training and 2,143,540 answers for validation (10 per question)\n\nThere is only one type of task\n- Open-ended task\n\n## VQA v1.0 release ##\nThis release consists of\n- Real \n\t- 82,783 MS COCO training images, 40,504 MS COCO validation images and 81,434 MS COCO testing images (images are obtained from [MS COCO website] (http://mscoco.org/dataset/#download))\n\t- 248,349 questions for training, 121,512 questions for validation and 244,302 questions for testing (3 per image)\n\t- 2,483,490 answers for training and 1,215,120 answers for validation (10 per question)\n- Abstract\n\t- 20,000 training images, 10,000 validation images and 20,000 MS COCO testing images\n\t- 60,000 questions for training, 30,000 questions for validation and 60,000 questions for testing (3 per image)\n\t- 600,000 answers for training and 300,000 answers for validation (10 per question)\n\nThere are two types of tasks\n- Open-ended task\n- Multiple-choice task (18 choices per question)\n\n## Requirements ##\n- python 2.7\n- scikit-image (visit [this page](http://scikit-image.org/docs/dev/install.html) for installation)\n- matplotlib (visit [this page](http://matplotlib.org/users/installing.html) for installation)\n\n## Files ##\n./Questions\n- For v2.0, download the question files from the [VQA download page](http://www.visualqa.org/download.html), extract them and place in this folder.\n- For v1.0, both real and abstract, question files can be found on the [VQA v1 download page](http://www.visualqa.org/vqa_v1_download.html).\n- Question files from Beta v0.9 release (123,287 MSCOCO train and val images, 369,861 questions, 3,698,610 answers) can be found below\n\t- [training question files](http://visualqa.org/data/mscoco/prev_rel/Beta_v0.9/Questions_Train_mscoco.zip)\n\t- [validation question files](http://visualqa.org/data/mscoco/prev_rel/Beta_v0.9/Questions_Val_mscoco.zip)\n- Question files from Beta v0.1 release (10k MSCOCO images, 30k questions, 300k answers) can be found [here](http://visualqa.org/data/mscoco/prev_rel/Beta_v0.1/Questions_Train_mscoco.zip).\n\n./Annotations\n- For v2.0, download the annotations files from the [VQA download page](http://www.visualqa.org/download.html), extract them and place in this folder.\n- For v1.0, for both real and abstract, annotation files can be found on the [VQA v1 download page](http://www.visualqa.org/vqa_v1_download.html).\n- Annotation files from Beta v0.9 release (123,287 MSCOCO train and val images, 369,861 questions, 3,698,610 answers) can be found below\n\t- [training annotation files](http://visualqa.org/data/mscoco/prev_rel/Beta_v0.9/Annotations_Train_mscoco.zip)\n\t- [validation annotation files](http://visualqa.org/data/mscoco/prev_rel/Beta_v0.9/Annotations_Val_mscoco.zip)\n- Annotation files from Beta v0.1 release (10k MSCOCO images, 30k questions, 300k answers) can be found [here](http://visualqa.org/data/mscoco/prev_rel/Beta_v0.1/Annotations_Train_mscoco.zip).\n\n./Images\n- For real, create a directory with name mscoco inside this directory. For each of train, val and test, create directories with names train2014, val2014 and test2015 respectively inside mscoco directory, download respective images from [MS COCO website](http://mscoco.org/dataset/#download) and place them in respective folders.\n- For abstract, create a directory with name abstract_v002 inside this directory. For each of train, val and test, create directories with names train2015, val2015 and test2015 respectively inside abstract_v002 directory, download respective images from [VQA download page](http://www.visualqa.org/download.html) and place them in respective folders.\n\n./PythonHelperTools\n- This directory contains the Python API to read and visualize the VQA dataset\n- vqaDemo.py (demo script)\n- vqaTools (API to read and visualize data)\n\n./PythonEvaluationTools\n- This directory contains the Python evaluation code\n- vqaEvalDemo.py (evaluation demo script)\n- vqaEvaluation (evaluation code)\n\n./Results\n- OpenEnded_mscoco_train2014_fake_results.json (an example of a fake results file for v1.0 to run the demo)\n- Visit [VQA evaluation page] (http://visualqa.org/evaluation) for more details.\n\n./QuestionTypes\n- This directory contains the following lists of question types for both real and abstract questions (question types are unchanged from v1.0 to v2.0). In a list, if there are question types of length n+k and length n with the same first n words, then the question type of length n does not include questions that belong to the question type of length n+k.\n- mscoco_question_types.txt\n- abstract_v002_question_types.txt\n\n## References ##\n- [VQA: Visual Question Answering](http://visualqa.org/)\n- [Microsoft COCO](http://mscoco.org/)\n\n## Developers ##\n- Aishwarya Agrawal (Virginia Tech)\n- Code for API is based on [MSCOCO API code](https://github.com/pdollar/coco).\n- The format of the code for evaluation is based on [MSCOCO evaluation code](https://github.com/tylin/coco-caption).\n" }, { "path": "minigpt4/common/vqa_tools/VQA/license.txt", "content": "Copyright (c) 2014, Aishwarya Agrawal\nAll rights reserved.\n\nRedistribution and use in source and binary forms, with or without\nmodification, are permitted provided that the following conditions are met: \n\n1. Redistributions of source code must retain the above copyright notice, this\n list of conditions and the following disclaimer. \n2. Redistributions in binary form must reproduce the above copyright notice,\n this list of conditions and the following disclaimer in the documentation\n and/or other materials provided with the distribution. \n\nTHIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS \"AS IS\"\nAND\nANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED\nWARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE\nDISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE\nFOR\nANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES\n(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;\nLOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND\nON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\n(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS\nSOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\n\nThe views and conclusions contained in the software and documentation are\nthose\nof the authors and should not be interpreted as representing official\npolicies, \neither expressed or implied, of the FreeBSD Project.\n" }, { "path": "minigpt4/common/vqa_tools/__init__.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\n__author__ = \"aagrawal\"\n" }, { "path": "minigpt4/common/vqa_tools/vqa.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\n__author__ = \"aagrawal\"\n__version__ = \"0.9\"\n\n# Interface for accessing the VQA dataset.\n\n# This code is based on the code written by Tsung-Yi Lin for MSCOCO Python API available at the following link:\n# (https://github.com/pdollar/coco/blob/master/PythonAPI/pycocotools/coco.py).\n\n# The following functions are defined:\n# VQA - VQA class that loads VQA annotation file and prepares data structures.\n# getQuesIds - Get question ids that satisfy given filter conditions.\n# getImgIds - Get image ids that satisfy given filter conditions.\n# loadQA - Load questions and answers with the specified question ids.\n# showQA - Display the specified questions and answers.\n# loadRes - Load result file and create result object.\n\n# Help on each function can be accessed by: \"help(COCO.function)\"\n\nimport json\nimport datetime\nimport copy\n\n\nclass VQA:\n def __init__(self, annotation_file=None, question_file=None):\n \"\"\"\n Constructor of VQA helper class for reading and visualizing questions and answers.\n :param annotation_file (str): location of VQA annotation file\n :return:\n \"\"\"\n # load dataset\n self.dataset = {}\n self.questions = {}\n self.qa = {}\n self.qqa = {}\n self.imgToQA = {}\n if not annotation_file == None and not question_file == None:\n print(\"loading VQA annotations and questions into memory...\")\n time_t = datetime.datetime.utcnow()\n dataset = json.load(open(annotation_file, \"r\"))\n questions = json.load(open(question_file, \"r\"))\n self.dataset = dataset\n self.questions = questions\n self.createIndex()\n\n def createIndex(self):\n # create index\n print(\"creating index...\")\n imgToQA = {ann[\"image_id\"]: [] for ann in self.dataset[\"annotations\"]}\n qa = {ann[\"question_id\"]: [] for ann in self.dataset[\"annotations\"]}\n qqa = {ann[\"question_id\"]: [] for ann in self.dataset[\"annotations\"]}\n for ann in self.dataset[\"annotations\"]:\n imgToQA[ann[\"image_id\"]] += [ann]\n qa[ann[\"question_id\"]] = ann\n for ques in self.questions[\"questions\"]:\n qqa[ques[\"question_id\"]] = ques\n print(\"index created!\")\n\n # create class members\n self.qa = qa\n self.qqa = qqa\n self.imgToQA = imgToQA\n\n def info(self):\n \"\"\"\n Print information about the VQA annotation file.\n :return:\n \"\"\"\n for key, value in self.datset[\"info\"].items():\n print(\"%s: %s\" % (key, value))\n\n def getQuesIds(self, imgIds=[], quesTypes=[], ansTypes=[]):\n \"\"\"\n Get question ids that satisfy given filter conditions. default skips that filter\n :param \timgIds (int array) : get question ids for given imgs\n quesTypes (str array) : get question ids for given question types\n ansTypes (str array) : get question ids for given answer types\n :return: ids (int array) : integer array of question ids\n \"\"\"\n imgIds = imgIds if type(imgIds) == list else [imgIds]\n quesTypes = quesTypes if type(quesTypes) == list else [quesTypes]\n ansTypes = ansTypes if type(ansTypes) == list else [ansTypes]\n\n if len(imgIds) == len(quesTypes) == len(ansTypes) == 0:\n anns = self.dataset[\"annotations\"]\n else:\n if not len(imgIds) == 0:\n anns = sum(\n [self.imgToQA[imgId] for imgId in imgIds if imgId in self.imgToQA],\n [],\n )\n else:\n anns = self.dataset[\"annotations\"]\n anns = (\n anns\n if len(quesTypes) == 0\n else [ann for ann in anns if ann[\"question_type\"] in quesTypes]\n )\n anns = (\n anns\n if len(ansTypes) == 0\n else [ann for ann in anns if ann[\"answer_type\"] in ansTypes]\n )\n ids = [ann[\"question_id\"] for ann in anns]\n return ids\n\n def getImgIds(self, quesIds=[], quesTypes=[], ansTypes=[]):\n \"\"\"\n Get image ids that satisfy given filter conditions. default skips that filter\n :param quesIds (int array) : get image ids for given question ids\n quesTypes (str array) : get image ids for given question types\n ansTypes (str array) : get image ids for given answer types\n :return: ids (int array) : integer array of image ids\n \"\"\"\n quesIds = quesIds if type(quesIds) == list else [quesIds]\n quesTypes = quesTypes if type(quesTypes) == list else [quesTypes]\n ansTypes = ansTypes if type(ansTypes) == list else [ansTypes]\n\n if len(quesIds) == len(quesTypes) == len(ansTypes) == 0:\n anns = self.dataset[\"annotations\"]\n else:\n if not len(quesIds) == 0:\n anns = sum(\n [self.qa[quesId] for quesId in quesIds if quesId in self.qa], []\n )\n else:\n anns = self.dataset[\"annotations\"]\n anns = (\n anns\n if len(quesTypes) == 0\n else [ann for ann in anns if ann[\"question_type\"] in quesTypes]\n )\n anns = (\n anns\n if len(ansTypes) == 0\n else [ann for ann in anns if ann[\"answer_type\"] in ansTypes]\n )\n ids = [ann[\"image_id\"] for ann in anns]\n return ids\n\n def loadQA(self, ids=[]):\n \"\"\"\n Load questions and answers with the specified question ids.\n :param ids (int array) : integer ids specifying question ids\n :return: qa (object array) : loaded qa objects\n \"\"\"\n if type(ids) == list:\n return [self.qa[id] for id in ids]\n elif type(ids) == int:\n return [self.qa[ids]]\n\n def showQA(self, anns):\n \"\"\"\n Display the specified annotations.\n :param anns (array of object): annotations to display\n :return: None\n \"\"\"\n if len(anns) == 0:\n return 0\n for ann in anns:\n quesId = ann[\"question_id\"]\n print(\"Question: %s\" % (self.qqa[quesId][\"question\"]))\n for ans in ann[\"answers\"]:\n print(\"Answer %d: %s\" % (ans[\"answer_id\"], ans[\"answer\"]))\n\n def loadRes(self, resFile, quesFile):\n \"\"\"\n Load result file and return a result object.\n :param resFile (str) : file name of result file\n :return: res (obj) : result api object\n \"\"\"\n res = VQA()\n res.questions = json.load(open(quesFile))\n res.dataset[\"info\"] = copy.deepcopy(self.questions[\"info\"])\n res.dataset[\"task_type\"] = copy.deepcopy(self.questions[\"task_type\"])\n res.dataset[\"data_type\"] = copy.deepcopy(self.questions[\"data_type\"])\n res.dataset[\"data_subtype\"] = copy.deepcopy(self.questions[\"data_subtype\"])\n res.dataset[\"license\"] = copy.deepcopy(self.questions[\"license\"])\n\n print(\"Loading and preparing results... \")\n time_t = datetime.datetime.utcnow()\n anns = json.load(open(resFile))\n assert type(anns) == list, \"results is not an array of objects\"\n annsQuesIds = [ann[\"question_id\"] for ann in anns]\n assert set(annsQuesIds) == set(\n self.getQuesIds()\n ), \"Results do not correspond to current VQA set. Either the results do not have predictions for all question ids in annotation file or there is atleast one question id that does not belong to the question ids in the annotation file.\"\n for ann in anns:\n quesId = ann[\"question_id\"]\n if res.dataset[\"task_type\"] == \"Multiple Choice\":\n assert (\n ann[\"answer\"] in self.qqa[quesId][\"multiple_choices\"]\n ), \"predicted answer is not one of the multiple choices\"\n qaAnn = self.qa[quesId]\n ann[\"image_id\"] = qaAnn[\"image_id\"]\n ann[\"question_type\"] = qaAnn[\"question_type\"]\n ann[\"answer_type\"] = qaAnn[\"answer_type\"]\n print(\n \"DONE (t=%0.2fs)\" % ((datetime.datetime.utcnow() - time_t).total_seconds())\n )\n\n res.dataset[\"annotations\"] = anns\n res.createIndex()\n return res\n" }, { "path": "minigpt4/common/vqa_tools/vqa_eval.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\n# coding=utf-8\n\n__author__ = \"aagrawal\"\n\n# This code is based on the code written by Tsung-Yi Lin for MSCOCO Python API available at the following link:\n# (https://github.com/tylin/coco-caption/blob/master/pycocoevalcap/eval.py).\nimport sys\nimport re\n\n\nclass VQAEval:\n def __init__(self, vqa=None, vqaRes=None, n=2):\n self.n = n\n self.accuracy = {}\n self.evalQA = {}\n self.evalQuesType = {}\n self.evalAnsType = {}\n self.vqa = vqa\n self.vqaRes = vqaRes\n if vqa is not None:\n self.params = {\"question_id\": vqa.getQuesIds()}\n self.contractions = {\n \"aint\": \"ain't\",\n \"arent\": \"aren't\",\n \"cant\": \"can't\",\n \"couldve\": \"could've\",\n \"couldnt\": \"couldn't\",\n \"couldn'tve\": \"couldn't've\",\n \"couldnt've\": \"couldn't've\",\n \"didnt\": \"didn't\",\n \"doesnt\": \"doesn't\",\n \"dont\": \"don't\",\n \"hadnt\": \"hadn't\",\n \"hadnt've\": \"hadn't've\",\n \"hadn'tve\": \"hadn't've\",\n \"hasnt\": \"hasn't\",\n \"havent\": \"haven't\",\n \"hed\": \"he'd\",\n \"hed've\": \"he'd've\",\n \"he'dve\": \"he'd've\",\n \"hes\": \"he's\",\n \"howd\": \"how'd\",\n \"howll\": \"how'll\",\n \"hows\": \"how's\",\n \"Id've\": \"I'd've\",\n \"I'dve\": \"I'd've\",\n \"Im\": \"I'm\",\n \"Ive\": \"I've\",\n \"isnt\": \"isn't\",\n \"itd\": \"it'd\",\n \"itd've\": \"it'd've\",\n \"it'dve\": \"it'd've\",\n \"itll\": \"it'll\",\n \"let's\": \"let's\",\n \"maam\": \"ma'am\",\n \"mightnt\": \"mightn't\",\n \"mightnt've\": \"mightn't've\",\n \"mightn'tve\": \"mightn't've\",\n \"mightve\": \"might've\",\n \"mustnt\": \"mustn't\",\n \"mustve\": \"must've\",\n \"neednt\": \"needn't\",\n \"notve\": \"not've\",\n \"oclock\": \"o'clock\",\n \"oughtnt\": \"oughtn't\",\n \"ow's'at\": \"'ow's'at\",\n \"'ows'at\": \"'ow's'at\",\n \"'ow'sat\": \"'ow's'at\",\n \"shant\": \"shan't\",\n \"shed've\": \"she'd've\",\n \"she'dve\": \"she'd've\",\n \"she's\": \"she's\",\n \"shouldve\": \"should've\",\n \"shouldnt\": \"shouldn't\",\n \"shouldnt've\": \"shouldn't've\",\n \"shouldn'tve\": \"shouldn't've\",\n \"somebody'd\": \"somebodyd\",\n \"somebodyd've\": \"somebody'd've\",\n \"somebody'dve\": \"somebody'd've\",\n \"somebodyll\": \"somebody'll\",\n \"somebodys\": \"somebody's\",\n \"someoned\": \"someone'd\",\n \"someoned've\": \"someone'd've\",\n \"someone'dve\": \"someone'd've\",\n \"someonell\": \"someone'll\",\n \"someones\": \"someone's\",\n \"somethingd\": \"something'd\",\n \"somethingd've\": \"something'd've\",\n \"something'dve\": \"something'd've\",\n \"somethingll\": \"something'll\",\n \"thats\": \"that's\",\n \"thered\": \"there'd\",\n \"thered've\": \"there'd've\",\n \"there'dve\": \"there'd've\",\n \"therere\": \"there're\",\n \"theres\": \"there's\",\n \"theyd\": \"they'd\",\n \"theyd've\": \"they'd've\",\n \"they'dve\": \"they'd've\",\n \"theyll\": \"they'll\",\n \"theyre\": \"they're\",\n \"theyve\": \"they've\",\n \"twas\": \"'twas\",\n \"wasnt\": \"wasn't\",\n \"wed've\": \"we'd've\",\n \"we'dve\": \"we'd've\",\n \"weve\": \"we've\",\n \"werent\": \"weren't\",\n \"whatll\": \"what'll\",\n \"whatre\": \"what're\",\n \"whats\": \"what's\",\n \"whatve\": \"what've\",\n \"whens\": \"when's\",\n \"whered\": \"where'd\",\n \"wheres\": \"where's\",\n \"whereve\": \"where've\",\n \"whod\": \"who'd\",\n \"whod've\": \"who'd've\",\n \"who'dve\": \"who'd've\",\n \"wholl\": \"who'll\",\n \"whos\": \"who's\",\n \"whove\": \"who've\",\n \"whyll\": \"why'll\",\n \"whyre\": \"why're\",\n \"whys\": \"why's\",\n \"wont\": \"won't\",\n \"wouldve\": \"would've\",\n \"wouldnt\": \"wouldn't\",\n \"wouldnt've\": \"wouldn't've\",\n \"wouldn'tve\": \"wouldn't've\",\n \"yall\": \"y'all\",\n \"yall'll\": \"y'all'll\",\n \"y'allll\": \"y'all'll\",\n \"yall'd've\": \"y'all'd've\",\n \"y'alld've\": \"y'all'd've\",\n \"y'all'dve\": \"y'all'd've\",\n \"youd\": \"you'd\",\n \"youd've\": \"you'd've\",\n \"you'dve\": \"you'd've\",\n \"youll\": \"you'll\",\n \"youre\": \"you're\",\n \"youve\": \"you've\",\n }\n self.manualMap = {\n \"none\": \"0\",\n \"zero\": \"0\",\n \"one\": \"1\",\n \"two\": \"2\",\n \"three\": \"3\",\n \"four\": \"4\",\n \"five\": \"5\",\n \"six\": \"6\",\n \"seven\": \"7\",\n \"eight\": \"8\",\n \"nine\": \"9\",\n \"ten\": \"10\",\n }\n self.articles = [\"a\", \"an\", \"the\"]\n\n self.periodStrip = re.compile(\"(?!<=\\d)(\\.)(?!\\d)\")\n self.commaStrip = re.compile(\"(\\d)(,)(\\d)\")\n self.punct = [\n \";\",\n r\"/\",\n \"[\",\n \"]\",\n '\"',\n \"{\",\n \"}\",\n \"(\",\n \")\",\n \"=\",\n \"+\",\n \"\\\\\",\n \"_\",\n \"-\",\n \">\",\n \"<\",\n \"@\",\n \"`\",\n \",\",\n \"?\",\n \"!\",\n ]\n\n def evaluate(self, quesIds=None):\n if quesIds == None:\n quesIds = [quesId for quesId in self.params[\"question_id\"]]\n gts = {}\n res = {}\n for quesId in quesIds:\n gts[quesId] = self.vqa.qa[quesId]\n res[quesId] = self.vqaRes.qa[quesId]\n\n # =================================================\n # Compute accuracy\n # =================================================\n accQA = []\n accQuesType = {}\n accAnsType = {}\n print(\"computing accuracy\")\n step = 0\n for quesId in quesIds:\n resAns = res[quesId][\"answer\"]\n resAns = resAns.replace(\"\\n\", \" \")\n resAns = resAns.replace(\"\\t\", \" \")\n resAns = resAns.strip()\n resAns = self.processPunctuation(resAns)\n resAns = self.processDigitArticle(resAns)\n gtAcc = []\n gtAnswers = [ans[\"answer\"] for ans in gts[quesId][\"answers\"]]\n if len(set(gtAnswers)) > 1:\n for ansDic in gts[quesId][\"answers\"]:\n ansDic[\"answer\"] = self.processPunctuation(ansDic[\"answer\"])\n for gtAnsDatum in gts[quesId][\"answers\"]:\n otherGTAns = [\n item for item in gts[quesId][\"answers\"] if item != gtAnsDatum\n ]\n matchingAns = [item for item in otherGTAns if item[\"answer\"] == resAns]\n acc = min(1, float(len(matchingAns)) / 3)\n gtAcc.append(acc)\n quesType = gts[quesId][\"question_type\"]\n ansType = gts[quesId][\"answer_type\"]\n avgGTAcc = float(sum(gtAcc)) / len(gtAcc)\n accQA.append(avgGTAcc)\n if quesType not in accQuesType:\n accQuesType[quesType] = []\n accQuesType[quesType].append(avgGTAcc)\n if ansType not in accAnsType:\n accAnsType[ansType] = []\n accAnsType[ansType].append(avgGTAcc)\n self.setEvalQA(quesId, avgGTAcc)\n self.setEvalQuesType(quesId, quesType, avgGTAcc)\n self.setEvalAnsType(quesId, ansType, avgGTAcc)\n if step % 100 == 0:\n self.updateProgress(step / float(len(quesIds)))\n step = step + 1\n\n self.setAccuracy(accQA, accQuesType, accAnsType)\n print(\"Done computing accuracy\")\n\n def processPunctuation(self, inText):\n outText = inText\n for p in self.punct:\n if (p + \" \" in inText or \" \" + p in inText) or (\n re.search(self.commaStrip, inText) != None\n ):\n outText = outText.replace(p, \"\")\n else:\n outText = outText.replace(p, \" \")\n outText = self.periodStrip.sub(\"\", outText, re.UNICODE)\n return outText\n\n def processDigitArticle(self, inText):\n outText = []\n tempText = inText.lower().split()\n for word in tempText:\n word = self.manualMap.setdefault(word, word)\n if word not in self.articles:\n outText.append(word)\n else:\n pass\n for wordId, word in enumerate(outText):\n if word in self.contractions:\n outText[wordId] = self.contractions[word]\n outText = \" \".join(outText)\n return outText\n\n def setAccuracy(self, accQA, accQuesType, accAnsType):\n self.accuracy[\"overall\"] = round(100 * float(sum(accQA)) / len(accQA), self.n)\n self.accuracy[\"perQuestionType\"] = {\n quesType: round(\n 100 * float(sum(accQuesType[quesType])) / len(accQuesType[quesType]),\n self.n,\n )\n for quesType in accQuesType\n }\n self.accuracy[\"perAnswerType\"] = {\n ansType: round(\n 100 * float(sum(accAnsType[ansType])) / len(accAnsType[ansType]), self.n\n )\n for ansType in accAnsType\n }\n\n def setEvalQA(self, quesId, acc):\n self.evalQA[quesId] = round(100 * acc, self.n)\n\n def setEvalQuesType(self, quesId, quesType, acc):\n if quesType not in self.evalQuesType:\n self.evalQuesType[quesType] = {}\n self.evalQuesType[quesType][quesId] = round(100 * acc, self.n)\n\n def setEvalAnsType(self, quesId, ansType, acc):\n if ansType not in self.evalAnsType:\n self.evalAnsType[ansType] = {}\n self.evalAnsType[ansType][quesId] = round(100 * acc, self.n)\n\n def updateProgress(self, progress):\n barLength = 20\n status = \"\"\n if isinstance(progress, int):\n progress = float(progress)\n if not isinstance(progress, float):\n progress = 0\n status = \"error: progress var must be float\\r\\n\"\n if progress < 0:\n progress = 0\n status = \"Halt...\\r\\n\"\n if progress >= 1:\n progress = 1\n status = \"Done...\\r\\n\"\n block = int(round(barLength * progress))\n text = \"\\rFinshed Percent: [{0}] {1}% {2}\".format(\n \"#\" * block + \"-\" * (barLength - block), int(progress * 100), status\n )\n sys.stdout.write(text)\n sys.stdout.flush()\n" }, { "path": "minigpt4/configs/datasets/aokvqa/defaults.yaml", "content": " # Copyright (c) 2022, salesforce.com, inc.\n # All rights reserved.\n # SPDX-License-Identifier: BSD-3-Clause\n # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\ndatasets:\n aok_vqa:\n # data_dir: ${env.data_dir}/datasets\n data_type: images # [images|videos|features]\n\n build_info:\n # Be careful not to append minus sign (-) before split to avoid itemizing\n annotations:\n train:\n url:\n - https://storage.googleapis.com/sfr-vision-language-research/LAVIS/datasets/aokvqa/aokvqa_v1p0_train.json\n storage:\n - /path/to/aokvqa_v1p0_train.json\n images:\n storage: /path/to/coco/images" }, { "path": "minigpt4/configs/datasets/cc_sbu/align.yaml", "content": "datasets:\n cc_sbu_align:\n data_type: images\n build_info:\n storage: /path/to/cc_sbu_align/\n" }, { "path": "minigpt4/configs/datasets/cc_sbu/defaults.yaml", "content": "datasets:\n cc_sbu:\n data_type: images\n build_info:\n storage: /path/to/cc_sbu_dataset/{00000..01255}.tar\n" }, { "path": "minigpt4/configs/datasets/coco/caption.yaml", "content": " # Copyright (c) 2022, salesforce.com, inc.\n # All rights reserved.\n # SPDX-License-Identifier: BSD-3-Clause\n # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\ndatasets:\n coco_caption: # name of the dataset builder\n # dataset_card: dataset_card/coco_caption.md\n # data_dir: ${env.data_dir}/datasets\n data_type: images # [images|videos|features]\n\n build_info:\n # Be careful not to append minus sign (-) before split to avoid itemizing\n annotations:\n train:\n url: https://storage.googleapis.com/sfr-vision-language-research/datasets/coco_karpathy_train.json\n md5: aa31ac474cf6250ebb81d18348a07ed8\n storage: /path/to/coco_caption/coco_karpathy_train.json\n images:\n storage: /path/to/coco/images\n \n" }, { "path": "minigpt4/configs/datasets/coco/defaults_vqa.yaml", "content": " # Copyright (c) 2022, salesforce.com, inc.\n # All rights reserved.\n # SPDX-License-Identifier: BSD-3-Clause\n # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\ndatasets:\n coco_vqa:\n # data_dir: ${env.data_dir}/datasets\n data_type: images # [images|videos|features]\n\n build_info:\n\n annotations:\n train:\n url:\n - https://storage.googleapis.com/sfr-vision-language-research/LAVIS/datasets/vqav2/vqa_train.json\n - https://storage.googleapis.com/sfr-vision-language-research/LAVIS/datasets/vqav2/vqa_val.json\n storage:\n - /path/to/vqav2/vqa_train.json\n - /path/to/vqav2/vqa_val.json\n images:\n storage: /path/to/coco/images\n\n " }, { "path": "minigpt4/configs/datasets/coco_bbox/invrefcoco.yaml", "content": "datasets:\n invrefcoco:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/refcoco_annotations\n dataset: invrefcoco\n splitBy: unc" }, { "path": "minigpt4/configs/datasets/coco_bbox/invrefcocog.yaml", "content": "datasets:\n invrefcocog:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/refcoco_annotations\n dataset: invrefcocog\n splitBy: umd" }, { "path": "minigpt4/configs/datasets/coco_bbox/invrefcocop.yaml", "content": "datasets:\n invrefcocop:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/refcoco_annotations\n dataset: invrefcoco+\n splitBy: unc" }, { "path": "minigpt4/configs/datasets/coco_bbox/refcoco.yaml", "content": "datasets:\n refcoco:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/refcoco_annotations\n dataset: refcoco\n splitBy: unc" }, { "path": "minigpt4/configs/datasets/coco_bbox/refcocog.yaml", "content": "datasets:\n refcocog:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/refcoco_annotations\n dataset: refcocog\n splitBy: umd" }, { "path": "minigpt4/configs/datasets/coco_bbox/refcocop.yaml", "content": "datasets:\n refcocop:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/refcoco_annotations\n dataset: refcoco+\n splitBy: unc" }, { "path": "minigpt4/configs/datasets/flickr/caption_to_phrase.yaml", "content": "datasets:\n flickr_CaptionToPhrase:\n data_type: images\n build_info:\n image_path: /path/to/filtered_flikcr/images\n ann_path: /path/to/filtered_flickr/captiontobbox.json\n" }, { "path": "minigpt4/configs/datasets/flickr/default.yaml", "content": "datasets:\n flickr_grounded_caption:\n data_type: images\n build_info:\n image_path: /path/to/filtered_flikcr/images\n ann_path: /path/to/filtered_flikcr/groundedcaption.json\n" }, { "path": "minigpt4/configs/datasets/flickr/object_to_phrase.yaml", "content": "datasets:\n flickr_ObjectToPhrase:\n data_type: images\n build_info:\n image_path: /path/to/filtered_flikcr/images\n ann_path: /path/to/filtered_flikcr/phrasetobbox.json\n" }, { "path": "minigpt4/configs/datasets/gqa/balanced_val.yaml", "content": " # Copyright (c) 2022, salesforce.com, inc.\n # All rights reserved.\n # SPDX-License-Identifier: BSD-3-Clause\n # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\ndatasets:\n gqa:\n # data_dir: ${env.data_dir}/datasets\n data_type: images # [images|videos|features]\n\n build_info:\n # Be careful not to append minus sign (-) before split to avoid itemizing\n annotations:\n train:\n url:\n - https://storage.googleapis.com/sfr-vision-language-research/LAVIS/datasets/gqa/train_balanced_questions.json\n storage:\n - /path/to/gqa/train_balanced_questions.json\n\n images:\n storage: /path/to/gqa/images\n" }, { "path": "minigpt4/configs/datasets/laion/defaults.yaml", "content": "datasets:\n laion:\n data_type: images\n build_info:\n storage: /path/to/laion_dataset/{00000..10488}.tar\n" }, { "path": "minigpt4/configs/datasets/llava/conversation.yaml", "content": "datasets:\n\n llava_conversation:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/llava/conversation_58k.json" }, { "path": "minigpt4/configs/datasets/llava/detail.yaml", "content": "datasets:\n llava_detail:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/llava/detail_23k.json" }, { "path": "minigpt4/configs/datasets/llava/reason.yaml", "content": "datasets:\n\n llava_reason:\n data_type: images\n build_info:\n image_path: /path/to/coco/images\n ann_path: /path/to/llava/complex_reasoning_77k.json" }, { "path": "minigpt4/configs/datasets/multitask_conversation/default.yaml", "content": "datasets:\n multitask_conversation:\n data_type: images\n build_info:\n \n image_path: /path/to/coco/images\n ann_path: /path/to/multitask_conversation/multi_task_conversation.json" }, { "path": "minigpt4/configs/datasets/nlp/unnatural_instruction.yaml", "content": "datasets:\n unnatural_instruction:\n data_type: text\n build_info:\n ann_path: /path/to/unnatural_instructions/filtered_unnatural_instruction.json" }, { "path": "minigpt4/configs/datasets/ocrvqa/ocrvqa.yaml", "content": "datasets:\n ocrvqa:\n data_type: images\n build_info:\n image_path: /path/to/ocrvqa/images\n ann_path: /path/to/ocrvqa/dataset.json" }, { "path": "minigpt4/configs/datasets/okvqa/defaults.yaml", "content": " # Copyright (c) 2022, salesforce.com, inc.\n # All rights reserved.\n # SPDX-License-Identifier: BSD-3-Clause\n # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\ndatasets:\n ok_vqa:\n # data_dir: ${env.data_dir}/datasets\n data_type: images # [images|videos|features]\n\n build_info:\n # Be careful not to append minus sign (-) before split to avoid itemizing\n annotations:\n train:\n url:\n # TODO make this order insensitive\n - https://storage.googleapis.com/sfr-vision-language-research/LAVIS/datasets/okvqa/okvqa_train.json\n storage:\n - /path/to/okvqa/okvqa_train.json\n images:\n storage: /path/to/coco/images" }, { "path": "minigpt4/configs/datasets/textcaps/caption.yaml", "content": "datasets:\n textcaps_caption:\n data_type: images\n \n build_info:\n image_path: /path/to/textcaps/train_images\n ann_path: /path/to/textcaps/TextCaps_0.1_train.json\n\n\n" }, { "path": "minigpt4/configs/datasets/vg/ref.yaml", "content": "datasets:\n refvg:\n data_type: images\n build_info:\n data_dir: /path/to/visual_genome" }, { "path": "minigpt4/configs/default.yaml", "content": "env:\n # For default users\n # cache_root: \"cache\"\n # For internal use with persistent storage\n cache_root: \"/export/home/.cache/minigpt4\"\n" }, { "path": "minigpt4/configs/models/minigpt4_llama2.yaml", "content": "model:\n arch: minigpt4\n\n # vit encoder\n image_size: 224\n drop_path_rate: 0\n use_grad_checkpoint: False\n vit_precision: \"fp16\"\n freeze_vit: True\n has_qformer: False\n\n # generation configs\n prompt: \"\"\n\n llama_model: \"please set this value to the path of llama2-chat-7b\"\n\npreprocess:\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n eval:\n name: \"blip2_image_eval\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n eval:\n name: \"blip_caption\"\n" }, { "path": "minigpt4/configs/models/minigpt4_vicuna0.yaml", "content": "model:\n arch: minigpt4\n\n # vit encoder\n image_size: 224\n drop_path_rate: 0\n use_grad_checkpoint: False\n vit_precision: \"fp16\"\n freeze_vit: True\n freeze_qformer: True\n\n # Q-Former\n num_query_token: 32\n\n # generation configs\n prompt: \"\"\n\n llama_model: \"please set this value to the path of vicuna model\"\n\npreprocess:\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n eval:\n name: \"blip2_image_eval\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n eval:\n name: \"blip_caption\"\n" }, { "path": "minigpt4/configs/models/minigpt_v2.yaml", "content": "model:\n arch: minigpt_v2\n\n # vit encoder\n image_size: 448\n drop_path_rate: 0\n use_grad_checkpoint: False\n vit_precision: \"fp16\"\n freeze_vit: True\n\n # generation configs\n prompt: \"\"\n\n llama_model: \"please set this value to the path of llama2-chat-7b\"\n lora_r: 64\n lora_alpha: 16\n\n\npreprocess:\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n eval:\n name: \"blip2_image_eval\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n eval:\n name: \"blip_caption\"\n" }, { "path": "minigpt4/conversation/__init__.py", "content": "" }, { "path": "minigpt4/conversation/conversation.py", "content": "import argparse\nimport time\nfrom threading import Thread\nfrom PIL import Image\n\nimport torch\nfrom transformers import AutoTokenizer, AutoModelForCausalLM, LlamaTokenizer\nfrom transformers import StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer\n\nimport dataclasses\nfrom enum import auto, Enum\nfrom typing import List, Tuple, Any\n\nfrom minigpt4.common.registry import registry\n\n\nclass SeparatorStyle(Enum):\n \"\"\"Different separator style.\"\"\"\n SINGLE = auto()\n TWO = auto()\n\n\n@dataclasses.dataclass\nclass Conversation:\n \"\"\"A class that keeps all conversation history.\"\"\"\n system: str\n roles: List[str]\n messages: List[List[str]]\n offset: int\n # system_img: List[Image.Image] = []\n sep_style: SeparatorStyle = SeparatorStyle.SINGLE\n sep: str = \"###\"\n sep2: str = None\n\n skip_next: bool = False\n conv_id: Any = None\n\n def get_prompt(self):\n if self.sep_style == SeparatorStyle.SINGLE:\n ret = self.system + self.sep\n for role, message in self.messages:\n if message:\n ret += role + message + self.sep\n else:\n ret += role\n return ret\n elif self.sep_style == SeparatorStyle.TWO:\n seps = [self.sep, self.sep2]\n ret = self.system + seps[0]\n for i, (role, message) in enumerate(self.messages):\n if message:\n ret += role + message + seps[i % 2]\n else:\n ret += role\n return ret\n else:\n raise ValueError(f\"Invalid style: {self.sep_style}\")\n\n def append_message(self, role, message):\n self.messages.append([role, message])\n\n def to_gradio_chatbot(self):\n ret = []\n for i, (role, msg) in enumerate(self.messages[self.offset:]):\n if i % 2 == 0:\n ret.append([msg, None])\n else:\n ret[-1][-1] = msg\n return ret\n\n def copy(self):\n return Conversation(\n system=self.system,\n # system_img=self.system_img,\n roles=self.roles,\n messages=[[x, y] for x, y in self.messages],\n offset=self.offset,\n sep_style=self.sep_style,\n sep=self.sep,\n sep2=self.sep2,\n conv_id=self.conv_id)\n\n def dict(self):\n return {\n \"system\": self.system,\n # \"system_img\": self.system_img,\n \"roles\": self.roles,\n \"messages\": self.messages,\n \"offset\": self.offset,\n \"sep\": self.sep,\n \"sep2\": self.sep2,\n \"conv_id\": self.conv_id,\n }\n\n\nclass StoppingCriteriaSub(StoppingCriteria):\n\n def __init__(self, stops=[], encounters=1):\n super().__init__()\n self.stops = stops\n\n def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor):\n for stop in self.stops:\n if torch.all(input_ids[:, -len(stop):] == stop).item():\n return True\n\n return False\n\n\nCONV_VISION_Vicuna0 = Conversation(\n system=\"Give the following image: ImageContent. \"\n \"You will be able to see the image once I provide it to you. Please answer my questions.\",\n roles=(\"Human: \", \"Assistant: \"),\n messages=[],\n offset=2,\n sep_style=SeparatorStyle.SINGLE,\n sep=\"###\",\n)\n\nCONV_VISION_LLama2 = Conversation(\n system=\"Give the following image: ImageContent. \"\n \"You will be able to see the image once I provide it to you. Please answer my questions.\",\n roles=(\"[INST] \", \" [/INST] \"),\n messages=[],\n offset=2,\n sep_style=SeparatorStyle.SINGLE,\n sep=\"\",\n)\n\nCONV_VISION_minigptv2 = Conversation(\n system=\"\",\n roles=(\"[INST] \", \" [/INST]\"),\n messages=[],\n offset=2,\n sep_style=SeparatorStyle.SINGLE,\n sep=\"\",\n)\n\nclass Chat:\n def __init__(self, model, vis_processor, device='cuda:0', stopping_criteria=None):\n self.device = device\n self.model = model\n self.vis_processor = vis_processor\n\n if stopping_criteria is not None:\n self.stopping_criteria = stopping_criteria\n else:\n stop_words_ids = [torch.tensor([2]).to(self.device)]\n self.stopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(stops=stop_words_ids)])\n\n def ask(self, text, conv):\n if len(conv.messages) > 0 and conv.messages[-1][0] == conv.roles[0] \\\n and conv.messages[-1][1][-6:] == '': # last message is image.\n conv.messages[-1][1] = ' '.join([conv.messages[-1][1], text])\n else:\n conv.append_message(conv.roles[0], text)\n\n def answer_prepare(self, conv, img_list, max_new_tokens=300, num_beams=1, min_length=1, top_p=0.9,\n repetition_penalty=1.05, length_penalty=1, temperature=1.0, max_length=2000):\n conv.append_message(conv.roles[1], None)\n prompt = conv.get_prompt()\n embs = self.model.get_context_emb(prompt, img_list)\n\n current_max_len = embs.shape[1] + max_new_tokens\n if current_max_len - max_length > 0:\n print('Warning: The number of tokens in current conversation exceeds the max length. '\n 'The model will not see the contexts outside the range.')\n begin_idx = max(0, current_max_len - max_length)\n embs = embs[:, begin_idx:]\n\n generation_kwargs = dict(\n inputs_embeds=embs,\n max_new_tokens=max_new_tokens,\n stopping_criteria=self.stopping_criteria,\n num_beams=num_beams,\n do_sample=True,\n min_length=min_length,\n top_p=top_p,\n repetition_penalty=repetition_penalty,\n length_penalty=length_penalty,\n temperature=float(temperature),\n )\n return generation_kwargs\n\n def answer(self, conv, img_list, **kargs):\n generation_dict = self.answer_prepare(conv, img_list, **kargs)\n output_token = self.model_generate(**generation_dict)[0]\n output_text = self.model.llama_tokenizer.decode(output_token, skip_special_tokens=True)\n\n output_text = output_text.split('###')[0] # remove the stop sign '###'\n output_text = output_text.split('Assistant:')[-1].strip()\n\n conv.messages[-1][1] = output_text\n return output_text, output_token.cpu().numpy()\n\n def stream_answer(self, conv, img_list, **kargs):\n generation_kwargs = self.answer_prepare(conv, img_list, **kargs)\n streamer = TextIteratorStreamer(self.model.llama_tokenizer, skip_special_tokens=True)\n generation_kwargs['streamer'] = streamer\n thread = Thread(target=self.model_generate, kwargs=generation_kwargs)\n thread.start()\n return streamer\n\n def model_generate(self, *args, **kwargs):\n # for 8 bit and 16 bit compatibility\n with self.model.maybe_autocast():\n output = self.model.llama_model.generate(*args, **kwargs)\n return output\n\n def encode_img(self, img_list):\n image = img_list[0]\n img_list.pop(0)\n if isinstance(image, str): # is a image path\n raw_image = Image.open(image).convert('RGB')\n image = self.vis_processor(raw_image).unsqueeze(0).to(self.device)\n elif isinstance(image, Image.Image):\n raw_image = image\n image = self.vis_processor(raw_image).unsqueeze(0).to(self.device)\n elif isinstance(image, torch.Tensor):\n if len(image.shape) == 3:\n image = image.unsqueeze(0)\n image = image.to(self.device)\n\n image_emb, _ = self.model.encode_img(image)\n img_list.append(image_emb)\n\n def upload_img(self, image, conv, img_list):\n conv.append_message(conv.roles[0], \"\")\n img_list.append(image)\n msg = \"Received.\"\n\n return msg\n\n" }, { "path": "minigpt4/datasets/__init__.py", "content": "" }, { "path": "minigpt4/datasets/builders/__init__.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nfrom minigpt4.datasets.builders.base_dataset_builder import load_dataset_config\nfrom minigpt4.datasets.builders.image_text_pair_builder import (\n CCSBUBuilder,\n LaionBuilder,\n CCSBUAlignBuilder\n)\nfrom minigpt4.common.registry import registry\n\n__all__ = [\n \"CCSBUBuilder\",\n \"LaionBuilder\",\n \"CCSBUAlignBuilder\"\n]\n\n\ndef load_dataset(name, cfg_path=None, vis_path=None, data_type=None):\n \"\"\"\n Example\n\n >>> dataset = load_dataset(\"coco_caption\", cfg=None)\n >>> splits = dataset.keys()\n >>> print([len(dataset[split]) for split in splits])\n\n \"\"\"\n if cfg_path is None:\n cfg = None\n else:\n cfg = load_dataset_config(cfg_path)\n\n try:\n builder = registry.get_builder_class(name)(cfg)\n except TypeError:\n print(\n f\"Dataset {name} not found. Available datasets:\\n\"\n + \", \".join([str(k) for k in dataset_zoo.get_names()])\n )\n exit(1)\n\n if vis_path is not None:\n if data_type is None:\n # use default data type in the config\n data_type = builder.config.data_type\n\n assert (\n data_type in builder.config.build_info\n ), f\"Invalid data_type {data_type} for {name}.\"\n\n builder.config.build_info.get(data_type).storage = vis_path\n\n dataset = builder.build_datasets()\n return dataset\n\n\nclass DatasetZoo:\n def __init__(self) -> None:\n self.dataset_zoo = {\n k: list(v.DATASET_CONFIG_DICT.keys())\n for k, v in sorted(registry.mapping[\"builder_name_mapping\"].items())\n }\n\n def get_names(self):\n return list(self.dataset_zoo.keys())\n\n\ndataset_zoo = DatasetZoo()\n" }, { "path": "minigpt4/datasets/builders/base_dataset_builder.py", "content": "\"\"\"\n This file is from\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport logging\nimport os\nimport shutil\nimport warnings\n\nfrom omegaconf import OmegaConf\nimport torch.distributed as dist\nfrom torchvision.datasets.utils import download_url\n\nimport minigpt4.common.utils as utils\nfrom minigpt4.common.dist_utils import is_dist_avail_and_initialized, is_main_process\nfrom minigpt4.common.registry import registry\nfrom minigpt4.processors.base_processor import BaseProcessor\n\n\n\nclass BaseDatasetBuilder:\n train_dataset_cls, eval_dataset_cls = None, None\n\n def __init__(self, cfg=None):\n super().__init__()\n\n if cfg is None:\n # help to create datasets from default config.\n self.config = load_dataset_config(self.default_config_path())\n elif isinstance(cfg, str):\n self.config = load_dataset_config(cfg)\n else:\n # when called from task.build_dataset()\n self.config = cfg\n\n self.data_type = self.config.data_type\n\n self.vis_processors = {\"train\": BaseProcessor(), \"eval\": BaseProcessor()}\n self.text_processors = {\"train\": BaseProcessor(), \"eval\": BaseProcessor()}\n\n def build_datasets(self):\n # download, split, etc...\n # only called on 1 GPU/TPU in distributed\n\n if is_main_process():\n self._download_data()\n\n if is_dist_avail_and_initialized():\n dist.barrier()\n\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n datasets = self.build() # dataset['train'/'val'/'test']\n\n return datasets\n\n def build_processors(self):\n vis_proc_cfg = self.config.get(\"vis_processor\")\n txt_proc_cfg = self.config.get(\"text_processor\")\n\n if vis_proc_cfg is not None:\n vis_train_cfg = vis_proc_cfg.get(\"train\")\n vis_eval_cfg = vis_proc_cfg.get(\"eval\")\n\n self.vis_processors[\"train\"] = self._build_proc_from_cfg(vis_train_cfg)\n self.vis_processors[\"eval\"] = self._build_proc_from_cfg(vis_eval_cfg)\n\n if txt_proc_cfg is not None:\n txt_train_cfg = txt_proc_cfg.get(\"train\")\n txt_eval_cfg = txt_proc_cfg.get(\"eval\")\n\n self.text_processors[\"train\"] = self._build_proc_from_cfg(txt_train_cfg)\n self.text_processors[\"eval\"] = self._build_proc_from_cfg(txt_eval_cfg)\n\n @staticmethod\n def _build_proc_from_cfg(cfg):\n return (\n registry.get_processor_class(cfg.name).from_config(cfg)\n if cfg is not None\n else None\n )\n\n @classmethod\n def default_config_path(cls, type=\"default\"):\n return utils.get_abs_path(cls.DATASET_CONFIG_DICT[type])\n\n def _download_data(self):\n self._download_ann()\n self._download_vis()\n\n def _download_ann(self):\n \"\"\"\n Download annotation files if necessary.\n All the vision-language datasets should have annotations of unified format.\n\n storage_path can be:\n (1) relative/absolute: will be prefixed with env.cache_root to make full path if relative.\n (2) basename/dirname: will be suffixed with base name of URL if dirname is provided.\n\n Local annotation paths should be relative.\n \"\"\"\n anns = self.config.build_info.annotations\n\n splits = anns.keys()\n\n cache_root = registry.get_path(\"cache_root\")\n\n for split in splits:\n info = anns[split]\n\n urls, storage_paths = info.get(\"url\", None), info.storage\n\n if isinstance(urls, str):\n urls = [urls]\n if isinstance(storage_paths, str):\n storage_paths = [storage_paths]\n\n assert len(urls) == len(storage_paths)\n\n for url_or_filename, storage_path in zip(urls, storage_paths):\n # if storage_path is relative, make it full by prefixing with cache_root.\n if not os.path.isabs(storage_path):\n storage_path = os.path.join(cache_root, storage_path)\n\n dirname = os.path.dirname(storage_path)\n if not os.path.exists(dirname):\n os.makedirs(dirname)\n\n if os.path.isfile(url_or_filename):\n src, dst = url_or_filename, storage_path\n if not os.path.exists(dst):\n shutil.copyfile(src=src, dst=dst)\n else:\n logging.info(\"Using existing file {}.\".format(dst))\n else:\n if os.path.isdir(storage_path):\n # if only dirname is provided, suffix with basename of URL.\n raise ValueError(\n \"Expecting storage_path to be a file path, got directory {}\".format(\n storage_path\n )\n )\n else:\n filename = os.path.basename(storage_path)\n\n download_url(url=url_or_filename, root=dirname, filename=filename)\n\n def _download_vis(self):\n\n storage_path = self.config.build_info.get(self.data_type).storage\n storage_path = utils.get_cache_path(storage_path)\n\n if not os.path.exists(storage_path):\n warnings.warn(\n f\"\"\"\n The specified path {storage_path} for visual inputs does not exist.\n Please provide a correct path to the visual inputs or\n refer to datasets/download_scripts/README.md for downloading instructions.\n \"\"\"\n )\n\n def build(self):\n \"\"\"\n Create by split datasets inheriting torch.utils.data.Datasets.\n\n # build() can be dataset-specific. Overwrite to customize.\n \"\"\"\n self.build_processors()\n\n build_info = self.config.build_info\n\n ann_info = build_info.annotations\n vis_info = build_info.get(self.data_type)\n\n datasets = dict()\n for split in ann_info.keys():\n if split not in [\"train\", \"val\", \"test\"]:\n continue\n\n is_train = split == \"train\"\n\n # processors\n vis_processor = (\n self.vis_processors[\"train\"]\n if is_train\n else self.vis_processors[\"eval\"]\n )\n text_processor = (\n self.text_processors[\"train\"]\n if is_train\n else self.text_processors[\"eval\"]\n )\n\n # annotation path\n ann_paths = ann_info.get(split).storage\n if isinstance(ann_paths, str):\n ann_paths = [ann_paths]\n\n abs_ann_paths = []\n for ann_path in ann_paths:\n if not os.path.isabs(ann_path):\n ann_path = utils.get_cache_path(ann_path)\n abs_ann_paths.append(ann_path)\n ann_paths = abs_ann_paths\n\n # visual data storage path\n vis_path = os.path.join(vis_info.storage, split)\n\n if not os.path.isabs(vis_path):\n # vis_path = os.path.join(utils.get_cache_path(), vis_path)\n vis_path = utils.get_cache_path(vis_path)\n\n if not os.path.exists(vis_path):\n warnings.warn(\"storage path {} does not exist.\".format(vis_path))\n\n # create datasets\n dataset_cls = self.train_dataset_cls if is_train else self.eval_dataset_cls\n datasets[split] = dataset_cls(\n vis_processor=vis_processor,\n text_processor=text_processor,\n ann_paths=ann_paths,\n vis_root=vis_path,\n )\n\n return datasets\n\n\ndef load_dataset_config(cfg_path):\n cfg = OmegaConf.load(cfg_path).datasets\n cfg = cfg[list(cfg.keys())[0]]\n\n return cfg\n" }, { "path": "minigpt4/datasets/builders/image_text_pair_builder.py", "content": "import os\nimport logging\nimport warnings\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.datasets.builders.base_dataset_builder import BaseDatasetBuilder\nfrom minigpt4.datasets.datasets.laion_dataset import LaionDataset\nfrom minigpt4.datasets.datasets.cc_sbu_dataset import CCSBUDataset, CCSBUAlignDataset\nfrom minigpt4.datasets.datasets.text_caps import TextCapDataset\nfrom minigpt4.datasets.datasets.llava_dataset import LlavaDetailDataset, LlavaReasonDataset, LlavaConversationDataset\nfrom minigpt4.datasets.datasets.unnatural_instruction import UnnaturalDataset\nfrom minigpt4.datasets.datasets.multitask_conversation import MultiTaskConversationDataset\nfrom minigpt4.datasets.datasets.flickr import GroundedDetailDataset,CaptionToObjectDataset,PhraseToObjectDataset\nfrom minigpt4.datasets.datasets.vg_dataset import ReferVisualGenomeDataset\nfrom minigpt4.datasets.datasets.coco_dataset import ReferCOCODataset, InvReferCOCODataset\nfrom minigpt4.datasets.datasets.gqa_datasets import GQADataset\nfrom minigpt4.datasets.datasets.aok_vqa_datasets import AOKVQADataset\nfrom minigpt4.datasets.datasets.coco_vqa_datasets import COCOVQADataset\nfrom minigpt4.datasets.datasets.ocrvqa_dataset import OCRVQADataset\nfrom minigpt4.datasets.datasets.coco_caption import COCOCapDataset\n\n\n@registry.register_builder(\"multitask_conversation\")\nclass MultitaskConversationBuilder(BaseDatasetBuilder):\n train_dataset_cls = MultiTaskConversationDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/multitask_conversation/default.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n\n\n@registry.register_builder(\"unnatural_instruction\")\nclass UnnaturalInstructionBuilder(BaseDatasetBuilder):\n train_dataset_cls = UnnaturalDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/nlp/unnatural_instruction.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n )\n\n return datasets\n\n\n\n@registry.register_builder(\"llava_detail\")\nclass LlavaDetailBuilder(BaseDatasetBuilder):\n train_dataset_cls = LlavaDetailDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/llava/detail.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n \n\n\n@registry.register_builder(\"llava_reason\")\nclass LlavaReasonBuilder(BaseDatasetBuilder):\n train_dataset_cls = LlavaReasonDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/llava/reason.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n\n@registry.register_builder(\"llava_conversation\")\nclass LlavaReasonBuilder(BaseDatasetBuilder):\n train_dataset_cls = LlavaConversationDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/llava/conversation.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n\n\nclass AllRefCOCOBuilder(BaseDatasetBuilder):\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n\n build_info = self.config.build_info\n image_path = build_info.image_path\n ann_path = build_info.ann_path\n\n datasets = dict()\n\n if not os.path.exists(image_path):\n warnings.warn(\"image path {} does not exist.\".format(image_path))\n if not os.path.exists(ann_path):\n warnings.warn(\"ann path {} does not exist.\".format(ann_path))\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=ann_path,\n vis_root=image_path,\n dataset=build_info.dataset,\n splitBy=build_info.splitBy\n )\n\n return datasets\n \n\n@registry.register_builder(\"refcoco\")\nclass RefCOCOBuilder(AllRefCOCOBuilder):\n train_dataset_cls = ReferCOCODataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco_bbox/refcoco.yaml\",\n }\n\n@registry.register_builder(\"refcocop\")\nclass RefCOCOPBuilder(AllRefCOCOBuilder):\n train_dataset_cls = ReferCOCODataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco_bbox/refcocop.yaml\",\n }\n\n\n@registry.register_builder(\"refcocog\")\nclass RefCOCOGBuilder(AllRefCOCOBuilder):\n train_dataset_cls = ReferCOCODataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco_bbox/refcocog.yaml\",\n }\n\n@registry.register_builder(\"invrefcoco\")\nclass RefCOCOBuilder(AllRefCOCOBuilder):\n train_dataset_cls = InvReferCOCODataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco_bbox/invrefcoco.yaml\",\n }\n\n\n@registry.register_builder(\"invrefcocop\")\nclass RefCOCOPBuilder(AllRefCOCOBuilder):\n train_dataset_cls = InvReferCOCODataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco_bbox/invrefcocop.yaml\",\n }\n\n\n@registry.register_builder(\"invrefcocog\")\nclass RefCOCOGBuilder(AllRefCOCOBuilder):\n train_dataset_cls = InvReferCOCODataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco_bbox/invrefcocog.yaml\",\n }\n\n@registry.register_builder(\"refvg\")\nclass RefVisualGenomeBuilder(BaseDatasetBuilder):\n train_dataset_cls = ReferVisualGenomeDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/vg/ref.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n\n build_info = self.config.build_info\n data_dir = build_info.data_dir\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n data_dir=data_dir,\n )\n\n return datasets\n\n\n@registry.register_builder(\"textcaps_caption\")\nclass TextcapCaptionBuilder(BaseDatasetBuilder):\n train_dataset_cls = TextCapDataset\n\n DATASET_CONFIG_DICT = {\"default\": \"configs/datasets/textcaps/caption.yaml\"}\n\n def _download_ann(self):\n pass\n\n def _download_vis(self):\n pass\n\n def build(self):\n self.build_processors()\n\n build_info = self.config.build_info\n\n datasets = dict()\n split = \"train\"\n\n # create datasets\n # [NOTE] return inner_datasets (wds.DataPipeline)\n dataset_cls = self.train_dataset_cls\n datasets[split] = dataset_cls(\n vis_processor=self.vis_processors[split],\n text_processor=self.text_processors[split],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n \n@registry.register_builder(\"coco_vqa\")\nclass COCOVQABuilder(BaseDatasetBuilder):\n train_dataset_cls = COCOVQADataset\n\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco/defaults_vqa.yaml\",\n }\n\n@registry.register_builder(\"ok_vqa\")\nclass OKVQABuilder(COCOVQABuilder):\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/okvqa/defaults.yaml\",\n }\n\n\n@registry.register_builder(\"aok_vqa\")\nclass AOKVQABuilder(BaseDatasetBuilder):\n train_dataset_cls = AOKVQADataset\n\n DATASET_CONFIG_DICT = {\"default\": \"configs/datasets/aokvqa/defaults.yaml\"}\n\n\n@registry.register_builder(\"gqa\")\nclass GQABuilder(BaseDatasetBuilder):\n train_dataset_cls = GQADataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/gqa/balanced_val.yaml\",\n }\n\n\n\n\n@registry.register_builder(\"flickr_grounded_caption\")\nclass GroundedCaptionBuilder(BaseDatasetBuilder):\n train_dataset_cls = GroundedDetailDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/flickr/default.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n\n\n@registry.register_builder(\"flickr_CaptionToPhrase\")\nclass CaptionToPhraseBuilder(BaseDatasetBuilder):\n train_dataset_cls = CaptionToObjectDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/flickr/caption_to_phrase.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n\n@registry.register_builder(\"flickr_ObjectToPhrase\")\nclass CaptionToPhraseBuilder(BaseDatasetBuilder):\n train_dataset_cls = PhraseToObjectDataset\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/flickr/object_to_phrase.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n build_info = self.config.build_info\n datasets = dict()\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_path=build_info.ann_path,\n vis_root=build_info.image_path,\n )\n\n return datasets\n\n\n\n\nclass DocumentVQABuilder(BaseDatasetBuilder):\n def _download_ann(self):\n pass\n\n def _download_vis(self):\n pass\n\n def build(self):\n self.build_processors()\n build_info = self.config.build_info\n\n datasets = dict()\n split = \"train\"\n\n dataset_cls = self.train_dataset_cls\n datasets[split] = dataset_cls(\n vis_processor=self.vis_processors[split],\n text_processor=self.text_processors[split],\n vis_root=build_info.image_path,\n ann_path=build_info.ann_path\n )\n\n return datasets\n \n\n@registry.register_builder(\"ocrvqa\")\nclass OCRVQABuilder(DocumentVQABuilder):\n train_dataset_cls = OCRVQADataset\n DATASET_CONFIG_DICT = {\"default\": \"configs/datasets/ocrvqa/ocrvqa.yaml\"}\n\n\n@registry.register_builder(\"cc_sbu\")\nclass CCSBUBuilder(BaseDatasetBuilder):\n train_dataset_cls = CCSBUDataset\n\n DATASET_CONFIG_DICT = {\"default\": \"configs/datasets/cc_sbu/defaults.yaml\"}\n\n def _download_ann(self):\n pass\n\n def _download_vis(self):\n pass\n\n def build(self):\n self.build_processors()\n\n build_info = self.config.build_info\n\n datasets = dict()\n split = \"train\"\n\n # create datasets\n # [NOTE] return inner_datasets (wds.DataPipeline)\n dataset_cls = self.train_dataset_cls\n datasets[split] = dataset_cls(\n vis_processor=self.vis_processors[split],\n text_processor=self.text_processors[split],\n location=build_info.storage,\n ).inner_dataset\n\n return datasets\n\n\n@registry.register_builder(\"laion\")\nclass LaionBuilder(BaseDatasetBuilder):\n train_dataset_cls = LaionDataset\n\n DATASET_CONFIG_DICT = {\"default\": \"configs/datasets/laion/defaults.yaml\"}\n\n def _download_ann(self):\n pass\n\n def _download_vis(self):\n pass\n\n def build(self):\n self.build_processors()\n\n build_info = self.config.build_info\n\n datasets = dict()\n split = \"train\"\n\n # create datasets\n # [NOTE] return inner_datasets (wds.DataPipeline)\n dataset_cls = self.train_dataset_cls\n datasets[split] = dataset_cls(\n vis_processor=self.vis_processors[split],\n text_processor=self.text_processors[split],\n location=build_info.storage,\n ).inner_dataset\n\n return datasets\n\n\n\n@registry.register_builder(\"coco_caption\")\nclass COCOCapBuilder(BaseDatasetBuilder):\n train_dataset_cls = COCOCapDataset\n\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/coco/caption.yaml\",\n }\n\n\n\n@registry.register_builder(\"cc_sbu_align\")\nclass CCSBUAlignBuilder(BaseDatasetBuilder):\n train_dataset_cls = CCSBUAlignDataset\n\n DATASET_CONFIG_DICT = {\n \"default\": \"configs/datasets/cc_sbu/align.yaml\",\n }\n\n def build_datasets(self):\n # at this point, all the annotations and image/videos should be all downloaded to the specified locations.\n logging.info(\"Building datasets...\")\n self.build_processors()\n\n build_info = self.config.build_info\n storage_path = build_info.storage\n\n datasets = dict()\n\n if not os.path.exists(storage_path):\n warnings.warn(\"storage path {} does not exist.\".format(storage_path))\n\n # create datasets\n dataset_cls = self.train_dataset_cls\n datasets['train'] = dataset_cls(\n vis_processor=self.vis_processors[\"train\"],\n text_processor=self.text_processors[\"train\"],\n ann_paths=[os.path.join(storage_path, 'filter_cap.json')],\n vis_root=os.path.join(storage_path, 'image'),\n )\n\n return datasets\n" }, { "path": "minigpt4/datasets/data_utils.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport gzip\nimport logging\nimport os\nimport random as rnd\nimport tarfile\nimport zipfile\nimport random\nfrom typing import List\nfrom tqdm import tqdm\n\nimport decord\nfrom decord import VideoReader\nimport webdataset as wds\nimport numpy as np\nimport torch\nfrom torch.utils.data.dataset import IterableDataset\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.datasets.datasets.base_dataset import ConcatDataset\n\n\ndecord.bridge.set_bridge(\"torch\")\nMAX_INT = registry.get(\"MAX_INT\")\n\n\nclass ChainDataset(wds.DataPipeline):\n r\"\"\"Dataset for chaining multiple :class:`DataPipeline` s.\n\n This class is useful to assemble different existing dataset streams. The\n chaining operation is done on-the-fly, so concatenating large-scale\n datasets with this class will be efficient.\n\n Args:\n datasets (iterable of IterableDataset): datasets to be chained together\n \"\"\"\n def __init__(self, datasets: List[wds.DataPipeline]) -> None:\n super().__init__()\n self.datasets = datasets\n self.prob = []\n self.names = []\n for dataset in self.datasets:\n if hasattr(dataset, 'name'):\n self.names.append(dataset.name)\n else:\n self.names.append('Unknown')\n if hasattr(dataset, 'sample_ratio'):\n self.prob.append(dataset.sample_ratio)\n else:\n self.prob.append(1)\n logging.info(\"One of the datapipeline doesn't define ratio and set to 1 automatically.\")\n\n def __iter__(self):\n datastreams = [iter(dataset) for dataset in self.datasets]\n while True:\n select_datastream = random.choices(datastreams, weights=self.prob, k=1)[0]\n yield next(select_datastream)\n\n\ndef apply_to_sample(f, sample):\n if len(sample) == 0:\n return {}\n\n def _apply(x):\n if torch.is_tensor(x):\n return f(x)\n elif isinstance(x, dict):\n return {key: _apply(value) for key, value in x.items()}\n elif isinstance(x, list):\n return [_apply(x) for x in x]\n else:\n return x\n\n return _apply(sample)\n\n\ndef move_to_cuda(sample):\n def _move_to_cuda(tensor):\n return tensor.cuda()\n\n return apply_to_sample(_move_to_cuda, sample)\n\n\ndef prepare_sample(samples, cuda_enabled=True):\n if cuda_enabled:\n samples = move_to_cuda(samples)\n\n # TODO fp16 support\n\n return samples\n\n\ndef reorg_datasets_by_split(datasets, batch_sizes):\n \"\"\"\n Organizes datasets by split.\n\n Args:\n datasets: dict of torch.utils.data.Dataset objects by name.\n\n Returns:\n Dict of datasets by split {split_name: List[Datasets]}.\n \"\"\"\n # if len(datasets) == 1:\n # return datasets[list(datasets.keys())[0]]\n # else:\n reorg_datasets = dict()\n reorg_batch_sizes = dict()\n\n # reorganize by split\n for dataset_name, dataset in datasets.items():\n for split_name, dataset_split in dataset.items():\n if split_name not in reorg_datasets:\n reorg_datasets[split_name] = [dataset_split]\n reorg_batch_sizes[split_name] = [batch_sizes[dataset_name]]\n else:\n reorg_datasets[split_name].append(dataset_split)\n reorg_batch_sizes[split_name].append(batch_sizes[dataset_name])\n\n return reorg_datasets, reorg_batch_sizes\n\n\ndef concat_datasets(datasets):\n \"\"\"\n Concatenates multiple datasets into a single dataset.\n\n It supports may-style datasets and DataPipeline from WebDataset. Currently, does not support\n generic IterableDataset because it requires creating separate samplers.\n\n Now only supports conctenating training datasets and assuming validation and testing\n have only a single dataset. This is because metrics should not be computed on the concatenated\n datasets.\n\n Args:\n datasets: dict of torch.utils.data.Dataset objects by split.\n\n Returns:\n Dict of concatenated datasets by split, \"train\" is the concatenation of multiple datasets,\n \"val\" and \"test\" remain the same.\n\n If the input training datasets contain both map-style and DataPipeline datasets, returns\n a tuple, where the first element is a concatenated map-style dataset and the second\n element is a chained DataPipeline dataset.\n\n \"\"\"\n # concatenate datasets in the same split\n for split_name in datasets:\n if split_name != \"train\":\n assert (\n len(datasets[split_name]) == 1\n ), \"Do not support multiple {} datasets.\".format(split_name)\n datasets[split_name] = datasets[split_name][0]\n else:\n iterable_datasets, map_datasets = [], []\n for dataset in datasets[split_name]:\n if isinstance(dataset, wds.DataPipeline):\n logging.info(\n \"Dataset {} is IterableDataset, can't be concatenated.\".format(\n dataset\n )\n )\n iterable_datasets.append(dataset)\n elif isinstance(dataset, IterableDataset):\n raise NotImplementedError(\n \"Do not support concatenation of generic IterableDataset.\"\n )\n else:\n map_datasets.append(dataset)\n\n # if len(iterable_datasets) > 0:\n # concatenate map-style datasets and iterable-style datasets separately\n if len(iterable_datasets) > 1:\n chained_datasets = (\n ChainDataset(iterable_datasets)\n )\n elif len(iterable_datasets) == 1:\n chained_datasets = iterable_datasets[0]\n else:\n chained_datasets = None\n\n concat_datasets = (\n ConcatDataset(map_datasets) if len(map_datasets) > 0 else None\n )\n\n train_datasets = concat_datasets, chained_datasets\n train_datasets = tuple([x for x in train_datasets if x is not None])\n train_datasets = (\n train_datasets[0] if len(train_datasets) == 1 else train_datasets\n )\n\n datasets[split_name] = train_datasets\n\n return datasets\n\n" }, { "path": "minigpt4/datasets/datasets/__init__.py", "content": "" }, { "path": "minigpt4/datasets/datasets/aok_vqa_datasets.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nfrom collections import OrderedDict\nimport json\nimport os\nimport random\nimport torch\n\nfrom PIL import Image\n\nfrom minigpt4.datasets.datasets.vqa_datasets import VQADataset #, VQAEvalDataset\n\n\nclass __DisplMixin:\n def displ_item(self, index):\n sample, ann = self.__getitem__(index), self.annotation[index]\n return OrderedDict(\n {\n \"file\": ann[\"image\"],\n \"question\": ann[\"question\"],\n \"question_id\": ann[\"question_id\"],\n \"direct_answers\": \"; \".join(ann[\"direct_answers\"]),\n \"choices\": \"; \".join(ann[\"choices\"]),\n \"correct_choice\": ann[\"choices\"][ann[\"correct_choice_idx\"]],\n \"image\": sample[\"image\"],\n }\n )\n\n\nclass AOKVQADataset(VQADataset, __DisplMixin):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n self.instruction_pool =[\n \"[vqa] {}\",\n \"[vqa] Based on the image, respond to this question with a short answer: {}\"\n ]\n\n exist_annotation = []\n for ann in self.annotation:\n image_path = os.path.join(self.vis_root, ann[\"image\"].split('/')[-1])\n if os.path.exists(image_path):\n exist_annotation.append(ann)\n self.annotation = exist_annotation\n\n def get_data(self, index):\n ann = self.annotation[index]\n\n image_path = os.path.join(self.vis_root, ann[\"image\"].split('/')[-1])\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n question = self.text_processor(ann[\"question\"])\n\n answer_key = \"direct_answers\"\n\n answer_weight = {}\n for answer in ann[answer_key]:\n if answer in answer_weight.keys():\n answer_weight[answer] += 1 / len(ann[answer_key])\n else:\n answer_weight[answer] = 1 / len(ann[answer_key])\n\n answers = list(answer_weight.keys())\n weights = list(answer_weight.values())\n\n answer = random.choices(answers, weights=weights, k=1)[0] # random sample an answer according to weights\n\n return {\n \"image\": image,\n \"question\": question,\n \"answer\": answer,\n }\n\n def __getitem__(self, index):\n data = self.get_data(index)\n question = self.text_processor(data[\"question\"])\n instruction = random.choice(self.instruction_pool).format(question)\n\n instruction = \" {} \".format(instruction)\n answer = self.text_processor(data['answer'])\n\n return {\n \"image\": data['image'],\n \"instruction_input\": instruction,\n \"answer\": answer,\n }\n\n\nclass AOKVQGDataset(AOKVQADataset):\n\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n self.instruction_pool = [\n 'Given the image, generate a question whose answer is: {}',\n 'Based on the image, provide a question with the answer: {}',\n 'Given the visual representation, create a question for which the answer is \"{}\"',\n 'From the image provided, craft a question that leads to the reply: {}',\n 'Considering the picture, come up with a question where the answer is: {}',\n 'Taking the image into account, generate an question that has the answer: {}'\n ]\n\n def __getitem__(self, index):\n data = self.get_data(index)\n instruction = random.choice(self.instruction_pool).format(data['answer'])\n\n return {\n \"image\": data['image'],\n \"instruction_input\": instruction,\n \"answer\": data['question'],\n }\n" }, { "path": "minigpt4/datasets/datasets/base_dataset.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport json\nfrom typing import Iterable\n\nfrom torch.utils.data import Dataset, ConcatDataset\nfrom torch.utils.data.dataloader import default_collate\n\n\n\n\nclass BaseDataset(Dataset):\n def __init__(\n self, vis_processor=None, text_processor=None, vis_root=None, ann_paths=[]\n ):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.annotation = []\n # print(\"ann paths\", ann_paths)\n for ann_path in ann_paths:\n # print(\"ann_path\", ann_path)\n ann = json.load(open(ann_path, \"r\"))\n if isinstance(ann, dict):\n self.annotation.extend(json.load(open(ann_path, \"r\"))['annotations'])\n # self.annotation.extend(json.load(open(ann_path, \"r\")))\n else:\n self.annotation.extend(json.load(open(ann_path, \"r\")))\n \n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self._add_instance_ids()\n\n def __len__(self):\n return len(self.annotation)\n\n def collater(self, samples):\n return default_collate(samples)\n\n def set_processors(self, vis_processor, text_processor):\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n def _add_instance_ids(self, key=\"instance_id\"):\n for idx, ann in enumerate(self.annotation):\n ann[key] = str(idx)\n\n\n\nclass ConcatDataset(ConcatDataset):\n def __init__(self, datasets: Iterable[Dataset]) -> None:\n super().__init__(datasets)\n\n def collater(self, samples):\n # TODO For now only supports datasets with same underlying collater implementations\n\n all_keys = set()\n for s in samples:\n all_keys.update(s)\n\n shared_keys = all_keys\n for s in samples:\n shared_keys = shared_keys & set(s.keys())\n\n samples_shared_keys = []\n for s in samples:\n samples_shared_keys.append({k: s[k] for k in s.keys() if k in shared_keys})\n\n return self.datasets[0].collater(samples_shared_keys)\n" }, { "path": "minigpt4/datasets/datasets/caption_datasets.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport os\nfrom collections import OrderedDict\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom PIL import Image\nimport random\n\n\nclass __DisplMixin:\n def displ_item(self, index):\n sample, ann = self.__getitem__(index), self.annotation[index]\n\n return OrderedDict(\n {\n \"file\": ann[\"image\"],\n \"caption\": ann[\"caption\"],\n \"image\": sample[\"image\"],\n }\n )\n\n\nclass CaptionDataset(BaseDataset, __DisplMixin):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n self.img_ids = {}\n n = 0\n for ann in self.annotation:\n img_id = ann[\"image_id\"]\n if img_id not in self.img_ids.keys():\n self.img_ids[img_id] = n\n n += 1\n\n def __getitem__(self, index):\n\n # TODO this assumes image input, not general enough\n ann = self.annotation[index]\n\n img_file = '{:0>12}.jpg'.format(ann[\"image_id\"])\n image_path = os.path.join(self.vis_root, img_file)\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n caption = self.text_processor(ann[\"caption\"])\n\n return {\n \"image\": image,\n \"text_input\": caption,\n \"image_id\": self.img_ids[ann[\"image_id\"]],\n }\n\n\n\nclass COCOCaptionDataset(BaseDataset, __DisplMixin):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n self.img_ids = {}\n n = 0\n\n self.filter_anntation = []\n \n for ann in self.annotation:\n if \"train\" in ann[\"image\"]:\n self.filter_anntation.append(ann)\n self.annotation = self.filter_anntation\n\n for ann in self.annotation:\n img_id = ann[\"image_id\"]\n if img_id not in self.img_ids.keys():\n self.img_ids[img_id] = n\n n += 1\n\n self.instruction_pool = [\n 'Briefly describe this image.',\n 'Provide a concise depiction of this image.',\n 'Present a short description of this image.',\n 'Summarize this image in a few words.',\n 'A short image caption:',\n 'A short image description:',\n 'A photo of ',\n 'An image that shows ',\n 'Write a short description for the image. ',\n 'Write a description for the photo.',\n 'Provide a description of what is presented in the photo.',\n 'Briefly describe the content of the image.',\n 'Can you briefly explain what you see in the image?',\n 'Could you use a few words to describe what you perceive in the photo?',\n 'Please provide a short depiction of the picture.',\n 'Using language, provide a short account of the image.',\n 'Use a few words to illustrate what is happening in the picture.',\n ]\n def __getitem__(self, index):\n\n # TODO this assumes image input, not general enough\n ann = self.annotation[index]\n\n img_file = ann[\"image\"].split(\"/\")[-1]\n image_path = os.path.join(self.vis_root, img_file)\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n caption = self.text_processor(ann[\"caption\"])\n\n instruction = random.choice(self.instruction_pool)\n instruction = \" [caption] {} \".format(instruction)\n\n return {\n \"image\": image,\n \"answer\": caption,\n \"instruction_input\": instruction,\n }\n\nclass CaptionEvalDataset(BaseDataset, __DisplMixin):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n split (string): val or test\n \"\"\"\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n def __getitem__(self, index):\n\n ann = self.annotation[index]\n\n image_path = os.path.join(self.vis_root, ann[\"image\"])\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n\n return {\n \"image\": image,\n \"image_id\": ann[\"image_id\"],\n \"instance_id\": ann[\"instance_id\"],\n }\n" }, { "path": "minigpt4/datasets/datasets/cc_sbu_dataset.py", "content": "import os\nfrom PIL import Image\nimport webdataset as wds\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\n\nclass CCSBUDataset(BaseDataset):\n def __init__(self, vis_processor, text_processor, location):\n super().__init__(vis_processor=vis_processor, text_processor=text_processor)\n\n self.inner_dataset = wds.DataPipeline(\n wds.ResampledShards(location),\n wds.tarfile_to_samples(handler=wds.warn_and_continue),\n wds.shuffle(1000, handler=wds.warn_and_continue),\n wds.decode(\"pilrgb\", handler=wds.warn_and_continue),\n wds.to_tuple(\"jpg\", \"json\", handler=wds.warn_and_continue),\n wds.map_tuple(self.vis_processor, handler=wds.warn_and_continue),\n wds.map(self.to_dict, handler=wds.warn_and_continue),\n )\n\n def to_dict(self, sample):\n return {\n \"image\": sample[0],\n \"answer\": self.text_processor(sample[1][\"caption\"]),\n }\n\n\nclass CCSBUAlignDataset(CaptionDataset):\n\n def __getitem__(self, index):\n\n # TODO this assumes image input, not general enough\n ann = self.annotation[index]\n\n img_file = '{}.jpg'.format(ann[\"image_id\"])\n image_path = os.path.join(self.vis_root, img_file)\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n caption = ann[\"caption\"]\n\n return {\n \"image\": image,\n \"answer\": caption,\n \"image_id\": self.img_ids[ann[\"image_id\"]],\n }" }, { "path": "minigpt4/datasets/datasets/coco_caption.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport os\nimport json\nimport torch\nimport numpy as np\n\nfrom PIL import Image\nfrom PIL import ImageFile\n\nImageFile.LOAD_TRUNCATED_IMAGES = True\n\nfrom minigpt4.datasets.datasets.caption_datasets import COCOCaptionDataset, CaptionEvalDataset\n\nCOCOCapDataset = COCOCaptionDataset\n\n\n\n\n\nclass COCOCapEvalDataset(CaptionEvalDataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n split (string): val or test\n \"\"\"\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n def __getitem__(self, index):\n ann = self.annotation[index]\n\n image_path = os.path.join(self.vis_root, ann[\"image\"])\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n\n img_id = ann[\"image\"].split(\"/\")[-1].strip(\".jpg\").split(\"_\")[-1]\n\n return {\n \"image\": image,\n \"image_id\": img_id,\n \"instance_id\": ann[\"instance_id\"],\n }\n\n\nclass NoCapsEvalDataset(CaptionEvalDataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n split (string): val or test\n \"\"\"\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n def __getitem__(self, index):\n ann = self.annotation[index]\n\n image_path = os.path.join(self.vis_root, ann[\"image\"])\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n\n img_id = ann[\"img_id\"]\n\n return {\n \"image\": image,\n \"image_id\": img_id,\n \"instance_id\": ann[\"instance_id\"],\n }\n\n\nclass RefCOCOEvalData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n\n def __len__(self):\n return len(self.loaded_data)\n \n def __getitem__(self, idx):\n data = self.loaded_data[idx]\n img_id = data['img_id']\n sent = data['sents']\n image_path = os.path.join(self.root_path, f'{img_id[:27]}.jpg')\n image = Image.open(image_path).convert('RGB')\n image = self.vis_processor(image)\n question = f\"[refer] give me the location of {sent}\"\n return image, question, img_id\n\nclass EvalCaptionData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n ann = dict()\n for item in self.loaded_data:\n image_id = item['image_id']\n ann[image_id] = item['image']\n self.ann = [{'image_id':image_id, 'image': ann[image_id]} for image_id in ann]\n\n def __len__(self):\n return len(self.ann)\n \n def __getitem__(self, idx):\n data = self.ann[idx]\n image_id = data['image_id']\n img_file = data['image'].split('/')[-1]\n image_path = os.path.join(self.root_path, img_file)\n image = Image.open(image_path).convert('RGB')\n \n image = self.vis_processor(image)\n question = f\"[caption] please describe this image?\"\n return image, question, image_id\n" }, { "path": "minigpt4/datasets/datasets/coco_dataset.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport itertools\n\nimport numpy as np\nfrom PIL import Image\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nfrom matplotlib.collections import PatchCollection\nfrom matplotlib.patches import Polygon, Rectangle\nfrom torch.utils.data import Dataset\nimport webdataset as wds\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\n\nclass ReferCOCODataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path, dataset='refcoco', splitBy='unc'):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self.refer = REFER(ann_path, vis_root, dataset, splitBy)\n self.ref_ids = self.refer.getRefIds(split=\"train\")\n\n self.instruction_pool = [\n \"[refer] {}\",\n \"[refer] give me the location of {}\",\n \"[refer] where is {} ?\",\n \"[refer] from this image, tell me the location of {}\",\n \"[refer] the location of {} is\",\n \"[refer] could you tell me the location for {} ?\",\n \"[refer] where can I locate the {} ?\",\n ]\n\n\n def __len__(self):\n return len(self.ref_ids)\n\n def preprocess(self, index):\n ref_id = self.ref_ids[index]\n ref = self.refer.loadRefs(ref_id)[0]\n\n image_file = 'COCO_train2014_{:0>12}.jpg'.format(ref[\"image_id\"])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image_orig_size = image.size\n image = self.vis_processor(image)\n image_new_size = [image.shape[1], image.shape[2]]\n\n image_new_size = [100,100]\n\n sample_sentence = random.choice(ref['sentences'])['raw']\n refer_sentence = self.text_processor(sample_sentence)\n\n\n bbox = self.refer.getRefBox(ref['ref_id'])\n bbox = [\n bbox[0] / image_orig_size[0] * image_new_size[0],\n bbox[1] / image_orig_size[1] * image_new_size[1],\n (bbox[0] + bbox[2]) / image_orig_size[0] * image_new_size[0],\n (bbox[1] + bbox[3]) / image_orig_size[1] * image_new_size[1]\n ]\n bbox = [int(x) for x in bbox]\n bbox = \"{{<{}><{}><{}><{}>}}\".format(*bbox)\n return {\n \"image\": image,\n \"refer_sentence\": refer_sentence,\n \"bbox\": bbox,\n \"image_id\": ref['image_id'],\n }\n\n def __getitem__(self, index):\n data = self.preprocess(index)\n instruction = random.choice(self.instruction_pool).format(data['refer_sentence'])\n\n instruction = \" {} \".format(instruction)\n\n return {\n \"image\": data['image'],\n \"instruction_input\": instruction,\n \"answer\": data['bbox'],\n \"image_id\": data['image_id'],\n }\n\n\nclass InvReferCOCODataset(ReferCOCODataset):\n def __init__(self, *args, **kwargs):\n super(InvReferCOCODataset, self).__init__(*args, **kwargs)\n\n self.instruction_pool = [\n \"[identify] {}\",\n \"[identify] what object is in this location {}\",\n \"[identify] identify the object present at this location {}\",\n \"[identify] what is it in {}\",\n \"[identify] describe this object in {}\",\n \"[identify] this {} is\",\n \"[identify] the object in {} is\",\n ]\n\n def __getitem__(self, index):\n data = self.preprocess(index)\n\n instruction = random.choice(self.instruction_pool).format(data['bbox'])\n\n instruction = \" {} \".format(instruction)\n \n return {\n \"image\": data['image'],\n \"instruction_input\": instruction,\n \"answer\": self.text_processor(data['refer_sentence']),\n \"image_id\": data['image_id'],\n }\n\n\nclass REFER:\n def __init__(self, data_root, vis_root, dataset='refcoco', splitBy='unc'):\n # provide data_root folder which contains refclef, refcoco, refcoco+ and refcocog\n # also provide dataset name and splitBy information\n # e.g., dataset = 'refcoco', splitBy = 'unc'\n dataset = dataset.split('inv')[-1] # inv dataset is stored in the same path as normal dataset\n print('loading dataset %s into memory...' % dataset)\n self.ann_dir = os.path.join(data_root, dataset)\n if dataset in ['refcoco', 'refcoco+', 'refcocog']:\n self.vis_root = vis_root\n elif dataset == 'refclef':\n raise 'No RefClef image data'\n else:\n raise 'No refer dataset is called [%s]' % dataset\n\n # load refs from data/dataset/refs(dataset).json\n tic = time.time()\n ref_file = os.path.join(self.ann_dir, 'refs(' + splitBy + ').p')\n self.data = {}\n self.data['dataset'] = dataset\n self.data['refs'] = pickle.load(open(ref_file, 'rb'))\n\n # load annotations from data/dataset/instances.json\n instances_file = os.path.join(self.ann_dir, 'instances.json')\n instances = json.load(open(instances_file, 'r'))\n self.data['images'] = instances['images']\n self.data['annotations'] = instances['annotations']\n self.data['categories'] = instances['categories']\n\n # create index\n self.createIndex()\n print('DONE (t=%.2fs)' % (time.time() - tic))\n\n def createIndex(self):\n # create sets of mapping\n # 1) Refs: \t \t{ref_id: ref}\n # 2) Anns: \t \t{ann_id: ann}\n # 3) Imgs:\t\t \t{image_id: image}\n # 4) Cats: \t \t{category_id: category_name}\n # 5) Sents: \t{sent_id: sent}\n # 6) imgToRefs: \t{image_id: refs}\n # 7) imgToAnns: \t{image_id: anns}\n # 8) refToAnn: \t{ref_id: ann}\n # 9) annToRef: \t{ann_id: ref}\n # 10) catToRefs: \t{category_id: refs}\n # 11) sentToRef: \t{sent_id: ref}\n # 12) sentToTokens: {sent_id: tokens}\n print('creating index...')\n # fetch info from instances\n Anns, Imgs, Cats, imgToAnns = {}, {}, {}, {}\n for ann in self.data['annotations']:\n Anns[ann['id']] = ann\n imgToAnns[ann['image_id']] = imgToAnns.get(ann['image_id'], []) + [ann]\n for img in self.data['images']:\n Imgs[img['id']] = img\n for cat in self.data['categories']:\n Cats[cat['id']] = cat['name']\n\n # fetch info from refs\n Refs, imgToRefs, refToAnn, annToRef, catToRefs = {}, {}, {}, {}, {}\n Sents, sentToRef, sentToTokens = {}, {}, {}\n for ref in self.data['refs']:\n # ids\n ref_id = ref['ref_id']\n ann_id = ref['ann_id']\n category_id = ref['category_id']\n image_id = ref['image_id']\n\n # add mapping related to ref\n Refs[ref_id] = ref\n imgToRefs[image_id] = imgToRefs.get(image_id, []) + [ref]\n catToRefs[category_id] = catToRefs.get(category_id, []) + [ref]\n refToAnn[ref_id] = Anns[ann_id]\n annToRef[ann_id] = ref\n\n # add mapping of sent\n for sent in ref['sentences']:\n Sents[sent['sent_id']] = sent\n sentToRef[sent['sent_id']] = ref\n sentToTokens[sent['sent_id']] = sent['tokens']\n\n # create class members\n self.Refs = Refs\n self.Anns = Anns\n self.Imgs = Imgs\n self.Cats = Cats\n self.Sents = Sents\n self.imgToRefs = imgToRefs\n self.imgToAnns = imgToAnns\n self.refToAnn = refToAnn\n self.annToRef = annToRef\n self.catToRefs = catToRefs\n self.sentToRef = sentToRef\n self.sentToTokens = sentToTokens\n print('index created.')\n\n def getRefIds(self, image_ids=[], cat_ids=[], ref_ids=[], split=''):\n image_ids = image_ids if type(image_ids) == list else [image_ids]\n cat_ids = cat_ids if type(cat_ids) == list else [cat_ids]\n ref_ids = ref_ids if type(ref_ids) == list else [ref_ids]\n\n if len(image_ids) == len(cat_ids) == len(ref_ids) == len(split) == 0:\n refs = self.data['refs']\n else:\n if not len(image_ids) == 0:\n refs = [self.imgToRefs[image_id] for image_id in image_ids]\n else:\n refs = self.data['refs']\n if not len(cat_ids) == 0:\n refs = [ref for ref in refs if ref['category_id'] in cat_ids]\n if not len(ref_ids) == 0:\n refs = [ref for ref in refs if ref['ref_id'] in ref_ids]\n if not len(split) == 0:\n if split in ['testA', 'testB', 'testC']:\n refs = [ref for ref in refs if\n split[-1] in ref['split']] # we also consider testAB, testBC, ...\n elif split in ['testAB', 'testBC', 'testAC']:\n refs = [ref for ref in refs if ref['split'] == split] # rarely used I guess...\n elif split == 'test':\n refs = [ref for ref in refs if 'test' in ref['split']]\n elif split == 'train' or split == 'val':\n refs = [ref for ref in refs if ref['split'] == split]\n else:\n raise 'No such split [%s]' % split\n ref_ids = [ref['ref_id'] for ref in refs]\n return ref_ids\n\n def getAnnIds(self, image_ids=[], cat_ids=[], ref_ids=[]):\n image_ids = image_ids if type(image_ids) == list else [image_ids]\n cat_ids = cat_ids if type(cat_ids) == list else [cat_ids]\n ref_ids = ref_ids if type(ref_ids) == list else [ref_ids]\n\n if len(image_ids) == len(cat_ids) == len(ref_ids) == 0:\n ann_ids = [ann['id'] for ann in self.data['annotations']]\n else:\n if not len(image_ids) == 0:\n lists = [self.imgToAnns[image_id] for image_id in image_ids if image_id in self.imgToAnns] # list of [anns]\n anns = list(itertools.chain.from_iterable(lists))\n else:\n anns = self.data['annotations']\n if not len(cat_ids) == 0:\n anns = [ann for ann in anns if ann['category_id'] in cat_ids]\n ann_ids = [ann['id'] for ann in anns]\n if not len(ref_ids) == 0:\n ids = set(ann_ids).intersection(set([self.Refs[ref_id]['ann_id'] for ref_id in ref_ids]))\n return ann_ids\n\n def getImgIds(self, ref_ids=[]):\n ref_ids = ref_ids if type(ref_ids) == list else [ref_ids]\n\n if not len(ref_ids) == 0:\n image_ids = list(set([self.Refs[ref_id]['image_id'] for ref_id in ref_ids]))\n else:\n image_ids = self.Imgs.keys()\n return image_ids\n\n def getCatIds(self):\n return self.Cats.keys()\n\n def loadRefs(self, ref_ids=[]):\n if type(ref_ids) == list:\n return [self.Refs[ref_id] for ref_id in ref_ids]\n elif type(ref_ids) == int:\n return [self.Refs[ref_ids]]\n\n def loadAnns(self, ann_ids=[]):\n if type(ann_ids) == list:\n return [self.Anns[ann_id] for ann_id in ann_ids]\n elif type(ann_ids) == int:\n return [self.Anns[ann_ids]]\n\n def loadImgs(self, image_ids=[]):\n if type(image_ids) == list:\n return [self.Imgs[image_id] for image_id in image_ids]\n elif type(image_ids) == int:\n return [self.Imgs[image_ids]]\n\n def loadCats(self, cat_ids=[]):\n if type(cat_ids) == list:\n return [self.Cats[cat_id] for cat_id in cat_ids]\n elif type(cat_ids) == int:\n return [self.Cats[cat_ids]]\n\n def getRefBox(self, ref_id):\n ref = self.Refs[ref_id]\n ann = self.refToAnn[ref_id]\n return ann['bbox'] # [x, y, w, h]\n\n def showRef(self, ref, seg_box='box'):\n ax = plt.gca()\n # show image\n image = self.Imgs[ref['image_id']]\n I = io.imread(os.path.join(self.vis_root, image['file_name']))\n ax.imshow(I)\n # show refer expression\n for sid, sent in enumerate(ref['sentences']):\n print('%s. %s' % (sid + 1, sent['sent']))\n # show segmentations\n if seg_box == 'seg':\n ann_id = ref['ann_id']\n ann = self.Anns[ann_id]\n polygons = []\n color = []\n c = 'none'\n if type(ann['segmentation'][0]) == list:\n # polygon used for refcoco*\n for seg in ann['segmentation']:\n poly = np.array(seg).reshape((len(seg) / 2, 2))\n polygons.append(Polygon(poly, True, alpha=0.4))\n color.append(c)\n p = PatchCollection(polygons, facecolors=color, edgecolors=(1, 1, 0, 0), linewidths=3, alpha=1)\n ax.add_collection(p) # thick yellow polygon\n p = PatchCollection(polygons, facecolors=color, edgecolors=(1, 0, 0, 0), linewidths=1, alpha=1)\n ax.add_collection(p) # thin red polygon\n else:\n # mask used for refclef\n raise NotImplementedError('RefClef is not downloaded')\n # show bounding-box\n elif seg_box == 'box':\n ann_id = ref['ann_id']\n ann = self.Anns[ann_id]\n bbox = self.getRefBox(ref['ref_id'])\n box_plot = Rectangle((bbox[0], bbox[1]), bbox[2], bbox[3], fill=False, edgecolor='green', linewidth=3)\n ax.add_patch(box_plot)\n" }, { "path": "minigpt4/datasets/datasets/coco_vqa_datasets.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport os\nimport json\nimport random\n\nfrom PIL import Image\n\nfrom minigpt4.datasets.datasets.vqa_datasets import VQADataset, VQAEvalDataset\n\nfrom collections import OrderedDict\n\n\nclass __DisplMixin:\n def displ_item(self, index):\n sample, ann = self.__getitem__(index), self.annotation[index]\n\n return OrderedDict(\n {\n \"file\": ann[\"image\"],\n \"question\": ann[\"question\"],\n \"question_id\": ann[\"question_id\"],\n \"answers\": \"; \".join(ann[\"answer\"]),\n \"image\": sample[\"image\"],\n }\n )\n\n\nclass COCOVQADataset(VQADataset, __DisplMixin):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n self.instruction_pool =[\n \"[vqa] {}\",\n \"[vqa] Based on the image, respond to this question with a short answer: {}\"\n ]\n\n exist_annotation = []\n for ann in self.annotation:\n image_path = os.path.join(self.vis_root, ann[\"image\"].split('/')[-1])\n if os.path.exists(image_path):\n exist_annotation.append(ann)\n self.annotation = exist_annotation\n\n\n def get_data(self, index):\n ann = self.annotation[index]\n\n image_path = os.path.join(self.vis_root, ann[\"image\"].split('/')[-1])\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n question = self.text_processor(ann[\"question\"])\n question_id = ann[\"question_id\"]\n\n answer_weight = {}\n for answer in ann[\"answer\"]:\n if answer in answer_weight.keys():\n answer_weight[answer] += 1 / len(ann[\"answer\"])\n else:\n answer_weight[answer] = 1 / len(ann[\"answer\"])\n\n answers = list(answer_weight.keys())\n weights = list(answer_weight.values())\n\n answer = random.choices(answers, weights=weights, k=1)[0] # random sample an answer according to weights\n\n\n return {\n \"image\": image,\n \"question\": question,\n \"question_id\": question_id,\n \"answer\": answer,\n }\n\n def __getitem__(self, index):\n data = self.get_data(index)\n instruction = random.choice(self.instruction_pool).format(data['question'])\n instruction = \" {} \".format(instruction)\n\n return {\n \"image\": data['image'],\n \"question_id\": data[\"question_id\"],\n \"instruction_input\": instruction,\n \"answer\": self.text_processor(data['answer']),\n }\n\n\nclass COCOVQAEvalDataset(VQAEvalDataset, __DisplMixin):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n \n self.instruction_pool = [\n 'Question: {} Short answer:',\n ]\n self.vis_root = vis_root\n\n self.annotation = json.load(open(ann_paths[0]))\n\n answer_list_path = ann_paths[1]\n if os.path.exists(answer_list_path):\n self.answer_list = json.load(open(answer_list_path))\n else:\n self.answer_list = None\n\n try:\n self.coco_fmt_qust_file = ann_paths[2]\n self.coco_fmt_anno_file = ann_paths[3]\n except IndexError:\n self.coco_fmt_qust_file = None\n self.coco_fmt_anno_file = None\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self._add_instance_ids()\n\n def __getitem__(self, index):\n ann = self.annotation[index]\n\n image_path = os.path.join(self.vis_root, ann[\"image\"])\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n question = self.text_processor(ann[\"question\"])\n \n instruction = random.choice(self.instruction_pool).format(question)\n instruction = \" {} \".format(instruction)\n \n return {\n \"image\": image,\n 'image_path': image_path,\n \"question\": question,\n \"question_id\": ann[\"question_id\"],\n \"instruction_input\": instruction,\n \"instance_id\": ann[\"instance_id\"],\n }\n" }, { "path": "minigpt4/datasets/datasets/dataloader_utils.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport time\nimport random\nimport torch\nfrom minigpt4.datasets.data_utils import move_to_cuda\nfrom torch.utils.data import DataLoader\n\n\nclass MultiIterLoader:\n \"\"\"\n A simple wrapper for iterating over multiple iterators.\n\n Args:\n loaders (List[Loader]): List of Iterator loaders.\n ratios (List[float]): List of ratios to sample from each loader. If None, all loaders are sampled uniformly.\n \"\"\"\n\n def __init__(self, loaders, ratios=None):\n # assert all loaders has __next__ method\n for loader in loaders:\n assert hasattr(\n loader, \"__next__\"\n ), \"Loader {} has no __next__ method.\".format(loader)\n\n if ratios is None:\n ratios = [1.0] * len(loaders)\n else:\n assert len(ratios) == len(loaders)\n ratios = [float(ratio) / sum(ratios) for ratio in ratios]\n\n self.loaders = loaders\n self.ratios = ratios\n\n def __next__(self):\n # random sample from each loader by ratio\n loader_idx = random.choices(range(len(self.loaders)), self.ratios, k=1)[0]\n return next(self.loaders[loader_idx])\n\n\nclass PrefetchLoader(object):\n \"\"\"\n Modified from https://github.com/ChenRocks/UNITER.\n\n overlap compute and cuda data transfer\n (copied and then modified from nvidia apex)\n \"\"\"\n\n def __init__(self, loader):\n self.loader = loader\n self.stream = torch.cuda.Stream()\n\n def __iter__(self):\n loader_it = iter(self.loader)\n self.preload(loader_it)\n batch = self.next(loader_it)\n while batch is not None:\n is_tuple = isinstance(batch, tuple)\n if is_tuple:\n task, batch = batch\n\n if is_tuple:\n yield task, batch\n else:\n yield batch\n batch = self.next(loader_it)\n\n def __len__(self):\n return len(self.loader)\n\n def preload(self, it):\n try:\n self.batch = next(it)\n except StopIteration:\n self.batch = None\n return\n # if record_stream() doesn't work, another option is to make sure\n # device inputs are created on the main stream.\n # self.next_input_gpu = torch.empty_like(self.next_input,\n # device='cuda')\n # self.next_target_gpu = torch.empty_like(self.next_target,\n # device='cuda')\n # Need to make sure the memory allocated for next_* is not still in use\n # by the main stream at the time we start copying to next_*:\n # self.stream.wait_stream(torch.cuda.current_stream())\n with torch.cuda.stream(self.stream):\n self.batch = move_to_cuda(self.batch)\n # more code for the alternative if record_stream() doesn't work:\n # copy_ will record the use of the pinned source tensor in this\n # side stream.\n # self.next_input_gpu.copy_(self.next_input, non_blocking=True)\n # self.next_target_gpu.copy_(self.next_target, non_blocking=True)\n # self.next_input = self.next_input_gpu\n # self.next_target = self.next_target_gpu\n\n def next(self, it):\n torch.cuda.current_stream().wait_stream(self.stream)\n batch = self.batch\n if batch is not None:\n record_cuda_stream(batch)\n self.preload(it)\n return batch\n\n def __getattr__(self, name):\n method = self.loader.__getattribute__(name)\n return method\n\n\ndef record_cuda_stream(batch):\n if isinstance(batch, torch.Tensor):\n batch.record_stream(torch.cuda.current_stream())\n elif isinstance(batch, list) or isinstance(batch, tuple):\n for t in batch:\n record_cuda_stream(t)\n elif isinstance(batch, dict):\n for t in batch.values():\n record_cuda_stream(t)\n else:\n pass\n\n\nclass IterLoader:\n \"\"\"\n A wrapper to convert DataLoader as an infinite iterator.\n\n Modified from:\n https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/iter_based_runner.py\n \"\"\"\n\n def __init__(self, dataloader: DataLoader, use_distributed: bool = False):\n self._dataloader = dataloader\n self.iter_loader = iter(self._dataloader)\n self._use_distributed = use_distributed\n self._epoch = 0\n\n @property\n def epoch(self) -> int:\n return self._epoch\n\n def __next__(self):\n try:\n data = next(self.iter_loader)\n except StopIteration:\n self._epoch += 1\n if hasattr(self._dataloader.sampler, \"set_epoch\") and self._use_distributed:\n self._dataloader.sampler.set_epoch(self._epoch)\n time.sleep(2) # Prevent possible deadlock during epoch transition\n self.iter_loader = iter(self._dataloader)\n data = next(self.iter_loader)\n\n return data\n\n def __iter__(self):\n return self\n\n def __len__(self):\n return len(self._dataloader)\n" }, { "path": "minigpt4/datasets/datasets/flickr.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport itertools\n\nimport numpy as np\nfrom PIL import Image\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nfrom matplotlib.collections import PatchCollection\nfrom matplotlib.patches import Polygon, Rectangle\nfrom torch.utils.data import Dataset\nimport webdataset as wds\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\n\nclass GroundedDetailDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self.instruction_pool = [\n '[grounding] please describe this image in details',\n '[grounding] describe this image as detailed as possible',\n '[grounding] summarize this image in details',\n '[grounding] give a thorough description of what you see in this image',\n ]\n\n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index]\n\n # image_file = 'COCO_train2014_{}.jpg'.format(info['image_id'])\n image_file = '{}.jpg'.format(info['image_id'])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n answer = info['grounded_caption']\n instruction = random.choice(self.instruction_pool)\n instruction = \" {} \".format(instruction)\n\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": answer,\n \"image_id\": info['image_id'],\n }\n\n\n\n\nclass CaptionToObjectDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self.instruction_pool = [\n '[detection] {}',\n ]\n\n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index]\n\n image_file = '{}.jpg'.format(info['image_id'])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n input = info[\"caption\"]\n answer = info[\"output\"]\n\n instruction = random.choice(self.instruction_pool).format(input)\n\n instruction = \" {} \".format(instruction)\n\n print(\"CaptionToObject instruction\", instruction)\n print(\"CaptionToObject answer\", answer)\n\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": answer,\n \"image_id\": info['image_id'],\n }\n\n\n\n\nclass PhraseToObjectDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self.instruction_pool = [\n '[detection] {}',\n ]\n\n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index]\n image_file = '{}.jpg'.format(info['image_id'])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n input = info[\"phrase\"]\n answer = \"

\"+input+\"

\"+info[\"bbox\"]\n instruction = random.choice(self.instruction_pool).format(input)\n\n instruction = \" {} \".format(instruction)\n\n print(\"PhraseToObject instruction\", instruction)\n print(\"PhraseToObject answer\", answer)\n\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": answer,\n \"image_id\": info['image_id'],\n }\n" }, { "path": "minigpt4/datasets/datasets/gqa_datasets.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport os\nimport json\n\nfrom PIL import Image\n\nfrom minigpt4.datasets.datasets.vqa_datasets import VQADataset\n\nfrom collections import OrderedDict\nimport random\n\nclass __DisplMixin:\n def displ_item(self, index):\n sample, ann = self.__getitem__(index), self.annotation[index]\n\n return OrderedDict(\n {\n \"file\": ann[\"image\"],\n \"question\": ann[\"question\"],\n \"question_id\": ann[\"question_id\"],\n \"answers\": \"; \".join(ann[\"answer\"]),\n \"image\": sample[\"image\"],\n }\n )\n\n\nclass GQADataset(VQADataset, __DisplMixin):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n self.instruction_pool =[\n \"[vqa] {}\",\n \"[vqa] Based on the image, respond to this question with a short answer: {}\"\n ]\n\n def __getitem__(self, index):\n ann = self.annotation[index]\n\n image_path = os.path.join(self.vis_root, ann[\"image\"])\n image = Image.open(image_path).convert(\"RGB\")\n\n image = self.vis_processor(image)\n question = self.text_processor(ann[\"question\"])\n\n instruction = random.choice(self.instruction_pool).format(question)\n instruction = \" {} \".format(instruction)\n\n answers = self.text_processor(ann[\"answer\"])\n\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": answers,\n }\n\n" }, { "path": "minigpt4/datasets/datasets/laion_dataset.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport webdataset as wds\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\n\n\nclass LaionDataset(BaseDataset):\n def __init__(self, vis_processor, text_processor, location):\n super().__init__(vis_processor=vis_processor, text_processor=text_processor)\n\n self.inner_dataset = wds.DataPipeline(\n wds.ResampledShards(location),\n wds.tarfile_to_samples(handler=wds.warn_and_continue),\n wds.shuffle(1000, handler=wds.warn_and_continue),\n wds.decode(\"pilrgb\", handler=wds.warn_and_continue),\n wds.to_tuple(\"jpg\", \"json\", handler=wds.warn_and_continue),\n wds.map_tuple(self.vis_processor, handler=wds.warn_and_continue),\n wds.map(self.to_dict, handler=wds.warn_and_continue),\n )\n\n def to_dict(self, sample):\n return {\n \"image\": sample[0],\n \"answer\": self.text_processor(sample[1][\"caption\"]),\n }\n\n" }, { "path": "minigpt4/datasets/datasets/llava_dataset.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport numpy as np\nfrom PIL import Image\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nfrom matplotlib.collections import PatchCollection\nfrom matplotlib.patches import Polygon, Rectangle\nfrom torch.utils.data import Dataset\nimport webdataset as wds\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\nclass LlavaDetailDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index]\n\n image_file = 'COCO_train2014_{}.jpg'.format(info['id'])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n answer = info['conversations'][1]['value']\n instruction = info['conversations'][0]['value'].replace('', '').replace('\\n', '').strip()\n \n instruction = ' {} '.format(self.text_processor(instruction))\n\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": answer,\n \"image_id\": info['id'],\n }\n\nclass LlavaReasonDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index]\n\n image_file = 'COCO_train2014_{}.jpg'.format(info['id'])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n answer = info['conversations'][1]['value']\n instruction = info['conversations'][0]['value'].replace('', '').replace('\\n', '').strip()\n\n instruction = ' {} '.format(self.text_processor(instruction))\n\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": answer,\n \"image_id\": info['id'],\n }\n\n\n\n\nclass LlavaConversationDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self.ann=[]\n\n \n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n self.connect_sym = \"!@#\"\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index]\n\n image_file = 'COCO_train2014_{}.jpg'.format(info['id'])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n first_instruction = info['conversations'][0]['value'].replace('', '').replace('\\n', '').strip()\n first_instruction = ' {} '.format(first_instruction)\n\n questions = [first_instruction]\n answers = []\n\n for i, item in enumerate(info[\"conversations\"][1:]):\n if i % 2 ==0: # assistant\n assistant_answer = item[\"value\"]\n answers.append(assistant_answer)\n else:\n human_instruction = item[\"value\"]+\" \"\n questions.append(human_instruction)\n\n questions = self.connect_sym.join(questions)\n answers = self.connect_sym.join(answers)\n\n\n return {\n \"image\": image,\n \"conv_q\": questions,\n 'conv_a': answers,\n \"image_id\": info['id'],\n \"connect_sym\": self.connect_sym\n }" }, { "path": "minigpt4/datasets/datasets/multitask_conversation.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport itertools\n\nimport numpy as np\nfrom PIL import Image\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nfrom matplotlib.collections import PatchCollection\nfrom matplotlib.patches import Polygon, Rectangle\nfrom torch.utils.data import Dataset\nimport webdataset as wds\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\n\n\n\nclass MultiTaskConversationDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n\n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n self.connect_sym = \"!@#\"\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index]\n\n image_file = 'COCO_train2014_{}.jpg'.format(info['id'])\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n first_instruction = info['conversations'][0]['value'].replace('', '').replace('\\n', '').strip()\n first_instruction = ' {} '.format(first_instruction)\n\n questions = [first_instruction]\n answers = []\n\n for i, item in enumerate(info[\"conversations\"][1:]):\n if i % 2 ==0: # assistant\n assistant_answer = item[\"value\"]\n answers.append(assistant_answer)\n else:\n human_instruction = item[\"value\"]+\" \"\n questions.append(human_instruction)\n\n questions = self.connect_sym.join(questions)\n answers = self.connect_sym.join(answers)\n\n\n return {\n \"image\": image,\n \"conv_q\": questions,\n 'conv_a': answers,\n \"image_id\": info['id'],\n \"connect_sym\": self.connect_sym\n }" }, { "path": "minigpt4/datasets/datasets/ocrvqa_dataset.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport itertools\n\nimport numpy as np\nfrom PIL import Image\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nfrom matplotlib.collections import PatchCollection\nfrom matplotlib.patches import Polygon, Rectangle\nfrom torch.utils.data import Dataset\nimport webdataset as wds\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\n\nclass OCRVQADataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n self.data = self.create_data(ann_path)\n\n self.instruction_pool =[\n \"[vqa] {}\",\n \"[vqa] Based on the image, respond to this question with a short answer: {}\"\n ]\n\n def create_data(self, ann_path):\n processed_data = []\n with open(ann_path, 'r') as f:\n data = json.load(f)\n for k in data.keys():\n if data[k]['split'] != 1: continue # 1 for training, 2 for validation, 3 for test\n ext = os.path.splitext(data[k]['imageURL'])[1]\n imageFile = k + ext\n assert len(data[k]['questions']) == len(data[k]['answers'])\n for q, a in zip(data[k]['questions'], data[k]['answers']):\n processed_data.append(\n {'question': q,\n 'answer': a,\n 'image_path': imageFile,\n 'image_id': k,\n 'title': data[k]['title'],\n 'genre': data[k]['genre'],\n }\n )\n return processed_data\n\n def __len__(self):\n return len(self.data)\n\n def __getitem__(self, index):\n sample = self.data[index]\n image = Image.open(os.path.join(self.vis_root, sample['image_path'])).convert(\"RGB\")\n image = self.vis_processor(image)\n question = self.text_processor(sample[\"question\"])\n answer = self.text_processor(sample[\"answer\"])\n\n instruction = random.choice(self.instruction_pool).format(question)\n instruction = \" {} \".format(instruction)\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": answer,\n \"image_id\": sample['image_id']\n }\n \n" }, { "path": "minigpt4/datasets/datasets/text_caps.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport itertools\n\nimport numpy as np\nfrom PIL import Image\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nfrom matplotlib.collections import PatchCollection\nfrom matplotlib.patches import Polygon, Rectangle\nfrom torch.utils.data import Dataset\nimport webdataset as wds\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\n\n\nclass TextCapDataset(Dataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.vis_root = vis_root\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n self.instruction_pool = [\n 'Briefly describe this image.',\n 'Provide a concise depiction of this image.',\n 'Present a short description of this image.',\n 'Summarize this image in a few words.',\n 'A short image caption:',\n 'A short image description:',\n 'A photo of ',\n 'An image that shows ',\n 'Write a short description for the image. ',\n 'Write a description for the photo.',\n 'Provide a description of what is presented in the photo.',\n 'Briefly describe the content of the image.',\n 'Can you briefly explain what you see in the image?',\n 'Could you use a few words to describe what you perceive in the photo?',\n 'Please provide a short depiction of the picture.',\n 'Using language, provide a short account of the image.',\n 'Use a few words to illustrate what is happening in the picture.',\n ]\n \n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n\n def __len__(self):\n return len(self.ann[\"data\"])\n\n\n def __getitem__(self, index):\n info = self.ann[\"data\"][index]\n\n image_file = '{}.jpg'.format(info['image_id'])\n\n image_path = os.path.join(self.vis_root, image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n\n caption = info[\"caption_str\"]\n caption = self.text_processor(caption)\n instruction = \" [caption] {} \".format(random.choice(self.instruction_pool))\n return {\n \"image\": image,\n \"instruction_input\": instruction,\n \"answer\": caption,\n }\n" }, { "path": "minigpt4/datasets/datasets/unnatural_instruction.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport itertools\n\nimport numpy as np\nfrom PIL import Image\nimport skimage.io as io\nimport matplotlib.pyplot as plt\nfrom matplotlib.collections import PatchCollection\nfrom matplotlib.patches import Polygon, Rectangle\nfrom torch.utils.data import Dataset\nimport webdataset as wds\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\nfrom minigpt4.datasets.datasets.caption_datasets import CaptionDataset\n\n\nclass UnnaturalDataset(Dataset):\n def __init__(self, text_processor, ann_path):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.text_processor = text_processor\n\n with open(ann_path, 'r') as f:\n self.ann = json.load(f)\n\n def __len__(self):\n return len(self.ann)\n\n def __getitem__(self, index):\n info = self.ann[index][\"instances\"][0]\n instruction = info[\"instruction_with_input\"]\n constraints = info[\"constraints\"]\n answer = info[\"output\"]\n if constraints != None:\n instruction = instruction+\" \"+constraints\n\n return {\n \"instruction_input\": self.text_processor(instruction),\n \"answer\": self.text_processor(answer),\n }\n" }, { "path": "minigpt4/datasets/datasets/vg_dataset.py", "content": "import os\nimport json\nimport pickle\nimport random\nimport time\nimport itertools\n\nimport numpy as np\nfrom PIL import Image\nfrom torch.utils.data import Dataset\nfrom visual_genome import local\n\n\n\n\nclass ReferVisualGenomeDataset(Dataset):\n def __init__(self, vis_processor, text_processor, data_dir):\n \"\"\"\n vis_root (string): Root directory of images (e.g. coco/images/)\n ann_root (string): directory to store the annotation file\n \"\"\"\n self.data_dir = data_dir\n\n self.vis_processor = vis_processor\n self.text_processor = text_processor\n\n all_regions = local.get_all_region_descriptions(self.data_dir)\n all_regions = [region for regions in all_regions for region in regions]\n\n # follow OFA practice, only regions smaller than 16384 pixels are used for refer\n self.regions = [region for region in all_regions if region.width * region.height < 16384]\n\n\n self.instruction_pool = [\n \"[refer] {}\",\n \"[refer] give me the location of {}\",\n \"[refer] where is {} ?\",\n \"[refer] from this image, tell me the location of {}\",\n \"[refer] the location of {} is\",\n \"[refer] could you tell me the location for {} ?\",\n \"[refer] where can I locate the {} ?\",\n ]\n\n\n def __len__(self):\n return len(self.regions)\n\n def preprocess(self, index):\n region = self.regions[index]\n image_file = region.image.url.split('/')[-2:]\n image_path = os.path.join(self.data_dir, *image_file)\n image = Image.open(image_path).convert(\"RGB\")\n image_orig_size = image.size\n image = self.vis_processor(image)\n image_new_size = [100,100]\n\n sample_sentence = region.phrase\n refer_sentence = self.text_processor(sample_sentence)\n\n bbox = [region.x, region.y, region.width, region.height]\n\n bbox = [\n bbox[0] / image_orig_size[0] * image_new_size[0],\n bbox[1] / image_orig_size[1] * image_new_size[1],\n (bbox[0] + bbox[2]) / image_orig_size[0] * image_new_size[0],\n (bbox[1] + bbox[3]) / image_orig_size[1] * image_new_size[1]\n ]\n bbox = [int(x) for x in bbox]\n bbox = \"{{<{}><{}><{}><{}>}}\".format(*bbox)\n return {\n \"image\": image,\n \"refer_sentence\": refer_sentence,\n \"bbox\": bbox,\n \"image_id\": region.image.id,\n }\n\n def __getitem__(self, index):\n data = self.preprocess(index)\n instruction = random.choice(self.instruction_pool).format(data['refer_sentence'])\n\n instruction = \" {} \".format(instruction)\n\n return {\n \"image\": data['image'],\n \"instruction_input\": instruction,\n \"answer\": data['bbox'],\n \"image_id\": data['image_id'],\n }\n\n\n" }, { "path": "minigpt4/datasets/datasets/vqa_datasets.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport torch\nfrom PIL import Image\nimport os\n\nfrom minigpt4.datasets.datasets.base_dataset import BaseDataset\n\n\nclass VQADataset(BaseDataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n\nclass VQAEvalDataset(BaseDataset):\n def __init__(self, vis_processor, text_processor, vis_root, ann_paths):\n super().__init__(vis_processor, text_processor, vis_root, ann_paths)\n\n\nclass OKVQAEvalData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n\n def __len__(self):\n return len(self.loaded_data)\n \n def __getitem__(self, idx):\n data = self.loaded_data[idx]\n img_id = data['image_id']\n question = data['question']\n question_id = data['question_id']\n img_file = '{:0>12}.jpg'.format(img_id)\n image_path = os.path.join(self.root_path, img_file)\n image = Image.open(image_path).convert('RGB')\n image = self.vis_processor(image)\n question = f\"[vqa] Based on the image, respond to this question with a short answer: {question}\"\n return image, question, question_id, img_id\n\nclass VizWizEvalData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n\n def __len__(self):\n return len(self.loaded_data)\n \n def __getitem__(self, idx):\n data = self.loaded_data[idx]\n img_id = data['image']\n question = data['question']\n answers = data['answers']\n answers = '_'.join([answer['answer'] for answer in answers])\n image_path = os.path.join(self.root_path, img_id)\n image = Image.open(image_path).convert('RGB')\n image = self.vis_processor(image)\n question = f\"[vqa] The question is '{question}' Based on the image, answer the question with a single word or phrase. and reply 'unanswerable' when the provided information is insufficient\"\n return image, question, answers\n\nclass IconQAEvalData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n\n def __len__(self):\n return len(self.loaded_data)\n \n def __getitem__(self, idx):\n data = self.loaded_data[idx]\n image_id = data['image_id']\n question = data['question']\n image_path = os.path.join(self.root_path, image_id, 'image.png')\n image = Image.open(image_path).convert('RGB')\n image = self.vis_processor(image).half().cuda()\n candidates = '_'.join(data['choices'])\n answer = data['answer']\n question = f\"[vqa] Based on the image, respond to this question with a short answer: {question}\"\n return image, question, candidates, answer\n\nclass GQAEvalData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n\n def __len__(self):\n return len(self.loaded_data)\n \n def __getitem__(self, idx):\n ann = self.loaded_data[idx]\n image_id = ann[\"image\"]\n image_path = os.path.join(self.root_path, f\"{image_id}\")\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n question = ann[\"question\"]\n question = f\"[vqa] Based on the image, respond to this question with a short answer: {question}\"\n labels = ann[\"answer\"]\n\n return image, question, labels\n\nclass HMEvalData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n\n def __len__(self):\n return len(self.loaded_data)\n \n def __getitem__(self, idx):\n ann = self.loaded_data[idx]\n image_id = ann[\"img\"]\n image_path = os.path.join(self.root_path, f\"{image_id}\")\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n question = ann[\"text\"]\n question = f\"This is an image writting '{question}'. Is this image hateful? Answer yes or no. Answer:\"\n labels = ann[\"label\"]\n\n return image, question, labels\n\nclass VSREvalData(torch.utils.data.Dataset):\n def __init__(self, loaded_data, vis_processor, root_path):\n self.loaded_data = loaded_data\n self.root_path = root_path\n self.vis_processor = vis_processor\n\n def __len__(self):\n return len(self.loaded_data)\n \n def __getitem__(self, idx):\n ann = self.loaded_data[idx]\n image_path = os.path.join(self.root_path, ann[\"image\"])\n image = Image.open(image_path).convert(\"RGB\")\n image = self.vis_processor(image)\n question = ann[\"caption\"]\n question = f'[vqa] Based on the image, is this statement true or false? {question}'\n labels = 'true' if ann[\"label\"] == 1 else 'false'\n\n return image, question, labels" }, { "path": "minigpt4/models/Qformer.py", "content": "\"\"\"\n * Copyright (c) 2023, salesforce.com, inc.\n * All rights reserved.\n * SPDX-License-Identifier: BSD-3-Clause\n * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n * By Junnan Li\n * Based on huggingface code base\n * https://github.com/huggingface/transformers/blob/v4.15.0/src/transformers/models/bert\n\"\"\"\n\nimport math\nimport os\nimport warnings\nfrom dataclasses import dataclass\nfrom typing import Optional, Tuple, Dict, Any\n\nimport torch\nfrom torch import Tensor, device, dtype, nn\nimport torch.utils.checkpoint\nfrom torch import nn\nfrom torch.nn import CrossEntropyLoss\nimport torch.nn.functional as F\n\nfrom transformers.activations import ACT2FN\nfrom transformers.file_utils import (\n ModelOutput,\n)\nfrom transformers.modeling_outputs import (\n BaseModelOutputWithPastAndCrossAttentions,\n BaseModelOutputWithPoolingAndCrossAttentions,\n CausalLMOutputWithCrossAttentions,\n MaskedLMOutput,\n MultipleChoiceModelOutput,\n NextSentencePredictorOutput,\n QuestionAnsweringModelOutput,\n SequenceClassifierOutput,\n TokenClassifierOutput,\n)\nfrom transformers.modeling_utils import (\n PreTrainedModel,\n apply_chunking_to_forward,\n find_pruneable_heads_and_indices,\n prune_linear_layer,\n)\nfrom transformers.utils import logging\nfrom transformers.models.bert.configuration_bert import BertConfig\n\nlogger = logging.get_logger(__name__)\n\n\nclass BertEmbeddings(nn.Module):\n \"\"\"Construct the embeddings from word and position embeddings.\"\"\"\n\n def __init__(self, config):\n super().__init__()\n self.word_embeddings = nn.Embedding(\n config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id\n )\n self.position_embeddings = nn.Embedding(\n config.max_position_embeddings, config.hidden_size\n )\n\n # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load\n # any TensorFlow checkpoint file\n self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)\n self.dropout = nn.Dropout(config.hidden_dropout_prob)\n\n # position_ids (1, len position emb) is contiguous in memory and exported when serialized\n self.register_buffer(\n \"position_ids\", torch.arange(config.max_position_embeddings).expand((1, -1))\n )\n self.position_embedding_type = getattr(\n config, \"position_embedding_type\", \"absolute\"\n )\n\n self.config = config\n\n def forward(\n self,\n input_ids=None,\n position_ids=None,\n query_embeds=None,\n past_key_values_length=0,\n ):\n if input_ids is not None:\n seq_length = input_ids.size()[1]\n else:\n seq_length = 0\n\n if position_ids is None:\n position_ids = self.position_ids[\n :, past_key_values_length : seq_length + past_key_values_length\n ].clone()\n\n if input_ids is not None:\n embeddings = self.word_embeddings(input_ids)\n if self.position_embedding_type == \"absolute\":\n position_embeddings = self.position_embeddings(position_ids)\n embeddings = embeddings + position_embeddings\n\n if query_embeds is not None:\n embeddings = torch.cat((query_embeds, embeddings), dim=1)\n else:\n embeddings = query_embeds\n\n embeddings = self.LayerNorm(embeddings)\n embeddings = self.dropout(embeddings)\n return embeddings\n\n\nclass BertSelfAttention(nn.Module):\n def __init__(self, config, is_cross_attention):\n super().__init__()\n self.config = config\n if config.hidden_size % config.num_attention_heads != 0 and not hasattr(\n config, \"embedding_size\"\n ):\n raise ValueError(\n \"The hidden size (%d) is not a multiple of the number of attention \"\n \"heads (%d)\" % (config.hidden_size, config.num_attention_heads)\n )\n\n self.num_attention_heads = config.num_attention_heads\n self.attention_head_size = int(config.hidden_size / config.num_attention_heads)\n self.all_head_size = self.num_attention_heads * self.attention_head_size\n\n self.query = nn.Linear(config.hidden_size, self.all_head_size)\n if is_cross_attention:\n self.key = nn.Linear(config.encoder_width, self.all_head_size)\n self.value = nn.Linear(config.encoder_width, self.all_head_size)\n else:\n self.key = nn.Linear(config.hidden_size, self.all_head_size)\n self.value = nn.Linear(config.hidden_size, self.all_head_size)\n\n self.dropout = nn.Dropout(config.attention_probs_dropout_prob)\n self.position_embedding_type = getattr(\n config, \"position_embedding_type\", \"absolute\"\n )\n if (\n self.position_embedding_type == \"relative_key\"\n or self.position_embedding_type == \"relative_key_query\"\n ):\n self.max_position_embeddings = config.max_position_embeddings\n self.distance_embedding = nn.Embedding(\n 2 * config.max_position_embeddings - 1, self.attention_head_size\n )\n self.save_attention = False\n\n def save_attn_gradients(self, attn_gradients):\n self.attn_gradients = attn_gradients\n\n def get_attn_gradients(self):\n return self.attn_gradients\n\n def save_attention_map(self, attention_map):\n self.attention_map = attention_map\n\n def get_attention_map(self):\n return self.attention_map\n\n def transpose_for_scores(self, x):\n new_x_shape = x.size()[:-1] + (\n self.num_attention_heads,\n self.attention_head_size,\n )\n x = x.view(*new_x_shape)\n return x.permute(0, 2, 1, 3)\n\n def forward(\n self,\n hidden_states,\n attention_mask=None,\n head_mask=None,\n encoder_hidden_states=None,\n encoder_attention_mask=None,\n past_key_value=None,\n output_attentions=False,\n ):\n\n # If this is instantiated as a cross-attention module, the keys\n # and values come from an encoder; the attention mask needs to be\n # such that the encoder's padding tokens are not attended to.\n is_cross_attention = encoder_hidden_states is not None\n\n if is_cross_attention:\n key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))\n value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))\n attention_mask = encoder_attention_mask\n elif past_key_value is not None:\n key_layer = self.transpose_for_scores(self.key(hidden_states))\n value_layer = self.transpose_for_scores(self.value(hidden_states))\n key_layer = torch.cat([past_key_value[0], key_layer], dim=2)\n value_layer = torch.cat([past_key_value[1], value_layer], dim=2)\n else:\n key_layer = self.transpose_for_scores(self.key(hidden_states))\n value_layer = self.transpose_for_scores(self.value(hidden_states))\n\n mixed_query_layer = self.query(hidden_states)\n\n query_layer = self.transpose_for_scores(mixed_query_layer)\n\n past_key_value = (key_layer, value_layer)\n\n # Take the dot product between \"query\" and \"key\" to get the raw attention scores.\n attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))\n\n if (\n self.position_embedding_type == \"relative_key\"\n or self.position_embedding_type == \"relative_key_query\"\n ):\n seq_length = hidden_states.size()[1]\n position_ids_l = torch.arange(\n seq_length, dtype=torch.long, device=hidden_states.device\n ).view(-1, 1)\n position_ids_r = torch.arange(\n seq_length, dtype=torch.long, device=hidden_states.device\n ).view(1, -1)\n distance = position_ids_l - position_ids_r\n positional_embedding = self.distance_embedding(\n distance + self.max_position_embeddings - 1\n )\n positional_embedding = positional_embedding.to(\n dtype=query_layer.dtype\n ) # fp16 compatibility\n\n if self.position_embedding_type == \"relative_key\":\n relative_position_scores = torch.einsum(\n \"bhld,lrd->bhlr\", query_layer, positional_embedding\n )\n attention_scores = attention_scores + relative_position_scores\n elif self.position_embedding_type == \"relative_key_query\":\n relative_position_scores_query = torch.einsum(\n \"bhld,lrd->bhlr\", query_layer, positional_embedding\n )\n relative_position_scores_key = torch.einsum(\n \"bhrd,lrd->bhlr\", key_layer, positional_embedding\n )\n attention_scores = (\n attention_scores\n + relative_position_scores_query\n + relative_position_scores_key\n )\n\n attention_scores = attention_scores / math.sqrt(self.attention_head_size)\n if attention_mask is not None:\n # Apply the attention mask is (precomputed for all layers in BertModel forward() function)\n attention_scores = attention_scores + attention_mask\n\n # Normalize the attention scores to probabilities.\n attention_probs = nn.Softmax(dim=-1)(attention_scores)\n\n if is_cross_attention and self.save_attention:\n self.save_attention_map(attention_probs)\n attention_probs.register_hook(self.save_attn_gradients)\n\n # This is actually dropping out entire tokens to attend to, which might\n # seem a bit unusual, but is taken from the original Transformer paper.\n attention_probs_dropped = self.dropout(attention_probs)\n\n # Mask heads if we want to\n if head_mask is not None:\n attention_probs_dropped = attention_probs_dropped * head_mask\n\n context_layer = torch.matmul(attention_probs_dropped, value_layer)\n\n context_layer = context_layer.permute(0, 2, 1, 3).contiguous()\n new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)\n context_layer = context_layer.view(*new_context_layer_shape)\n\n outputs = (\n (context_layer, attention_probs) if output_attentions else (context_layer,)\n )\n\n outputs = outputs + (past_key_value,)\n return outputs\n\n\nclass BertSelfOutput(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.dense = nn.Linear(config.hidden_size, config.hidden_size)\n self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)\n self.dropout = nn.Dropout(config.hidden_dropout_prob)\n\n def forward(self, hidden_states, input_tensor):\n hidden_states = self.dense(hidden_states)\n hidden_states = self.dropout(hidden_states)\n hidden_states = self.LayerNorm(hidden_states + input_tensor)\n return hidden_states\n\n\nclass BertAttention(nn.Module):\n def __init__(self, config, is_cross_attention=False):\n super().__init__()\n self.self = BertSelfAttention(config, is_cross_attention)\n self.output = BertSelfOutput(config)\n self.pruned_heads = set()\n\n def prune_heads(self, heads):\n if len(heads) == 0:\n return\n heads, index = find_pruneable_heads_and_indices(\n heads,\n self.self.num_attention_heads,\n self.self.attention_head_size,\n self.pruned_heads,\n )\n\n # Prune linear layers\n self.self.query = prune_linear_layer(self.self.query, index)\n self.self.key = prune_linear_layer(self.self.key, index)\n self.self.value = prune_linear_layer(self.self.value, index)\n self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)\n\n # Update hyper params and store pruned heads\n self.self.num_attention_heads = self.self.num_attention_heads - len(heads)\n self.self.all_head_size = (\n self.self.attention_head_size * self.self.num_attention_heads\n )\n self.pruned_heads = self.pruned_heads.union(heads)\n\n def forward(\n self,\n hidden_states,\n attention_mask=None,\n head_mask=None,\n encoder_hidden_states=None,\n encoder_attention_mask=None,\n past_key_value=None,\n output_attentions=False,\n ):\n self_outputs = self.self(\n hidden_states,\n attention_mask,\n head_mask,\n encoder_hidden_states,\n encoder_attention_mask,\n past_key_value,\n output_attentions,\n )\n attention_output = self.output(self_outputs[0], hidden_states)\n\n outputs = (attention_output,) + self_outputs[\n 1:\n ] # add attentions if we output them\n return outputs\n\n\nclass BertIntermediate(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.dense = nn.Linear(config.hidden_size, config.intermediate_size)\n if isinstance(config.hidden_act, str):\n self.intermediate_act_fn = ACT2FN[config.hidden_act]\n else:\n self.intermediate_act_fn = config.hidden_act\n\n def forward(self, hidden_states):\n hidden_states = self.dense(hidden_states)\n hidden_states = self.intermediate_act_fn(hidden_states)\n return hidden_states\n\n\nclass BertOutput(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.dense = nn.Linear(config.intermediate_size, config.hidden_size)\n self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)\n self.dropout = nn.Dropout(config.hidden_dropout_prob)\n\n def forward(self, hidden_states, input_tensor):\n hidden_states = self.dense(hidden_states)\n hidden_states = self.dropout(hidden_states)\n hidden_states = self.LayerNorm(hidden_states + input_tensor)\n return hidden_states\n\n\nclass BertLayer(nn.Module):\n def __init__(self, config, layer_num):\n super().__init__()\n self.config = config\n self.chunk_size_feed_forward = config.chunk_size_feed_forward\n self.seq_len_dim = 1\n self.attention = BertAttention(config)\n self.layer_num = layer_num\n if (\n self.config.add_cross_attention\n and layer_num % self.config.cross_attention_freq == 0\n ):\n self.crossattention = BertAttention(\n config, is_cross_attention=self.config.add_cross_attention\n )\n self.has_cross_attention = True\n else:\n self.has_cross_attention = False\n self.intermediate = BertIntermediate(config)\n self.output = BertOutput(config)\n\n self.intermediate_query = BertIntermediate(config)\n self.output_query = BertOutput(config)\n\n def forward(\n self,\n hidden_states,\n attention_mask=None,\n head_mask=None,\n encoder_hidden_states=None,\n encoder_attention_mask=None,\n past_key_value=None,\n output_attentions=False,\n query_length=0,\n ):\n # decoder uni-directional self-attention cached key/values tuple is at positions 1,2\n self_attn_past_key_value = (\n past_key_value[:2] if past_key_value is not None else None\n )\n self_attention_outputs = self.attention(\n hidden_states,\n attention_mask,\n head_mask,\n output_attentions=output_attentions,\n past_key_value=self_attn_past_key_value,\n )\n attention_output = self_attention_outputs[0]\n outputs = self_attention_outputs[1:-1]\n\n present_key_value = self_attention_outputs[-1]\n\n if query_length > 0:\n query_attention_output = attention_output[:, :query_length, :]\n\n if self.has_cross_attention:\n assert (\n encoder_hidden_states is not None\n ), \"encoder_hidden_states must be given for cross-attention layers\"\n cross_attention_outputs = self.crossattention(\n query_attention_output,\n attention_mask,\n head_mask,\n encoder_hidden_states,\n encoder_attention_mask,\n output_attentions=output_attentions,\n )\n query_attention_output = cross_attention_outputs[0]\n outputs = (\n outputs + cross_attention_outputs[1:-1]\n ) # add cross attentions if we output attention weights\n\n layer_output = apply_chunking_to_forward(\n self.feed_forward_chunk_query,\n self.chunk_size_feed_forward,\n self.seq_len_dim,\n query_attention_output,\n )\n if attention_output.shape[1] > query_length:\n layer_output_text = apply_chunking_to_forward(\n self.feed_forward_chunk,\n self.chunk_size_feed_forward,\n self.seq_len_dim,\n attention_output[:, query_length:, :],\n )\n layer_output = torch.cat([layer_output, layer_output_text], dim=1)\n else:\n layer_output = apply_chunking_to_forward(\n self.feed_forward_chunk,\n self.chunk_size_feed_forward,\n self.seq_len_dim,\n attention_output,\n )\n outputs = (layer_output,) + outputs\n\n outputs = outputs + (present_key_value,)\n\n return outputs\n\n def feed_forward_chunk(self, attention_output):\n intermediate_output = self.intermediate(attention_output)\n layer_output = self.output(intermediate_output, attention_output)\n return layer_output\n\n def feed_forward_chunk_query(self, attention_output):\n intermediate_output = self.intermediate_query(attention_output)\n layer_output = self.output_query(intermediate_output, attention_output)\n return layer_output\n\n\nclass BertEncoder(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.config = config\n self.layer = nn.ModuleList(\n [BertLayer(config, i) for i in range(config.num_hidden_layers)]\n )\n\n def forward(\n self,\n hidden_states,\n attention_mask=None,\n head_mask=None,\n encoder_hidden_states=None,\n encoder_attention_mask=None,\n past_key_values=None,\n use_cache=None,\n output_attentions=False,\n output_hidden_states=False,\n return_dict=True,\n query_length=0,\n ):\n all_hidden_states = () if output_hidden_states else None\n all_self_attentions = () if output_attentions else None\n all_cross_attentions = (\n () if output_attentions and self.config.add_cross_attention else None\n )\n\n next_decoder_cache = () if use_cache else None\n\n for i in range(self.config.num_hidden_layers):\n layer_module = self.layer[i]\n if output_hidden_states:\n all_hidden_states = all_hidden_states + (hidden_states,)\n\n layer_head_mask = head_mask[i] if head_mask is not None else None\n past_key_value = past_key_values[i] if past_key_values is not None else None\n\n if getattr(self.config, \"gradient_checkpointing\", False) and self.training:\n\n if use_cache:\n logger.warn(\n \"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...\"\n )\n use_cache = False\n\n def create_custom_forward(module):\n def custom_forward(*inputs):\n return module(\n *inputs, past_key_value, output_attentions, query_length\n )\n\n return custom_forward\n\n layer_outputs = torch.utils.checkpoint.checkpoint(\n create_custom_forward(layer_module),\n hidden_states,\n attention_mask,\n layer_head_mask,\n encoder_hidden_states,\n encoder_attention_mask,\n )\n else:\n layer_outputs = layer_module(\n hidden_states,\n attention_mask,\n layer_head_mask,\n encoder_hidden_states,\n encoder_attention_mask,\n past_key_value,\n output_attentions,\n query_length,\n )\n\n hidden_states = layer_outputs[0]\n if use_cache:\n next_decoder_cache += (layer_outputs[-1],)\n if output_attentions:\n all_self_attentions = all_self_attentions + (layer_outputs[1],)\n all_cross_attentions = all_cross_attentions + (layer_outputs[2],)\n\n if output_hidden_states:\n all_hidden_states = all_hidden_states + (hidden_states,)\n\n if not return_dict:\n return tuple(\n v\n for v in [\n hidden_states,\n next_decoder_cache,\n all_hidden_states,\n all_self_attentions,\n all_cross_attentions,\n ]\n if v is not None\n )\n return BaseModelOutputWithPastAndCrossAttentions(\n last_hidden_state=hidden_states,\n past_key_values=next_decoder_cache,\n hidden_states=all_hidden_states,\n attentions=all_self_attentions,\n cross_attentions=all_cross_attentions,\n )\n\n\nclass BertPooler(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.dense = nn.Linear(config.hidden_size, config.hidden_size)\n self.activation = nn.Tanh()\n\n def forward(self, hidden_states):\n # We \"pool\" the model by simply taking the hidden state corresponding\n # to the first token.\n first_token_tensor = hidden_states[:, 0]\n pooled_output = self.dense(first_token_tensor)\n pooled_output = self.activation(pooled_output)\n return pooled_output\n\n\nclass BertPredictionHeadTransform(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.dense = nn.Linear(config.hidden_size, config.hidden_size)\n if isinstance(config.hidden_act, str):\n self.transform_act_fn = ACT2FN[config.hidden_act]\n else:\n self.transform_act_fn = config.hidden_act\n self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)\n\n def forward(self, hidden_states):\n hidden_states = self.dense(hidden_states)\n hidden_states = self.transform_act_fn(hidden_states)\n hidden_states = self.LayerNorm(hidden_states)\n return hidden_states\n\n\nclass BertLMPredictionHead(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.transform = BertPredictionHeadTransform(config)\n\n # The output weights are the same as the input embeddings, but there is\n # an output-only bias for each token.\n self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)\n\n self.bias = nn.Parameter(torch.zeros(config.vocab_size))\n\n # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`\n self.decoder.bias = self.bias\n\n def forward(self, hidden_states):\n hidden_states = self.transform(hidden_states)\n hidden_states = self.decoder(hidden_states)\n return hidden_states\n\n\nclass BertOnlyMLMHead(nn.Module):\n def __init__(self, config):\n super().__init__()\n self.predictions = BertLMPredictionHead(config)\n\n def forward(self, sequence_output):\n prediction_scores = self.predictions(sequence_output)\n return prediction_scores\n\n\nclass BertPreTrainedModel(PreTrainedModel):\n \"\"\"\n An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained\n models.\n \"\"\"\n\n config_class = BertConfig\n base_model_prefix = \"bert\"\n _keys_to_ignore_on_load_missing = [r\"position_ids\"]\n\n def _init_weights(self, module):\n \"\"\"Initialize the weights\"\"\"\n if isinstance(module, (nn.Linear, nn.Embedding)):\n # Slightly different from the TF version which uses truncated_normal for initialization\n # cf https://github.com/pytorch/pytorch/pull/5617\n module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)\n elif isinstance(module, nn.LayerNorm):\n module.bias.data.zero_()\n module.weight.data.fill_(1.0)\n if isinstance(module, nn.Linear) and module.bias is not None:\n module.bias.data.zero_()\n\n\nclass BertModel(BertPreTrainedModel):\n \"\"\"\n The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of\n cross-attention is added between the self-attention layers, following the architecture described in `Attention is\n all you need `__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,\n Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.\n argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an\n input to the forward pass.\n \"\"\"\n\n def __init__(self, config, add_pooling_layer=False):\n super().__init__(config)\n self.config = config\n\n self.embeddings = BertEmbeddings(config)\n\n self.encoder = BertEncoder(config)\n\n self.pooler = BertPooler(config) if add_pooling_layer else None\n\n self.init_weights()\n\n def get_input_embeddings(self):\n return self.embeddings.word_embeddings\n\n def set_input_embeddings(self, value):\n self.embeddings.word_embeddings = value\n\n def _prune_heads(self, heads_to_prune):\n \"\"\"\n Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base\n class PreTrainedModel\n \"\"\"\n for layer, heads in heads_to_prune.items():\n self.encoder.layer[layer].attention.prune_heads(heads)\n\n def get_extended_attention_mask(\n self,\n attention_mask: Tensor,\n input_shape: Tuple[int],\n device: device,\n is_decoder: bool,\n has_query: bool = False,\n ) -> Tensor:\n \"\"\"\n Makes broadcastable attention and causal masks so that future and masked tokens are ignored.\n\n Arguments:\n attention_mask (:obj:`torch.Tensor`):\n Mask with ones indicating tokens to attend to, zeros for tokens to ignore.\n input_shape (:obj:`Tuple[int]`):\n The shape of the input to the model.\n device: (:obj:`torch.device`):\n The device of the input to the model.\n\n Returns:\n :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.\n \"\"\"\n # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]\n # ourselves in which case we just need to make it broadcastable to all heads.\n if attention_mask.dim() == 3:\n extended_attention_mask = attention_mask[:, None, :, :]\n elif attention_mask.dim() == 2:\n # Provided a padding mask of dimensions [batch_size, seq_length]\n # - if the model is a decoder, apply a causal mask in addition to the padding mask\n # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]\n if is_decoder:\n batch_size, seq_length = input_shape\n\n seq_ids = torch.arange(seq_length, device=device)\n causal_mask = (\n seq_ids[None, None, :].repeat(batch_size, seq_length, 1)\n <= seq_ids[None, :, None]\n )\n\n # add a prefix ones mask to the causal mask\n # causal and attention masks must have same type with pytorch version < 1.3\n causal_mask = causal_mask.to(attention_mask.dtype)\n\n if causal_mask.shape[1] < attention_mask.shape[1]:\n prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]\n if has_query: # UniLM style attention mask\n causal_mask = torch.cat(\n [\n torch.zeros(\n (batch_size, prefix_seq_len, seq_length),\n device=device,\n dtype=causal_mask.dtype,\n ),\n causal_mask,\n ],\n axis=1,\n )\n causal_mask = torch.cat(\n [\n torch.ones(\n (batch_size, causal_mask.shape[1], prefix_seq_len),\n device=device,\n dtype=causal_mask.dtype,\n ),\n causal_mask,\n ],\n axis=-1,\n )\n extended_attention_mask = (\n causal_mask[:, None, :, :] * attention_mask[:, None, None, :]\n )\n else:\n extended_attention_mask = attention_mask[:, None, None, :]\n else:\n raise ValueError(\n \"Wrong shape for input_ids (shape {}) or attention_mask (shape {})\".format(\n input_shape, attention_mask.shape\n )\n )\n\n # Since attention_mask is 1.0 for positions we want to attend and 0.0 for\n # masked positions, this operation will create a tensor which is 0.0 for\n # positions we want to attend and -10000.0 for masked positions.\n # Since we are adding it to the raw scores before the softmax, this is\n # effectively the same as removing these entirely.\n extended_attention_mask = extended_attention_mask.to(\n dtype=self.dtype\n ) # fp16 compatibility\n extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0\n return extended_attention_mask\n\n def forward(\n self,\n input_ids=None,\n attention_mask=None,\n position_ids=None,\n head_mask=None,\n query_embeds=None,\n encoder_hidden_states=None,\n encoder_attention_mask=None,\n past_key_values=None,\n use_cache=None,\n output_attentions=None,\n output_hidden_states=None,\n return_dict=None,\n is_decoder=False,\n ):\n r\"\"\"\n encoder_hidden_states (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):\n Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if\n the model is configured as a decoder.\n encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):\n Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in\n the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:\n - 1 for tokens that are **not masked**,\n - 0 for tokens that are **masked**.\n past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):\n Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.\n If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`\n (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`\n instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.\n use_cache (:obj:`bool`, `optional`):\n If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up\n decoding (see :obj:`past_key_values`).\n \"\"\"\n output_attentions = (\n output_attentions\n if output_attentions is not None\n else self.config.output_attentions\n )\n output_hidden_states = (\n output_hidden_states\n if output_hidden_states is not None\n else self.config.output_hidden_states\n )\n return_dict = (\n return_dict if return_dict is not None else self.config.use_return_dict\n )\n\n # use_cache = use_cache if use_cache is not None else self.config.use_cache\n\n if input_ids is None:\n assert (\n query_embeds is not None\n ), \"You have to specify query_embeds when input_ids is None\"\n\n # past_key_values_length\n past_key_values_length = (\n past_key_values[0][0].shape[2] - self.config.query_length\n if past_key_values is not None\n else 0\n )\n\n query_length = query_embeds.shape[1] if query_embeds is not None else 0\n\n embedding_output = self.embeddings(\n input_ids=input_ids,\n position_ids=position_ids,\n query_embeds=query_embeds,\n past_key_values_length=past_key_values_length,\n )\n\n input_shape = embedding_output.size()[:-1]\n batch_size, seq_length = input_shape\n device = embedding_output.device\n\n if attention_mask is None:\n attention_mask = torch.ones(\n ((batch_size, seq_length + past_key_values_length)), device=device\n )\n\n # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]\n # ourselves in which case we just need to make it broadcastable to all heads.\n if is_decoder:\n extended_attention_mask = self.get_extended_attention_mask(\n attention_mask,\n input_ids.shape,\n device,\n is_decoder,\n has_query=(query_embeds is not None),\n )\n else:\n extended_attention_mask = self.get_extended_attention_mask(\n attention_mask, input_shape, device, is_decoder\n )\n\n # If a 2D or 3D attention mask is provided for the cross-attention\n # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]\n if encoder_hidden_states is not None:\n if type(encoder_hidden_states) == list:\n encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[\n 0\n ].size()\n else:\n (\n encoder_batch_size,\n encoder_sequence_length,\n _,\n ) = encoder_hidden_states.size()\n encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)\n\n if type(encoder_attention_mask) == list:\n encoder_extended_attention_mask = [\n self.invert_attention_mask(mask) for mask in encoder_attention_mask\n ]\n elif encoder_attention_mask is None:\n encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)\n encoder_extended_attention_mask = self.invert_attention_mask(\n encoder_attention_mask\n )\n else:\n encoder_extended_attention_mask = self.invert_attention_mask(\n encoder_attention_mask\n )\n else:\n encoder_extended_attention_mask = None\n\n # Prepare head mask if needed\n # 1.0 in head_mask indicate we keep the head\n # attention_probs has shape bsz x n_heads x N x N\n # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]\n # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]\n head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)\n\n encoder_outputs = self.encoder(\n embedding_output,\n attention_mask=extended_attention_mask,\n head_mask=head_mask,\n encoder_hidden_states=encoder_hidden_states,\n encoder_attention_mask=encoder_extended_attention_mask,\n past_key_values=past_key_values,\n use_cache=use_cache,\n output_attentions=output_attentions,\n output_hidden_states=output_hidden_states,\n return_dict=return_dict,\n query_length=query_length,\n )\n sequence_output = encoder_outputs[0]\n pooled_output = (\n self.pooler(sequence_output) if self.pooler is not None else None\n )\n\n if not return_dict:\n return (sequence_output, pooled_output) + encoder_outputs[1:]\n\n return BaseModelOutputWithPoolingAndCrossAttentions(\n last_hidden_state=sequence_output,\n pooler_output=pooled_output,\n past_key_values=encoder_outputs.past_key_values,\n hidden_states=encoder_outputs.hidden_states,\n attentions=encoder_outputs.attentions,\n cross_attentions=encoder_outputs.cross_attentions,\n )\n\n\nclass BertLMHeadModel(BertPreTrainedModel):\n\n _keys_to_ignore_on_load_unexpected = [r\"pooler\"]\n _keys_to_ignore_on_load_missing = [r\"position_ids\", r\"predictions.decoder.bias\"]\n\n def __init__(self, config):\n super().__init__(config)\n\n self.bert = BertModel(config, add_pooling_layer=False)\n self.cls = BertOnlyMLMHead(config)\n\n self.init_weights()\n\n def get_output_embeddings(self):\n return self.cls.predictions.decoder\n\n def set_output_embeddings(self, new_embeddings):\n self.cls.predictions.decoder = new_embeddings\n\n def forward(\n self,\n input_ids=None,\n attention_mask=None,\n position_ids=None,\n head_mask=None,\n query_embeds=None,\n encoder_hidden_states=None,\n encoder_attention_mask=None,\n labels=None,\n past_key_values=None,\n use_cache=True,\n output_attentions=None,\n output_hidden_states=None,\n return_dict=None,\n return_logits=False,\n is_decoder=True,\n reduction=\"mean\",\n ):\n r\"\"\"\n encoder_hidden_states (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):\n Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if\n the model is configured as a decoder.\n encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):\n Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in\n the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:\n - 1 for tokens that are **not masked**,\n - 0 for tokens that are **masked**.\n labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):\n Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in\n ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are\n ignored (masked), the loss is only computed for the tokens with labels n ``[0, ..., config.vocab_size]``\n past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):\n Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.\n If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`\n (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`\n instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.\n use_cache (:obj:`bool`, `optional`):\n If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up\n decoding (see :obj:`past_key_values`).\n Returns:\n Example::\n >>> from transformers import BertTokenizer, BertLMHeadModel, BertConfig\n >>> import torch\n >>> tokenizer = BertTokenizer.from_pretrained('bert-base-cased')\n >>> config = BertConfig.from_pretrained(\"bert-base-cased\")\n >>> model = BertLMHeadModel.from_pretrained('bert-base-cased', config=config)\n >>> inputs = tokenizer(\"Hello, my dog is cute\", return_tensors=\"pt\")\n >>> outputs = model(**inputs)\n >>> prediction_logits = outputs.logits\n \"\"\"\n return_dict = (\n return_dict if return_dict is not None else self.config.use_return_dict\n )\n if labels is not None:\n use_cache = False\n if past_key_values is not None:\n query_embeds = None\n\n outputs = self.bert(\n input_ids,\n attention_mask=attention_mask,\n position_ids=position_ids,\n head_mask=head_mask,\n query_embeds=query_embeds,\n encoder_hidden_states=encoder_hidden_states,\n encoder_attention_mask=encoder_attention_mask,\n past_key_values=past_key_values,\n use_cache=use_cache,\n output_attentions=output_attentions,\n output_hidden_states=output_hidden_states,\n return_dict=return_dict,\n is_decoder=is_decoder,\n )\n\n sequence_output = outputs[0]\n if query_embeds is not None:\n sequence_output = outputs[0][:, query_embeds.shape[1] :, :]\n\n prediction_scores = self.cls(sequence_output)\n\n if return_logits:\n return prediction_scores[:, :-1, :].contiguous()\n\n lm_loss = None\n if labels is not None:\n # we are doing next-token prediction; shift prediction scores and input ids by one\n shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()\n labels = labels[:, 1:].contiguous()\n loss_fct = CrossEntropyLoss(reduction=reduction, label_smoothing=0.1)\n lm_loss = loss_fct(\n shifted_prediction_scores.view(-1, self.config.vocab_size),\n labels.view(-1),\n )\n if reduction == \"none\":\n lm_loss = lm_loss.view(prediction_scores.size(0), -1).sum(1)\n\n if not return_dict:\n output = (prediction_scores,) + outputs[2:]\n return ((lm_loss,) + output) if lm_loss is not None else output\n\n return CausalLMOutputWithCrossAttentions(\n loss=lm_loss,\n logits=prediction_scores,\n past_key_values=outputs.past_key_values,\n hidden_states=outputs.hidden_states,\n attentions=outputs.attentions,\n cross_attentions=outputs.cross_attentions,\n )\n\n def prepare_inputs_for_generation(\n self, input_ids, query_embeds, past=None, attention_mask=None, **model_kwargs\n ):\n # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly\n if attention_mask is None:\n attention_mask = input_ids.new_ones(input_ids.shape)\n query_mask = input_ids.new_ones(query_embeds.shape[:-1])\n attention_mask = torch.cat([query_mask, attention_mask], dim=-1)\n\n # cut decoder_input_ids if past is used\n if past is not None:\n input_ids = input_ids[:, -1:]\n\n return {\n \"input_ids\": input_ids,\n \"query_embeds\": query_embeds,\n \"attention_mask\": attention_mask,\n \"past_key_values\": past,\n \"encoder_hidden_states\": model_kwargs.get(\"encoder_hidden_states\", None),\n \"encoder_attention_mask\": model_kwargs.get(\"encoder_attention_mask\", None),\n \"is_decoder\": True,\n }\n\n def _reorder_cache(self, past, beam_idx):\n reordered_past = ()\n for layer_past in past:\n reordered_past += (\n tuple(\n past_state.index_select(0, beam_idx) for past_state in layer_past\n ),\n )\n return reordered_past\n\n\nclass BertForMaskedLM(BertPreTrainedModel):\n\n _keys_to_ignore_on_load_unexpected = [r\"pooler\"]\n _keys_to_ignore_on_load_missing = [r\"position_ids\", r\"predictions.decoder.bias\"]\n\n def __init__(self, config):\n super().__init__(config)\n\n self.bert = BertModel(config, add_pooling_layer=False)\n self.cls = BertOnlyMLMHead(config)\n\n self.init_weights()\n\n def get_output_embeddings(self):\n return self.cls.predictions.decoder\n\n def set_output_embeddings(self, new_embeddings):\n self.cls.predictions.decoder = new_embeddings\n\n def forward(\n self,\n input_ids=None,\n attention_mask=None,\n position_ids=None,\n head_mask=None,\n query_embeds=None,\n encoder_hidden_states=None,\n encoder_attention_mask=None,\n labels=None,\n output_attentions=None,\n output_hidden_states=None,\n return_dict=None,\n return_logits=False,\n is_decoder=False,\n ):\n r\"\"\"\n labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):\n Labels for computing the masked language modeling loss. Indices should be in ``[-100, 0, ...,\n config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are ignored\n (masked), the loss is only computed for the tokens with labels in ``[0, ..., config.vocab_size]``\n \"\"\"\n\n return_dict = (\n return_dict if return_dict is not None else self.config.use_return_dict\n )\n\n outputs = self.bert(\n input_ids,\n attention_mask=attention_mask,\n position_ids=position_ids,\n head_mask=head_mask,\n query_embeds=query_embeds,\n encoder_hidden_states=encoder_hidden_states,\n encoder_attention_mask=encoder_attention_mask,\n output_attentions=output_attentions,\n output_hidden_states=output_hidden_states,\n return_dict=return_dict,\n is_decoder=is_decoder,\n )\n\n if query_embeds is not None:\n sequence_output = outputs[0][:, query_embeds.shape[1] :, :]\n prediction_scores = self.cls(sequence_output)\n\n if return_logits:\n return prediction_scores\n\n masked_lm_loss = None\n if labels is not None:\n loss_fct = CrossEntropyLoss() # -100 index = padding token\n masked_lm_loss = loss_fct(\n prediction_scores.view(-1, self.config.vocab_size), labels.view(-1)\n )\n\n if not return_dict:\n output = (prediction_scores,) + outputs[2:]\n return (\n ((masked_lm_loss,) + output) if masked_lm_loss is not None else output\n )\n\n return MaskedLMOutput(\n loss=masked_lm_loss,\n logits=prediction_scores,\n hidden_states=outputs.hidden_states,\n attentions=outputs.attentions,\n )\n" }, { "path": "minigpt4/models/__init__.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport logging\nimport torch\nfrom omegaconf import OmegaConf\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.models.base_model import BaseModel\nfrom minigpt4.models.minigpt_base import MiniGPTBase\nfrom minigpt4.models.minigpt4 import MiniGPT4\nfrom minigpt4.models.minigpt_v2 import MiniGPTv2\nfrom minigpt4.processors.base_processor import BaseProcessor\n\n\n__all__ = [\n \"load_model\",\n \"BaseModel\",\n \"MiniGPTBase\",\n \"MiniGPT4\",\n \"MiniGPTv2\"\n]\n\n\ndef load_model(name, model_type, is_eval=False, device=\"cpu\", checkpoint=None):\n \"\"\"\n Load supported models.\n\n To list all available models and types in registry:\n >>> from minigpt4.models import model_zoo\n >>> print(model_zoo)\n\n Args:\n name (str): name of the model.\n model_type (str): type of the model.\n is_eval (bool): whether the model is in eval mode. Default: False.\n device (str): device to use. Default: \"cpu\".\n checkpoint (str): path or to checkpoint. Default: None.\n Note that expecting the checkpoint to have the same keys in state_dict as the model.\n\n Returns:\n model (torch.nn.Module): model.\n \"\"\"\n\n model = registry.get_model_class(name).from_pretrained(model_type=model_type)\n\n if checkpoint is not None:\n model.load_checkpoint(checkpoint)\n\n if is_eval:\n model.eval()\n\n if device == \"cpu\":\n model = model.float()\n\n return model.to(device)\n\n\ndef load_preprocess(config):\n \"\"\"\n Load preprocessor configs and construct preprocessors.\n\n If no preprocessor is specified, return BaseProcessor, which does not do any preprocessing.\n\n Args:\n config (dict): preprocessor configs.\n\n Returns:\n vis_processors (dict): preprocessors for visual inputs.\n txt_processors (dict): preprocessors for text inputs.\n\n Key is \"train\" or \"eval\" for processors used in training and evaluation respectively.\n \"\"\"\n\n def _build_proc_from_cfg(cfg):\n return (\n registry.get_processor_class(cfg.name).from_config(cfg)\n if cfg is not None\n else BaseProcessor()\n )\n\n vis_processors = dict()\n txt_processors = dict()\n\n vis_proc_cfg = config.get(\"vis_processor\")\n txt_proc_cfg = config.get(\"text_processor\")\n\n if vis_proc_cfg is not None:\n vis_train_cfg = vis_proc_cfg.get(\"train\")\n vis_eval_cfg = vis_proc_cfg.get(\"eval\")\n else:\n vis_train_cfg = None\n vis_eval_cfg = None\n\n vis_processors[\"train\"] = _build_proc_from_cfg(vis_train_cfg)\n vis_processors[\"eval\"] = _build_proc_from_cfg(vis_eval_cfg)\n\n if txt_proc_cfg is not None:\n txt_train_cfg = txt_proc_cfg.get(\"train\")\n txt_eval_cfg = txt_proc_cfg.get(\"eval\")\n else:\n txt_train_cfg = None\n txt_eval_cfg = None\n\n txt_processors[\"train\"] = _build_proc_from_cfg(txt_train_cfg)\n txt_processors[\"eval\"] = _build_proc_from_cfg(txt_eval_cfg)\n\n return vis_processors, txt_processors\n\n\ndef load_model_and_preprocess(name, model_type, is_eval=False, device=\"cpu\"):\n \"\"\"\n Load model and its related preprocessors.\n\n List all available models and types in registry:\n >>> from minigpt4.models import model_zoo\n >>> print(model_zoo)\n\n Args:\n name (str): name of the model.\n model_type (str): type of the model.\n is_eval (bool): whether the model is in eval mode. Default: False.\n device (str): device to use. Default: \"cpu\".\n\n Returns:\n model (torch.nn.Module): model.\n vis_processors (dict): preprocessors for visual inputs.\n txt_processors (dict): preprocessors for text inputs.\n \"\"\"\n model_cls = registry.get_model_class(name)\n\n # load model\n model = model_cls.from_pretrained(model_type=model_type)\n\n if is_eval:\n model.eval()\n\n # load preprocess\n cfg = OmegaConf.load(model_cls.default_config_path(model_type))\n if cfg is not None:\n preprocess_cfg = cfg.preprocess\n\n vis_processors, txt_processors = load_preprocess(preprocess_cfg)\n else:\n vis_processors, txt_processors = None, None\n logging.info(\n f\"\"\"No default preprocess for model {name} ({model_type}).\n This can happen if the model is not finetuned on downstream datasets,\n or it is not intended for direct use without finetuning.\n \"\"\"\n )\n\n if device == \"cpu\" or device == torch.device(\"cpu\"):\n model = model.float()\n\n return model.to(device), vis_processors, txt_processors\n\n\nclass ModelZoo:\n \"\"\"\n A utility class to create string representation of available model architectures and types.\n\n >>> from minigpt4.models import model_zoo\n >>> # list all available models\n >>> print(model_zoo)\n >>> # show total number of models\n >>> print(len(model_zoo))\n \"\"\"\n\n def __init__(self) -> None:\n self.model_zoo = {\n k: list(v.PRETRAINED_MODEL_CONFIG_DICT.keys())\n for k, v in registry.mapping[\"model_name_mapping\"].items()\n }\n\n def __str__(self) -> str:\n return (\n \"=\" * 50\n + \"\\n\"\n + f\"{'Architectures':<30} {'Types'}\\n\"\n + \"=\" * 50\n + \"\\n\"\n + \"\\n\".join(\n [\n f\"{name:<30} {', '.join(types)}\"\n for name, types in self.model_zoo.items()\n ]\n )\n )\n\n def __iter__(self):\n return iter(self.model_zoo.items())\n\n def __len__(self):\n return sum([len(v) for v in self.model_zoo.values()])\n\n\nmodel_zoo = ModelZoo()\n" }, { "path": "minigpt4/models/base_model.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport os\nimport logging\nimport contextlib\n\nfrom omegaconf import OmegaConf\nimport numpy as np\nimport torch\nimport torch.nn as nn\nfrom transformers import LlamaTokenizer\nfrom peft import (\n LoraConfig,\n get_peft_model,\n prepare_model_for_int8_training,\n)\n\nfrom minigpt4.common.dist_utils import download_cached_file\nfrom minigpt4.common.utils import get_abs_path, is_url\nfrom minigpt4.models.eva_vit import create_eva_vit_g\nfrom minigpt4.models.modeling_llama import LlamaForCausalLM\n\n\n\nclass BaseModel(nn.Module):\n \"\"\"Base class for models.\"\"\"\n\n def __init__(self):\n super().__init__()\n\n @property\n def device(self):\n return list(self.parameters())[-1].device\n\n def load_checkpoint(self, url_or_filename):\n \"\"\"\n Load from a finetuned checkpoint.\n\n This should expect no mismatch in the model keys and the checkpoint keys.\n \"\"\"\n\n if is_url(url_or_filename):\n cached_file = download_cached_file(\n url_or_filename, check_hash=False, progress=True\n )\n checkpoint = torch.load(cached_file, map_location=\"cpu\")\n elif os.path.isfile(url_or_filename):\n checkpoint = torch.load(url_or_filename, map_location=\"cpu\")\n else:\n raise RuntimeError(\"checkpoint url or path is invalid\")\n\n if \"model\" in checkpoint.keys():\n state_dict = checkpoint[\"model\"]\n else:\n state_dict = checkpoint\n\n msg = self.load_state_dict(state_dict, strict=False)\n\n logging.info(\"Missing keys {}\".format(msg.missing_keys))\n logging.info(\"load checkpoint from %s\" % url_or_filename)\n\n return msg\n\n @classmethod\n def from_pretrained(cls, model_type):\n \"\"\"\n Build a pretrained model from default configuration file, specified by model_type.\n\n Args:\n - model_type (str): model type, specifying architecture and checkpoints.\n\n Returns:\n - model (nn.Module): pretrained or finetuned model, depending on the configuration.\n \"\"\"\n model_cfg = OmegaConf.load(cls.default_config_path(model_type)).model\n model = cls.from_config(model_cfg)\n\n return model\n\n @classmethod\n def default_config_path(cls, model_type):\n assert (\n model_type in cls.PRETRAINED_MODEL_CONFIG_DICT\n ), \"Unknown model type {}\".format(model_type)\n return get_abs_path(cls.PRETRAINED_MODEL_CONFIG_DICT[model_type])\n\n def load_checkpoint_from_config(self, cfg, **kwargs):\n \"\"\"\n Load checkpoint as specified in the config file.\n\n If load_finetuned is True, load the finetuned model; otherwise, load the pretrained model.\n When loading the pretrained model, each task-specific architecture may define their\n own load_from_pretrained() method.\n \"\"\"\n load_finetuned = cfg.get(\"load_finetuned\", True)\n if load_finetuned:\n finetune_path = cfg.get(\"finetuned\", None)\n assert (\n finetune_path is not None\n ), \"Found load_finetuned is True, but finetune_path is None.\"\n self.load_checkpoint(url_or_filename=finetune_path)\n else:\n # load pre-trained weights\n pretrain_path = cfg.get(\"pretrained\", None)\n assert \"Found load_finetuned is False, but pretrain_path is None.\"\n self.load_from_pretrained(url_or_filename=pretrain_path, **kwargs)\n\n def before_evaluation(self, **kwargs):\n pass\n\n def show_n_params(self, return_str=True):\n tot = 0\n for p in self.parameters():\n w = 1\n for x in p.shape:\n w *= x\n tot += w\n if return_str:\n if tot >= 1e6:\n return \"{:.1f}M\".format(tot / 1e6)\n else:\n return \"{:.1f}K\".format(tot / 1e3)\n else:\n return tot\n\n def maybe_autocast(self, dtype=torch.float16):\n # if on cpu, don't use autocast\n # if on gpu, use autocast with dtype if provided, otherwise use torch.float16\n enable_autocast = self.device != torch.device(\"cpu\")\n\n if enable_autocast:\n return torch.cuda.amp.autocast(dtype=dtype)\n else:\n return contextlib.nullcontext()\n\n @classmethod\n def init_vision_encoder(\n cls, model_name, img_size, drop_path_rate, use_grad_checkpoint, precision, freeze\n ):\n logging.info('Loading VIT')\n\n assert model_name == \"eva_clip_g\", \"vit model must be eva_clip_g for current version of MiniGPT-4\"\n if not freeze:\n precision = \"fp32\" # fp16 is not for training\n\n visual_encoder = create_eva_vit_g(\n img_size, drop_path_rate, use_grad_checkpoint, precision\n )\n\n ln_vision = LayerNorm(visual_encoder.num_features)\n\n if freeze:\n for name, param in visual_encoder.named_parameters():\n param.requires_grad = False\n visual_encoder = visual_encoder.eval()\n visual_encoder.train = disabled_train\n for name, param in ln_vision.named_parameters():\n param.requires_grad = False\n ln_vision = ln_vision.eval()\n ln_vision.train = disabled_train\n logging.info(\"freeze vision encoder\")\n\n logging.info('Loading VIT Done')\n return visual_encoder, ln_vision\n\n def init_llm(cls, llama_model_path, low_resource=False, low_res_device=0, lora_r=0,\n lora_target_modules=[\"q_proj\",\"v_proj\"], **lora_kargs):\n logging.info('Loading LLAMA')\n llama_tokenizer = LlamaTokenizer.from_pretrained(llama_model_path, use_fast=False)\n llama_tokenizer.pad_token = \"$$\"\n\n if low_resource:\n llama_model = LlamaForCausalLM.from_pretrained(\n llama_model_path,\n torch_dtype=torch.float16,\n load_in_8bit=True,\n device_map={'': low_res_device}\n )\n else:\n llama_model = LlamaForCausalLM.from_pretrained(\n llama_model_path,\n torch_dtype=torch.float16,\n )\n\n if lora_r > 0:\n llama_model = prepare_model_for_int8_training(llama_model)\n loraconfig = LoraConfig(\n r=lora_r,\n bias=\"none\",\n task_type=\"CAUSAL_LM\",\n target_modules=lora_target_modules,\n **lora_kargs\n )\n llama_model = get_peft_model(llama_model, loraconfig)\n\n llama_model.print_trainable_parameters()\n\n else:\n for name, param in llama_model.named_parameters():\n param.requires_grad = False\n logging.info('Loading LLAMA Done')\n return llama_model, llama_tokenizer\n\n\n def load_from_pretrained(self, url_or_filename):\n if is_url(url_or_filename):\n cached_file = download_cached_file(\n url_or_filename, check_hash=False, progress=True\n )\n checkpoint = torch.load(cached_file, map_location=\"cpu\")\n elif os.path.isfile(url_or_filename):\n checkpoint = torch.load(url_or_filename, map_location=\"cpu\")\n else:\n raise RuntimeError(\"checkpoint url or path is invalid\")\n\n state_dict = checkpoint[\"model\"]\n\n msg = self.load_state_dict(state_dict, strict=False)\n\n # logging.info(\"Missing keys {}\".format(msg.missing_keys))\n logging.info(\"load checkpoint from %s\" % url_or_filename)\n\n return msg\n\n\ndef disabled_train(self, mode=True):\n \"\"\"Overwrite model.train with this function to make sure train/eval mode\n does not change anymore.\"\"\"\n return self\n\n\nclass LayerNorm(nn.LayerNorm):\n \"\"\"Subclass torch's LayerNorm to handle fp16.\"\"\"\n\n def forward(self, x: torch.Tensor):\n orig_type = x.dtype\n ret = super().forward(x.type(torch.float32))\n return ret.type(orig_type)\n\n\n\n\n\n" }, { "path": "minigpt4/models/eva_vit.py", "content": "# Based on EVA, BEIT, timm and DeiT code bases\n# https://github.com/baaivision/EVA\n# https://github.com/rwightman/pytorch-image-models/tree/master/timm\n# https://github.com/microsoft/unilm/tree/master/beit\n# https://github.com/facebookresearch/deit/\n# https://github.com/facebookresearch/dino\n# --------------------------------------------------------'\nimport math\nfrom functools import partial\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.utils.checkpoint as checkpoint\nfrom timm.models.layers import drop_path, to_2tuple, trunc_normal_\nfrom timm.models.registry import register_model\n\nfrom minigpt4.common.dist_utils import download_cached_file\n\ndef _cfg(url='', **kwargs):\n return {\n 'url': url,\n 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,\n 'crop_pct': .9, 'interpolation': 'bicubic',\n 'mean': (0.5, 0.5, 0.5), 'std': (0.5, 0.5, 0.5),\n **kwargs\n }\n\n\nclass DropPath(nn.Module):\n \"\"\"Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).\n \"\"\"\n def __init__(self, drop_prob=None):\n super(DropPath, self).__init__()\n self.drop_prob = drop_prob\n\n def forward(self, x):\n return drop_path(x, self.drop_prob, self.training)\n \n def extra_repr(self) -> str:\n return 'p={}'.format(self.drop_prob)\n\n\nclass Mlp(nn.Module):\n def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):\n super().__init__()\n out_features = out_features or in_features\n hidden_features = hidden_features or in_features\n self.fc1 = nn.Linear(in_features, hidden_features)\n self.act = act_layer()\n self.fc2 = nn.Linear(hidden_features, out_features)\n self.drop = nn.Dropout(drop)\n\n def forward(self, x):\n x = self.fc1(x)\n x = self.act(x)\n # x = self.drop(x)\n # commit this for the orignal BERT implement \n x = self.fc2(x)\n x = self.drop(x)\n return x\n\n\nclass Attention(nn.Module):\n def __init__(\n self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0.,\n proj_drop=0., window_size=None, attn_head_dim=None):\n super().__init__()\n self.num_heads = num_heads\n head_dim = dim // num_heads\n if attn_head_dim is not None:\n head_dim = attn_head_dim\n all_head_dim = head_dim * self.num_heads\n self.scale = qk_scale or head_dim ** -0.5\n\n self.qkv = nn.Linear(dim, all_head_dim * 3, bias=False)\n if qkv_bias:\n self.q_bias = nn.Parameter(torch.zeros(all_head_dim))\n self.v_bias = nn.Parameter(torch.zeros(all_head_dim))\n else:\n self.q_bias = None\n self.v_bias = None\n\n if window_size:\n self.window_size = window_size\n self.num_relative_distance = (2 * window_size[0] - 1) * (2 * window_size[1] - 1) + 3\n self.relative_position_bias_table = nn.Parameter(\n torch.zeros(self.num_relative_distance, num_heads)) # 2*Wh-1 * 2*Ww-1, nH\n # cls to token & token 2 cls & cls to cls\n\n # get pair-wise relative position index for each token inside the window\n coords_h = torch.arange(window_size[0])\n coords_w = torch.arange(window_size[1])\n coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww\n coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww\n relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww\n relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2\n relative_coords[:, :, 0] += window_size[0] - 1 # shift to start from 0\n relative_coords[:, :, 1] += window_size[1] - 1\n relative_coords[:, :, 0] *= 2 * window_size[1] - 1\n relative_position_index = \\\n torch.zeros(size=(window_size[0] * window_size[1] + 1, ) * 2, dtype=relative_coords.dtype)\n relative_position_index[1:, 1:] = relative_coords.sum(-1) # Wh*Ww, Wh*Ww\n relative_position_index[0, 0:] = self.num_relative_distance - 3\n relative_position_index[0:, 0] = self.num_relative_distance - 2\n relative_position_index[0, 0] = self.num_relative_distance - 1\n\n self.register_buffer(\"relative_position_index\", relative_position_index)\n else:\n self.window_size = None\n self.relative_position_bias_table = None\n self.relative_position_index = None\n\n self.attn_drop = nn.Dropout(attn_drop)\n self.proj = nn.Linear(all_head_dim, dim)\n self.proj_drop = nn.Dropout(proj_drop)\n\n def forward(self, x, rel_pos_bias=None):\n B, N, C = x.shape\n qkv_bias = None\n if self.q_bias is not None:\n qkv_bias = torch.cat((self.q_bias, torch.zeros_like(self.v_bias, requires_grad=False), self.v_bias))\n # qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)\n qkv = F.linear(input=x, weight=self.qkv.weight, bias=qkv_bias)\n qkv = qkv.reshape(B, N, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)\n q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple)\n\n q = q * self.scale\n attn = (q @ k.transpose(-2, -1))\n\n if self.relative_position_bias_table is not None:\n relative_position_bias = \\\n self.relative_position_bias_table[self.relative_position_index.view(-1)].view(\n self.window_size[0] * self.window_size[1] + 1,\n self.window_size[0] * self.window_size[1] + 1, -1) # Wh*Ww,Wh*Ww,nH\n relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww\n attn = attn + relative_position_bias.unsqueeze(0)\n\n if rel_pos_bias is not None:\n attn = attn + rel_pos_bias\n \n attn = attn.softmax(dim=-1)\n attn = self.attn_drop(attn)\n\n x = (attn @ v).transpose(1, 2).reshape(B, N, -1)\n x = self.proj(x)\n x = self.proj_drop(x)\n return x\n\n\nclass Block(nn.Module):\n\n def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,\n drop_path=0., init_values=None, act_layer=nn.GELU, norm_layer=nn.LayerNorm,\n window_size=None, attn_head_dim=None):\n super().__init__()\n self.norm1 = norm_layer(dim)\n self.attn = Attention(\n dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,\n attn_drop=attn_drop, proj_drop=drop, window_size=window_size, attn_head_dim=attn_head_dim)\n # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here\n self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()\n self.norm2 = norm_layer(dim)\n mlp_hidden_dim = int(dim * mlp_ratio)\n self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)\n\n if init_values is not None and init_values > 0:\n self.gamma_1 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)\n self.gamma_2 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)\n else:\n self.gamma_1, self.gamma_2 = None, None\n\n def forward(self, x, rel_pos_bias=None):\n if self.gamma_1 is None:\n x = x + self.drop_path(self.attn(self.norm1(x), rel_pos_bias=rel_pos_bias))\n x = x + self.drop_path(self.mlp(self.norm2(x)))\n else:\n x = x + self.drop_path(self.gamma_1 * self.attn(self.norm1(x), rel_pos_bias=rel_pos_bias))\n x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x)))\n return x\n\n\nclass PatchEmbed(nn.Module):\n \"\"\" Image to Patch Embedding\n \"\"\"\n def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):\n super().__init__()\n img_size = to_2tuple(img_size)\n patch_size = to_2tuple(patch_size)\n num_patches = (img_size[1] // patch_size[1]) * (img_size[0] // patch_size[0])\n self.patch_shape = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])\n self.img_size = img_size\n self.patch_size = patch_size\n self.num_patches = num_patches\n\n self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)\n\n def forward(self, x, **kwargs):\n B, C, H, W = x.shape\n # FIXME look at relaxing size constraints\n assert H == self.img_size[0] and W == self.img_size[1], \\\n f\"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]}).\"\n x = self.proj(x).flatten(2).transpose(1, 2)\n return x\n\n\nclass RelativePositionBias(nn.Module):\n\n def __init__(self, window_size, num_heads):\n super().__init__()\n self.window_size = window_size\n self.num_relative_distance = (2 * window_size[0] - 1) * (2 * window_size[1] - 1) + 3\n self.relative_position_bias_table = nn.Parameter(\n torch.zeros(self.num_relative_distance, num_heads)) # 2*Wh-1 * 2*Ww-1, nH\n # cls to token & token 2 cls & cls to cls\n\n # get pair-wise relative position index for each token inside the window\n coords_h = torch.arange(window_size[0])\n coords_w = torch.arange(window_size[1])\n coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww\n coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww\n relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww\n relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2\n relative_coords[:, :, 0] += window_size[0] - 1 # shift to start from 0\n relative_coords[:, :, 1] += window_size[1] - 1\n relative_coords[:, :, 0] *= 2 * window_size[1] - 1\n relative_position_index = \\\n torch.zeros(size=(window_size[0] * window_size[1] + 1,) * 2, dtype=relative_coords.dtype)\n relative_position_index[1:, 1:] = relative_coords.sum(-1) # Wh*Ww, Wh*Ww\n relative_position_index[0, 0:] = self.num_relative_distance - 3\n relative_position_index[0:, 0] = self.num_relative_distance - 2\n relative_position_index[0, 0] = self.num_relative_distance - 1\n\n self.register_buffer(\"relative_position_index\", relative_position_index)\n\n # trunc_normal_(self.relative_position_bias_table, std=.02)\n\n def forward(self):\n relative_position_bias = \\\n self.relative_position_bias_table[self.relative_position_index.view(-1)].view(\n self.window_size[0] * self.window_size[1] + 1,\n self.window_size[0] * self.window_size[1] + 1, -1) # Wh*Ww,Wh*Ww,nH\n return relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww\n\n\nclass VisionTransformer(nn.Module):\n \"\"\" Vision Transformer with support for patch or hybrid CNN input stage\n \"\"\"\n def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,\n num_heads=12, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop_rate=0., attn_drop_rate=0.,\n drop_path_rate=0., norm_layer=nn.LayerNorm, init_values=None,\n use_abs_pos_emb=True, use_rel_pos_bias=False, use_shared_rel_pos_bias=False,\n use_mean_pooling=True, init_scale=0.001, use_checkpoint=False):\n super().__init__()\n self.image_size = img_size\n self.num_classes = num_classes\n self.num_features = self.embed_dim = embed_dim # num_features for consistency with other models\n\n self.patch_embed = PatchEmbed(\n img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)\n num_patches = self.patch_embed.num_patches\n\n self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))\n if use_abs_pos_emb:\n self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))\n else:\n self.pos_embed = None\n self.pos_drop = nn.Dropout(p=drop_rate)\n\n if use_shared_rel_pos_bias:\n self.rel_pos_bias = RelativePositionBias(window_size=self.patch_embed.patch_shape, num_heads=num_heads)\n else:\n self.rel_pos_bias = None\n self.use_checkpoint = use_checkpoint\n \n dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)] # stochastic depth decay rule\n self.use_rel_pos_bias = use_rel_pos_bias\n self.blocks = nn.ModuleList([\n Block(\n dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,\n drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,\n init_values=init_values, window_size=self.patch_embed.patch_shape if use_rel_pos_bias else None)\n for i in range(depth)])\n# self.norm = nn.Identity() if use_mean_pooling else norm_layer(embed_dim)\n# self.fc_norm = norm_layer(embed_dim) if use_mean_pooling else None\n# self.head = nn.Linear(embed_dim, num_classes) if num_classes > 0 else nn.Identity()\n\n if self.pos_embed is not None:\n trunc_normal_(self.pos_embed, std=.02)\n trunc_normal_(self.cls_token, std=.02)\n # trunc_normal_(self.mask_token, std=.02)\n# if isinstance(self.head, nn.Linear):\n# trunc_normal_(self.head.weight, std=.02)\n self.apply(self._init_weights)\n self.fix_init_weight()\n# if isinstance(self.head, nn.Linear):\n# self.head.weight.data.mul_(init_scale)\n# self.head.bias.data.mul_(init_scale)\n\n def fix_init_weight(self):\n def rescale(param, layer_id):\n param.div_(math.sqrt(2.0 * layer_id))\n\n for layer_id, layer in enumerate(self.blocks):\n rescale(layer.attn.proj.weight.data, layer_id + 1)\n rescale(layer.mlp.fc2.weight.data, layer_id + 1)\n\n def _init_weights(self, m):\n if isinstance(m, nn.Linear):\n trunc_normal_(m.weight, std=.02)\n if isinstance(m, nn.Linear) and m.bias is not None:\n nn.init.constant_(m.bias, 0)\n elif isinstance(m, nn.LayerNorm):\n nn.init.constant_(m.bias, 0)\n nn.init.constant_(m.weight, 1.0)\n\n def get_classifier(self):\n return self.head\n\n def reset_classifier(self, num_classes, global_pool=''):\n self.num_classes = num_classes\n self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()\n\n def forward_features(self, x):\n x = self.patch_embed(x)\n batch_size, seq_len, _ = x.size()\n\n cls_tokens = self.cls_token.expand(batch_size, -1, -1) # stole cls_tokens impl from Phil Wang, thanks\n x = torch.cat((cls_tokens, x), dim=1)\n if self.pos_embed is not None:\n x = x + self.pos_embed\n x = self.pos_drop(x)\n\n rel_pos_bias = self.rel_pos_bias() if self.rel_pos_bias is not None else None\n for blk in self.blocks:\n if self.use_checkpoint:\n x = checkpoint.checkpoint(blk, x, rel_pos_bias)\n else:\n x = blk(x, rel_pos_bias)\n return x\n# x = self.norm(x)\n\n# if self.fc_norm is not None:\n# t = x[:, 1:, :]\n# return self.fc_norm(t.mean(1))\n# else:\n# return x[:, 0]\n\n def forward(self, x):\n x = self.forward_features(x)\n# x = self.head(x)\n return x\n\n def get_intermediate_layers(self, x):\n x = self.patch_embed(x)\n batch_size, seq_len, _ = x.size()\n\n cls_tokens = self.cls_token.expand(batch_size, -1, -1) # stole cls_tokens impl from Phil Wang, thanks\n x = torch.cat((cls_tokens, x), dim=1)\n if self.pos_embed is not None:\n x = x + self.pos_embed\n x = self.pos_drop(x)\n\n features = []\n rel_pos_bias = self.rel_pos_bias() if self.rel_pos_bias is not None else None\n for blk in self.blocks:\n x = blk(x, rel_pos_bias)\n features.append(x)\n\n return features\n \n \ndef interpolate_pos_embed(model, checkpoint_model):\n if 'pos_embed' in checkpoint_model:\n pos_embed_checkpoint = checkpoint_model['pos_embed'].float()\n embedding_size = pos_embed_checkpoint.shape[-1]\n num_patches = model.patch_embed.num_patches\n num_extra_tokens = model.pos_embed.shape[-2] - num_patches\n # height (== width) for the checkpoint position embedding\n orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)\n # height (== width) for the new position embedding\n new_size = int(num_patches ** 0.5)\n # class_token and dist_token are kept unchanged\n if orig_size != new_size:\n print(\"Position interpolate from %dx%d to %dx%d\" % (orig_size, orig_size, new_size, new_size))\n extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]\n # only the position tokens are interpolated\n pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]\n pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)\n pos_tokens = torch.nn.functional.interpolate(\n pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)\n pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)\n new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)\n checkpoint_model['pos_embed'] = new_pos_embed\n \n \ndef convert_weights_to_fp16(model: nn.Module):\n \"\"\"Convert applicable model parameters to fp16\"\"\"\n\n def _convert_weights_to_fp16(l):\n if isinstance(l, (nn.Conv1d, nn.Conv2d, nn.Linear)):\n l.weight.data = l.weight.data.half()\n if l.bias is not None:\n l.bias.data = l.bias.data.half()\n\n# if isinstance(l, (nn.MultiheadAttention, Attention)):\n# for attr in [*[f\"{s}_proj_weight\" for s in [\"in\", \"q\", \"k\", \"v\"]], \"in_proj_bias\", \"bias_k\", \"bias_v\"]:\n# tensor = getattr(l, attr)\n# if tensor is not None:\n# tensor.data = tensor.data.half()\n\n model.apply(_convert_weights_to_fp16)\n \n \ndef create_eva_vit_g(img_size=224,drop_path_rate=0.4,use_checkpoint=False,precision=\"fp16\"):\n model = VisionTransformer(\n img_size=img_size,\n patch_size=14,\n use_mean_pooling=False,\n embed_dim=1408,\n depth=39,\n num_heads=1408//88,\n mlp_ratio=4.3637,\n qkv_bias=True,\n drop_path_rate=drop_path_rate,\n norm_layer=partial(nn.LayerNorm, eps=1e-6),\n use_checkpoint=use_checkpoint,\n ) \n url = \"https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/BLIP2/eva_vit_g.pth\"\n cached_file = download_cached_file(\n url, check_hash=False, progress=True\n )\n state_dict = torch.load(cached_file, map_location=\"cpu\") \n interpolate_pos_embed(model,state_dict)\n \n incompatible_keys = model.load_state_dict(state_dict, strict=False)\n# print(incompatible_keys)\n \n if precision == \"fp16\":\n# model.to(\"cuda\") \n convert_weights_to_fp16(model)\n return model" }, { "path": "minigpt4/models/minigpt4.py", "content": "import logging\nimport random\n\nimport torch\nfrom torch.cuda.amp import autocast as autocast\nimport torch.nn as nn\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.models.base_model import disabled_train\nfrom minigpt4.models.minigpt_base import MiniGPTBase\nfrom minigpt4.models.Qformer import BertConfig, BertLMHeadModel\n\n\n@registry.register_model(\"minigpt4\")\nclass MiniGPT4(MiniGPTBase):\n \"\"\"\n MiniGPT-4 model\n \"\"\"\n\n PRETRAINED_MODEL_CONFIG_DICT = {\n \"pretrain_vicuna0\": \"configs/models/minigpt4_vicuna0.yaml\",\n \"pretrain_llama2\": \"configs/models/minigpt4_llama2.yaml\",\n }\n\n def __init__(\n self,\n vit_model=\"eva_clip_g\",\n q_former_model=\"https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/BLIP2/blip2_pretrained_flant5xxl.pth\",\n img_size=224,\n drop_path_rate=0,\n use_grad_checkpoint=False,\n vit_precision=\"fp16\",\n freeze_vit=True,\n has_qformer=True,\n freeze_qformer=True,\n num_query_token=32,\n llama_model=\"\",\n prompt_path=\"\",\n prompt_template=\"\",\n max_txt_len=32,\n end_sym='\\n',\n low_resource=False, # use 8 bit and put vit in cpu\n device_8bit=0, # the device of 8bit model should be set when loading and cannot be changed anymore.\n ):\n super().__init__(\n vit_model=vit_model,\n img_size=img_size,\n drop_path_rate=drop_path_rate,\n use_grad_checkpoint=use_grad_checkpoint,\n vit_precision=vit_precision,\n freeze_vit=freeze_vit,\n llama_model=llama_model,\n max_txt_len=max_txt_len,\n end_sym=end_sym,\n low_resource=low_resource,\n device_8bit=device_8bit,\n )\n\n self.has_qformer = has_qformer\n if self.has_qformer:\n print('Loading Q-Former')\n self.Qformer, self.query_tokens = self.init_Qformer(\n num_query_token, self.visual_encoder.num_features, freeze_qformer\n )\n self.load_from_pretrained(url_or_filename=q_former_model) # load q-former weights here\n\n img_f_dim = self.Qformer.config.hidden_size\n print('Loading Q-Former Done')\n else:\n img_f_dim = self.visual_encoder.num_features * 4\n print('Do not use Q-Former here.')\n\n self.llama_proj = nn.Linear(\n img_f_dim, self.llama_model.config.hidden_size\n )\n\n if prompt_path:\n with open(prompt_path, 'r') as f:\n raw_prompts = f.read().splitlines()\n filted_prompts = [raw_prompt for raw_prompt in raw_prompts if \"\" in raw_prompt]\n self.prompt_list = [prompt_template.format(p) for p in filted_prompts]\n print('Load {} training prompts'.format(len(self.prompt_list)))\n print('Prompt Example \\n{}'.format(random.choice(self.prompt_list)))\n else:\n self.prompt_list = []\n\n @classmethod\n def init_Qformer(cls, num_query_token, vision_width, freeze):\n encoder_config = BertConfig.from_pretrained(\"bert-base-uncased\")\n encoder_config.encoder_width = vision_width\n # insert cross-attention layer every other block\n encoder_config.add_cross_attention = True\n encoder_config.cross_attention_freq = 2\n encoder_config.query_length = num_query_token\n Qformer = BertLMHeadModel(config=encoder_config)\n query_tokens = nn.Parameter(\n torch.zeros(1, num_query_token, encoder_config.hidden_size)\n )\n query_tokens.data.normal_(mean=0.0, std=encoder_config.initializer_range)\n\n Qformer.cls = None\n Qformer.bert.embeddings.word_embeddings = None\n Qformer.bert.embeddings.position_embeddings = None\n for layer in Qformer.bert.encoder.layer:\n layer.output = None\n layer.intermediate = None\n\n if freeze:\n for name, param in Qformer.named_parameters():\n param.requires_grad = False\n Qformer = Qformer.eval()\n Qformer.train = disabled_train\n query_tokens.requires_grad = False\n logging.info(\"freeze Qformer\")\n\n return Qformer, query_tokens\n\n def encode_img(self, image):\n device = image.device\n\n if len(image.shape) > 4:\n image = image.reshape(-1, *image.shape[-3:])\n\n with self.maybe_autocast():\n image_embeds = self.ln_vision(self.visual_encoder(image)).to(device)\n if self.has_qformer:\n image_atts = torch.ones(image_embeds.size()[:-1], dtype=torch.long).to(device)\n\n query_tokens = self.query_tokens.expand(image_embeds.shape[0], -1, -1)\n query_output = self.Qformer.bert(\n query_embeds=query_tokens,\n encoder_hidden_states=image_embeds,\n encoder_attention_mask=image_atts,\n return_dict=True,\n )\n\n inputs_llama = self.llama_proj(query_output.last_hidden_state)\n else:\n image_embeds = image_embeds[:, 1:, :]\n bs, pn, hs = image_embeds.shape\n image_embeds = image_embeds.view(bs, int(pn / 4), int(hs * 4))\n\n inputs_llama = self.llama_proj(image_embeds)\n atts_llama = torch.ones(inputs_llama.size()[:-1], dtype=torch.long).to(image.device)\n return inputs_llama, atts_llama\n\n @classmethod\n def from_config(cls, cfg):\n vit_model = cfg.get(\"vit_model\", \"eva_clip_g\")\n q_former_model = cfg.get(\"q_former_model\", \"https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/BLIP2/blip2_pretrained_flant5xxl.pth\")\n img_size = cfg.get(\"image_size\")\n num_query_token = cfg.get(\"num_query_token\")\n llama_model = cfg.get(\"llama_model\")\n\n drop_path_rate = cfg.get(\"drop_path_rate\", 0)\n use_grad_checkpoint = cfg.get(\"use_grad_checkpoint\", False)\n vit_precision = cfg.get(\"vit_precision\", \"fp16\")\n freeze_vit = cfg.get(\"freeze_vit\", True)\n has_qformer = cfg.get(\"has_qformer\", True)\n freeze_qformer = cfg.get(\"freeze_qformer\", True)\n low_resource = cfg.get(\"low_resource\", False)\n device_8bit = cfg.get(\"device_8bit\", 0)\n\n prompt_path = cfg.get(\"prompt_path\", \"\")\n prompt_template = cfg.get(\"prompt_template\", \"\")\n max_txt_len = cfg.get(\"max_txt_len\", 32)\n end_sym = cfg.get(\"end_sym\", '\\n')\n\n model = cls(\n vit_model=vit_model,\n q_former_model=q_former_model,\n img_size=img_size,\n drop_path_rate=drop_path_rate,\n use_grad_checkpoint=use_grad_checkpoint,\n vit_precision=vit_precision,\n freeze_vit=freeze_vit,\n has_qformer=has_qformer,\n freeze_qformer=freeze_qformer,\n num_query_token=num_query_token,\n llama_model=llama_model,\n prompt_path=prompt_path,\n prompt_template=prompt_template,\n max_txt_len=max_txt_len,\n end_sym=end_sym,\n low_resource=low_resource,\n device_8bit=device_8bit,\n )\n\n ckpt_path = cfg.get(\"ckpt\", \"\") # load weights of MiniGPT-4\n if ckpt_path:\n print(\"Load MiniGPT-4 Checkpoint: {}\".format(ckpt_path))\n ckpt = torch.load(ckpt_path, map_location=\"cpu\")\n msg = model.load_state_dict(ckpt['model'], strict=False)\n\n return model\n" }, { "path": "minigpt4/models/minigpt_base.py", "content": "import logging\nimport random\n\nimport torch\nfrom torch.cuda.amp import autocast as autocast\nimport torch.nn as nn\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.models.base_model import BaseModel\nfrom transformers import StoppingCriteria, StoppingCriteriaList\n\nfrom minigpt4.conversation.conversation import StoppingCriteriaSub\n\nclass MiniGPTBase(BaseModel):\n \"\"\"\n Base class for MiniGPT-4 and MiniGPT-v2\n \"\"\"\n\n def __init__(\n self,\n vit_model=\"eva_clip_g\",\n img_size=224,\n drop_path_rate=0,\n use_grad_checkpoint=False,\n vit_precision=\"fp16\",\n freeze_vit=True,\n llama_model=\"\",\n max_txt_len=32,\n max_context_len=3800,\n prompt_template=\"\",\n end_sym='\\n',\n low_resource=False, # use 8 bit and put vit in cpu\n device_8bit=0, # the device of 8bit model should be set when loading and cannot be changed anymore.\n lora_r=0, # lora_r means lora is not used\n lora_target_modules=[\"q_proj\", \"v_proj\"],\n lora_alpha=16,\n lora_dropout=0.05,\n ):\n super().__init__()\n\n self.llama_model, self.llama_tokenizer = self.init_llm(\n llama_model_path=llama_model,\n low_resource=low_resource,\n low_res_device=device_8bit,\n lora_r=lora_r,\n lora_target_modules=lora_target_modules,\n lora_alpha=lora_alpha,\n lora_dropout=lora_dropout,\n )\n\n self.visual_encoder, self.ln_vision = self.init_vision_encoder(\n vit_model, img_size, drop_path_rate, use_grad_checkpoint, vit_precision, freeze_vit\n )\n\n self.max_txt_len = max_txt_len\n self.max_context_len = max_context_len\n self.end_sym = end_sym\n\n self.prompt_template = prompt_template\n self.prompt_list = []\n\n def vit_to_cpu(self):\n self.ln_vision.to(\"cpu\")\n self.ln_vision.float()\n self.visual_encoder.to(\"cpu\")\n self.visual_encoder.float()\n\n def get_context_emb(self, prompt, img_list):\n device = img_list[0].device\n prompt_segs = prompt.split('')\n assert len(prompt_segs) == len(img_list) + 1, \"Unmatched numbers of image placeholders and images.\"\n seg_tokens = [\n self.llama_tokenizer(\n seg, return_tensors=\"pt\", add_special_tokens=i==0).to(device).input_ids # only add bos to the first seg\n for i, seg in enumerate(prompt_segs)\n ]\n seg_embs = [self.embed_tokens(seg_t) for seg_t in seg_tokens]\n\n mixed_embs = [emb for pair in zip(seg_embs[:-1], img_list) for emb in pair] + [seg_embs[-1]]\n mixed_embs = torch.cat(mixed_embs, dim=1)\n return mixed_embs\n\n def prompt_wrap(self, img_embeds, atts_img, prompts, lengths=None):\n if prompts is None or len(prompts) == 0:\n # prompts is not provided, just return the original image embedding\n return img_embeds, atts_img\n elif img_embeds is None:\n # prompt is provided but there is no image embedding. return the prompt embedding in right padding\n self.llama_tokenizer.padding_side = \"right\"\n prompt_tokens = self.llama_tokenizer(\n prompts,\n return_tensors=\"pt\",\n padding=\"longest\",\n add_special_tokens=False\n ).to(self.device)\n prompt_embeds = self.embed_tokens(prompt_tokens.input_ids)\n atts_prompt = prompt_tokens.attention_mask\n return prompt_embeds, atts_prompt\n else:\n # return the multi-modal embedding in right padding\n emb_lists = []\n if isinstance(prompts, str):\n prompts = [prompts] * len(img_embeds)\n\n for idx, (each_img_embed, each_prompt) in enumerate(zip(img_embeds, prompts)):\n pn = each_img_embed.shape[-2]\n if lengths is not None:\n each_img_embed = each_img_embed.reshape(-1, each_img_embed.shape[-1])\n each_img_embed = each_img_embed[:lengths[idx] * pn]\n p_segs = each_prompt.split('')\n interleave_emb = []\n for idx, seg in enumerate(p_segs[:-1]):\n p_tokens = self.llama_tokenizer(\n seg, return_tensors=\"pt\", add_special_tokens=False).to(img_embeds.device)\n p_embed = self.embed_tokens(p_tokens.input_ids)\n interleave_emb.append(torch.cat([p_embed, each_img_embed[None][:, idx * pn:(idx + 1) * pn]], dim=1))\n wrapped_emb = torch.cat(interleave_emb, dim=1)\n p_tokens = self.llama_tokenizer(\n p_segs[-1], return_tensors=\"pt\", add_special_tokens=False).to(img_embeds.device)\n p_embed = self.embed_tokens(p_tokens.input_ids)\n wrapped_emb = torch.cat([wrapped_emb, p_embed], dim=1)\n emb_lists.append(wrapped_emb)\n\n emb_lens = [emb.shape[1] for emb in emb_lists]\n pad_emb = self.embed_tokens(torch.tensor(self.llama_tokenizer.pad_token_id, device=img_embeds.device))\n\n max_length = max(emb_lens) if max(emb_lens) < self.max_context_len else self.max_context_len\n wrapped_embs = pad_emb.expand(len(emb_lens), max_length, -1).clone()\n wrapped_atts = torch.zeros([len(emb_lens), max_length], dtype=torch.int, device=img_embeds.device)\n \n for i, emb in enumerate(emb_lists):\n length = emb_lens[i] if emb_lens[i] < self.max_context_len else self.max_context_len\n wrapped_embs[i, :length] = emb[:, :length]\n wrapped_atts[i, :length] = 1\n return wrapped_embs, wrapped_atts\n\n def concat_emb_input_output(self, input_embs, input_atts, output_embs, output_atts):\n \"\"\"\n Concatenate the batched input embedding and batched output embedding together.\n Both the input and the output embedding should be right padded.\n \"\"\"\n input_lens = []\n cat_embs = []\n cat_atts = []\n for i in range(input_embs.size(0)):\n input_len = input_atts[i].sum()\n input_lens.append(input_len)\n cat_embs.append(\n torch.cat([\n input_embs[i][:input_len],\n output_embs[i],\n input_embs[i][input_len:]\n ])\n )\n cat_atts.append(\n torch.cat([\n input_atts[i][:input_len],\n output_atts[i],\n input_atts[i][input_len:]\n ])\n )\n cat_embs = torch.stack(cat_embs)\n cat_atts = torch.stack(cat_atts)\n return cat_embs, cat_atts, input_lens\n\n def tokenize_conversation(self, conv_q, conv_a):\n \"\"\"concatenate conversation and make sure the model is only trained to regress the answer\"\"\"\n\n to_regress_token_ids_list = []\n targets_list = []\n\n batch_size = len(conv_q)\n for batch_idx in range(batch_size):\n questions, answers = conv_q[batch_idx], conv_a[batch_idx]\n questions = [self.llama_tokenizer(self.llama_tokenizer.bos_token + q,\n return_tensors=\"pt\",\n add_special_tokens=False).to(self.device) for q in questions[1:]] # the first question is handled in the prompt wrap function, skip it\n answers = [self.llama_tokenizer(a + self.end_sym,\n return_tensors=\"pt\",\n add_special_tokens=False).to(self.device) for a in answers]\n cur_id = []\n cur_target = []\n for i in range(len(questions)):\n cur_id.append(answers[i].input_ids)\n cur_target.append(answers[i].input_ids)\n cur_id.append(questions[i].input_ids)\n cur_target.append(torch.ones_like(questions[i].input_ids) * -100)\n\n cur_id.append(answers[-1].input_ids)\n cur_target.append(answers[-1].input_ids)\n\n cur_id = torch.cat(cur_id, dim=1)\n cur_target = torch.cat(cur_target, dim=1)\n to_regress_token_ids_list.append(cur_id)\n targets_list.append(cur_target)\n\n max_len = min(max([target.shape[1] for target in targets_list]), self.max_txt_len)\n to_regress_token_ids = torch.ones([batch_size, max_len],\n dtype=cur_id.dtype, device=self.device) * self.llama_tokenizer.pad_token_id\n targets = torch.ones([batch_size, max_len],\n dtype=cur_id.dtype, device=self.device) * -100\n for batch_idx in range(batch_size):\n cur_len = to_regress_token_ids_list[batch_idx].shape[1]\n to_regress_token_ids[batch_idx, :cur_len] = to_regress_token_ids_list[batch_idx][0, :max_len]\n targets[batch_idx, :cur_len] = targets_list[batch_idx][0, :max_len]\n\n to_regress_token_attn = (to_regress_token_ids != self.llama_tokenizer.pad_token_id).to(torch.int)\n\n return to_regress_token_ids, to_regress_token_attn, targets\n\n def preparing_embedding(self, samples):\n ### prepare input tokens\n if 'image' in samples:\n img_embeds, img_atts = self.encode_img(samples[\"image\"])\n else:\n img_embeds = img_atts = None\n\n if 'conv_q' in samples:\n # handeling conversation datasets\n conv_q, conv_a = samples['conv_q'], samples['conv_a']\n\n connect_sym = samples['connect_sym'][0]\n conv_q = [q.split(connect_sym)for q in conv_q]\n conv_a = [a.split(connect_sym) for a in conv_a]\n\n conv_q = [[self.prompt_template.format(item) for item in items] for items in conv_q]\n\n cond_embeds, cond_atts = self.prompt_wrap(img_embeds, img_atts, [q[0] for q in conv_q])\n regress_token_ids, regress_atts, part_targets = self.tokenize_conversation(conv_q, conv_a)\n\n else:\n if \"instruction_input\" in samples:\n instruction = samples[\"instruction_input\"]\n elif self.prompt_list:\n instruction = random.choice(self.prompt_list)\n else:\n instruction = None\n\n if hasattr(self, 'chat_template') and self.chat_template:\n instruction = [self.prompt_template.format(instruct) for instruct in instruction]\n\n if 'length' in samples:\n # the input is a image train (like videos)\n bsz, pn, hs = img_embeds.shape\n img_embeds = img_embeds.reshape(len(samples['image']), -1, pn, hs)\n cond_embeds, cond_atts = self.prompt_wrap(img_embeds, img_atts, instruction, samples['length'])\n else:\n cond_embeds, cond_atts = self.prompt_wrap(img_embeds, img_atts, instruction)\n\n ### prepare target tokens\n self.llama_tokenizer.padding_side = \"right\"\n text = [t + self.end_sym for t in samples[\"answer\"]]\n\n regress_tokens = self.llama_tokenizer(\n text,\n return_tensors=\"pt\",\n padding=\"longest\",\n truncation=True,\n max_length=self.max_txt_len,\n add_special_tokens=False\n ).to(self.device)\n\n regress_token_ids = regress_tokens.input_ids\n regress_atts = regress_tokens.attention_mask\n part_targets = regress_token_ids.masked_fill(\n regress_token_ids == self.llama_tokenizer.pad_token_id, -100\n )\n\n regress_embeds = self.embed_tokens(regress_token_ids)\n\n return cond_embeds, cond_atts, regress_embeds, regress_atts, part_targets\n\n def forward(self, samples, reduction='mean'):\n # prepare the embedding to condition and the embedding to regress\n cond_embeds, cond_atts, regress_embeds, regress_atts, part_targets = \\\n self.preparing_embedding(samples)\n\n # concat the embedding to condition and the embedding to regress\n inputs_embeds, attention_mask, input_lens = \\\n self.concat_emb_input_output(cond_embeds, cond_atts, regress_embeds, regress_atts)\n\n # get bos token embedding\n bos = torch.ones_like(part_targets[:, :1]) * self.llama_tokenizer.bos_token_id\n bos_embeds = self.embed_tokens(bos)\n bos_atts = cond_atts[:, :1]\n\n # add bos token at the begining\n inputs_embeds = torch.cat([bos_embeds, inputs_embeds], dim=1)\n attention_mask = torch.cat([bos_atts, attention_mask], dim=1)\n\n # ensemble the final targets\n targets = torch.ones([inputs_embeds.shape[0], inputs_embeds.shape[1]],\n dtype=torch.long).to(self.device).fill_(-100)\n\n for i, target in enumerate(part_targets):\n targets[i, input_lens[i]+1:input_lens[i]+len(target)+1] = target # plus 1 for bos\n\n with self.maybe_autocast():\n outputs = self.llama_model(\n inputs_embeds=inputs_embeds,\n attention_mask=attention_mask,\n return_dict=True,\n labels=targets,\n reduction=reduction\n )\n loss = outputs.loss\n\n return {\"loss\": loss}\n\n def embed_tokens(self, token_ids):\n if hasattr(self.llama_model.base_model, 'model'): ## lora wrapped model\n embeds = self.llama_model.base_model.model.model.embed_tokens(token_ids)\n else:\n embeds = self.llama_model.base_model.embed_tokens(token_ids)\n return embeds\n\n @torch.no_grad()\n def generate(\n self,\n images,\n texts,\n num_beams=1,\n max_new_tokens=20,\n min_length=1,\n top_p=0.9,\n repetition_penalty=1,\n length_penalty=1,\n temperature=1,\n do_sample=False,\n stop_words_ids=[2],\n ):\n '''\n function for generate test use\n '''\n\n stopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(\n stops=[torch.tensor([i]).to(self.device) for i in stop_words_ids])])\n\n img_embeds, atts_img = self.encode_img(images.to(self.device))\n image_lists = [[image_emb[None]] for image_emb in img_embeds]\n\n batch_embs = [self.get_context_emb(text, img_list) for text, img_list in zip(texts, image_lists)]\n\n batch_size = len(batch_embs)\n max_len = max([emb.shape[1] for emb in batch_embs])\n emb_dim = batch_embs[0].shape[2]\n dtype = batch_embs[0].dtype\n device = batch_embs[0].device\n\n embs = torch.zeros([batch_size, max_len, emb_dim], dtype=dtype, device=device)\n attn_mask = torch.zeros([batch_size, max_len], dtype=torch.int, device=device)\n for i, emb in enumerate(batch_embs):\n emb_len = emb.shape[1]\n embs[i, -emb_len:] = emb[0]\n attn_mask[i, -emb_len:] = 1\n\n with self.maybe_autocast():\n outputs = self.llama_model.generate(\n inputs_embeds=embs,\n attention_mask=attn_mask,\n max_new_tokens=max_new_tokens,\n num_beams=num_beams,\n length_penalty=length_penalty,\n temperature=temperature,\n do_sample=do_sample,\n min_length=min_length,\n top_p=top_p,\n repetition_penalty=repetition_penalty,\n # stopping_criteria=stopping_criteria,\n )\n\n # with self.maybe_autocast():\n # outputs = self.llama_model.generate(\n # inputs_embeds=embs,\n # attention_mask=attn_mask,\n # max_new_tokens=max_new_tokens,\n # num_beams=num_beams,\n # do_sample=do_sample,\n # # stopping_criteria=stopping_criteria,\n # )\n answers = []\n for output_token in outputs:\n if output_token[0] == 0:\n output_token = output_token[1:]\n output_texts = self.llama_tokenizer.decode(output_token, skip_special_tokens=True)\n output_texts = output_texts.split('')[0] # remove the stop sign \n output_texts = output_texts.replace(\"\", \"\")\n output_texts = output_texts.split(r'[/INST]')[-1].strip()\n answers.append(output_texts)\n\n return answers\n\n @torch.no_grad()\n def multi_select(self, images, texts, answers, num_cand=None):\n all_losses = []\n for answer in answers:\n choice_samples = {\n 'image': images,\n 'instruction_input': texts,\n 'answer': answer\n }\n loss = self.forward(choice_samples, reduction='none')['loss'].reshape(-1, 1)\n all_losses.append(loss)\n torch.cuda.empty_cache()\n all_losses = torch.cat(all_losses, dim=-1)\n if num_cand is not None:\n for i in range(all_losses.shape[0]):\n all_losses[i, num_cand[i]:] = 9999\n output_class_ranks = torch.argsort(all_losses, dim=-1)\n return output_class_ranks.tolist()\n" }, { "path": "minigpt4/models/minigpt_v2.py", "content": "import logging\nimport random\n\nimport torch\nfrom torch.cuda.amp import autocast as autocast\nimport torch.nn as nn\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.models.base_model import disabled_train\nfrom minigpt4.models.minigpt_base import MiniGPTBase\nfrom minigpt4.models.Qformer import BertConfig, BertLMHeadModel\n\n\n@registry.register_model(\"minigpt_v2\")\nclass MiniGPTv2(MiniGPTBase):\n \"\"\"\n MiniGPT-v2 model\n \"\"\"\n\n PRETRAINED_MODEL_CONFIG_DICT = {\n \"pretrain\": \"configs/models/minigpt_v2.yaml\",\n }\n\n def __init__(\n self,\n vit_model=\"eva_clip_g\",\n img_size=448,\n drop_path_rate=0,\n use_grad_checkpoint=False,\n vit_precision=\"fp16\",\n freeze_vit=True,\n llama_model=\"\",\n prompt_template='[INST] {} [/INST]',\n max_txt_len=300,\n end_sym='\\n',\n lora_r=64,\n lora_target_modules=[\"q_proj\", \"v_proj\"],\n lora_alpha=16,\n lora_dropout=0.05,\n chat_template=False,\n use_grad_checkpoint_llm=False,\n max_context_len=3800,\n low_resource=False, # use 8 bit and put vit in cpu\n device_8bit=0, # the device of 8bit model should be set when loading and cannot be changed anymore.\n ):\n super().__init__(\n vit_model=vit_model,\n img_size=img_size,\n drop_path_rate=drop_path_rate,\n use_grad_checkpoint=use_grad_checkpoint,\n vit_precision=vit_precision,\n freeze_vit=freeze_vit,\n llama_model=llama_model,\n max_txt_len=max_txt_len,\n max_context_len=max_context_len,\n end_sym=end_sym,\n prompt_template=prompt_template,\n low_resource=low_resource,\n device_8bit=device_8bit,\n lora_r=lora_r,\n lora_target_modules=lora_target_modules,\n lora_alpha=lora_alpha,\n lora_dropout=lora_dropout,\n )\n\n img_f_dim = self.visual_encoder.num_features * 4\n self.llama_proj = nn.Linear(\n img_f_dim, self.llama_model.config.hidden_size\n )\n self.chat_template = chat_template\n\n if use_grad_checkpoint_llm:\n self.llama_model.gradient_checkpointing_enable()\n\n def encode_img(self, image):\n device = image.device\n\n if len(image.shape) > 4:\n image = image.reshape(-1, *image.shape[-3:])\n\n with self.maybe_autocast():\n image_embeds = self.ln_vision(self.visual_encoder(image)).to(device)\n image_embeds = image_embeds[:, 1:, :]\n bs, pn, hs = image_embeds.shape\n image_embeds = image_embeds.view(bs, int(pn / 4), int(hs * 4))\n\n inputs_llama = self.llama_proj(image_embeds)\n atts_llama = torch.ones(inputs_llama.size()[:-1], dtype=torch.long).to(image.device)\n return inputs_llama, atts_llama\n\n @classmethod\n def from_config(cls, cfg):\n vit_model = cfg.get(\"vit_model\", \"eva_clip_g\")\n img_size = cfg.get(\"image_size\")\n llama_model = cfg.get(\"llama_model\")\n\n drop_path_rate = cfg.get(\"drop_path_rate\", 0)\n use_grad_checkpoint = cfg.get(\"use_grad_checkpoint\", False)\n vit_precision = cfg.get(\"vit_precision\", \"fp16\")\n freeze_vit = cfg.get(\"freeze_vit\", True)\n low_resource = cfg.get(\"low_resource\", False)\n\n prompt_template = cfg.get(\"prompt_template\", '[INST] {} [/INST]')\n max_txt_len = cfg.get(\"max_txt_len\", 300)\n end_sym = cfg.get(\"end_sym\", '\\n')\n\n lora_r = cfg.get(\"lora_r\", 64)\n lora_alpha = cfg.get(\"lora_alpha\", 16)\n chat_template = cfg.get(\"chat_template\", False)\n\n use_grad_checkpoint_llm = cfg.get(\"use_grad_checkpoint_llm\", False)\n max_context_len = cfg.get(\"max_context_len\", 3800)\n\n model = cls(\n vit_model=vit_model,\n img_size=img_size,\n drop_path_rate=drop_path_rate,\n use_grad_checkpoint=use_grad_checkpoint,\n vit_precision=vit_precision,\n freeze_vit=freeze_vit,\n llama_model=llama_model,\n prompt_template=prompt_template,\n max_txt_len=max_txt_len,\n low_resource=low_resource,\n end_sym=end_sym,\n lora_r=lora_r,\n lora_alpha=lora_alpha,\n chat_template=chat_template,\n use_grad_checkpoint_llm=use_grad_checkpoint_llm,\n max_context_len=max_context_len,\n )\n\n ckpt_path = cfg.get(\"ckpt\", \"\") # load weights of MiniGPT-4\n if ckpt_path:\n print(\"Load Minigpt-4-LLM Checkpoint: {}\".format(ckpt_path))\n ckpt = torch.load(ckpt_path, map_location=\"cpu\")\n msg = model.load_state_dict(ckpt['model'], strict=False)\n\n return model\n" }, { "path": "minigpt4/models/modeling_llama.py", "content": "import math\nfrom typing import List, Optional, Tuple, Union\n\nimport torch\nimport torch.nn.functional as F\nfrom torch.nn import CrossEntropyLoss\n\nfrom transformers.utils import add_start_docstrings_to_model_forward, replace_return_docstrings\nfrom transformers.modeling_outputs import CausalLMOutputWithPast\nfrom transformers.models.llama.modeling_llama import LLAMA_INPUTS_DOCSTRING, _CONFIG_FOR_DOC\nfrom transformers.models.llama.modeling_llama import LlamaForCausalLM as LlamaForCausalLMOrig\n\n\nclass LlamaForCausalLM(LlamaForCausalLMOrig):\n\n @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)\n @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)\n def forward(\n self,\n input_ids: torch.LongTensor = None,\n attention_mask: Optional[torch.Tensor] = None,\n position_ids: Optional[torch.LongTensor] = None,\n past_key_values: Optional[List[torch.FloatTensor]] = None,\n inputs_embeds: Optional[torch.FloatTensor] = None,\n labels: Optional[torch.LongTensor] = None,\n use_cache: Optional[bool] = None,\n output_attentions: Optional[bool] = None,\n output_hidden_states: Optional[bool] = None,\n return_dict: Optional[bool] = None,\n reduction: Optional[str] = \"mean\",\n ) -> Union[Tuple, CausalLMOutputWithPast]:\n r\"\"\"\n Args:\n labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):\n Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,\n config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored\n (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.\n\n Returns:\n\n Example:\n\n ```python\n >>> from transformers import AutoTokenizer, LlamaForCausalLM\n\n >>> model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)\n >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)\n\n >>> prompt = \"Hey, are you conscious? Can you talk to me?\"\n >>> inputs = tokenizer(prompt, return_tensors=\"pt\")\n\n >>> # Generate\n >>> generate_ids = model.generate(inputs.input_ids, max_length=30)\n >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]\n \"Hey, are you conscious? Can you talk to me?\\nI'm not conscious, but I can talk to you.\"\n ```\"\"\"\n\n output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions\n output_hidden_states = (\n output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states\n )\n return_dict = return_dict if return_dict is not None else self.config.use_return_dict\n\n # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)\n outputs = self.model(\n input_ids=input_ids,\n attention_mask=attention_mask,\n position_ids=position_ids,\n past_key_values=past_key_values,\n inputs_embeds=inputs_embeds,\n use_cache=use_cache,\n output_attentions=output_attentions,\n output_hidden_states=output_hidden_states,\n return_dict=return_dict,\n )\n\n hidden_states = outputs[0]\n if hasattr(self.config, 'pretraining_tp') and self.config.pretraining_tp > 1:\n lm_head_slices = self.lm_head.weight.split(self.vocab_size // self.config.pretraining_tp, dim=0)\n logits = [F.linear(hidden_states, lm_head_slices[i]) for i in range(self.config.pretraining_tp)]\n logits = torch.cat(logits, dim=-1)\n else:\n logits = self.lm_head(hidden_states)\n logits = logits.float()\n\n loss = None\n if labels is not None:\n # Shift so that tokens < n predict n\n shift_logits = logits[..., :-1, :].contiguous()\n shift_labels = labels[..., 1:].contiguous()\n # Flatten the tokens\n loss_fct = CrossEntropyLoss(reduction=reduction)\n shift_logits = shift_logits.view(-1, self.config.vocab_size)\n shift_labels = shift_labels.view(-1)\n # Enable model parallelism\n shift_labels = shift_labels.to(shift_logits.device)\n loss = loss_fct(shift_logits, shift_labels)\n if reduction == \"none\":\n loss = loss.view(logits.size(0), -1).mean(1)\n\n if not return_dict:\n output = (logits,) + outputs[1:]\n return (loss,) + output if loss is not None else output\n\n return CausalLMOutputWithPast(\n loss=loss,\n logits=logits,\n past_key_values=outputs.past_key_values,\n hidden_states=outputs.hidden_states,\n attentions=outputs.attentions,\n )\n" }, { "path": "minigpt4/processors/__init__.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nfrom minigpt4.processors.base_processor import BaseProcessor\nfrom minigpt4.processors.blip_processors import (\n Blip2ImageTrainProcessor,\n Blip2ImageEvalProcessor,\n BlipCaptionProcessor,\n)\n\nfrom minigpt4.common.registry import registry\n\n__all__ = [\n \"BaseProcessor\",\n \"Blip2ImageTrainProcessor\",\n \"Blip2ImageEvalProcessor\",\n \"BlipCaptionProcessor\",\n]\n\n\ndef load_processor(name, cfg=None):\n \"\"\"\n Example\n\n >>> processor = load_processor(\"alpro_video_train\", cfg=None)\n \"\"\"\n processor = registry.get_processor_class(name).from_config(cfg)\n\n return processor\n" }, { "path": "minigpt4/processors/base_processor.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nfrom omegaconf import OmegaConf\n\n\nclass BaseProcessor:\n def __init__(self):\n self.transform = lambda x: x\n return\n\n def __call__(self, item):\n return self.transform(item)\n\n @classmethod\n def from_config(cls, cfg=None):\n return cls()\n\n def build(self, **kwargs):\n cfg = OmegaConf.create(kwargs)\n\n return self.from_config(cfg)\n" }, { "path": "minigpt4/processors/blip_processors.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport re\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.processors.base_processor import BaseProcessor\nfrom minigpt4.processors.randaugment import RandomAugment\nfrom omegaconf import OmegaConf\nfrom torchvision import transforms\nfrom torchvision.transforms.functional import InterpolationMode\n\n\nclass BlipImageBaseProcessor(BaseProcessor):\n def __init__(self, mean=None, std=None):\n if mean is None:\n mean = (0.48145466, 0.4578275, 0.40821073)\n if std is None:\n std = (0.26862954, 0.26130258, 0.27577711)\n\n self.normalize = transforms.Normalize(mean, std)\n\n\n@registry.register_processor(\"blip_caption\")\nclass BlipCaptionProcessor(BaseProcessor):\n def __init__(self, prompt=\"\", max_words=50):\n self.prompt = prompt\n self.max_words = max_words\n\n def __call__(self, caption):\n caption = self.prompt + self.pre_caption(caption)\n\n return caption\n\n @classmethod\n def from_config(cls, cfg=None):\n if cfg is None:\n cfg = OmegaConf.create()\n\n prompt = cfg.get(\"prompt\", \"\")\n max_words = cfg.get(\"max_words\", 50)\n\n return cls(prompt=prompt, max_words=max_words)\n\n def pre_caption(self, caption):\n caption = re.sub(\n r\"([.!\\\"()*#:;~])\",\n \" \",\n caption.lower(),\n )\n caption = re.sub(\n r\"\\s{2,}\",\n \" \",\n caption,\n )\n caption = caption.rstrip(\"\\n\")\n caption = caption.strip(\" \")\n\n # truncate caption\n caption_words = caption.split(\" \")\n if len(caption_words) > self.max_words:\n caption = \" \".join(caption_words[: self.max_words])\n\n return caption\n\n\n@registry.register_processor(\"blip2_image_train\")\nclass Blip2ImageTrainProcessor(BlipImageBaseProcessor):\n def __init__(self, image_size=224, mean=None, std=None, min_scale=0.5, max_scale=1.0):\n super().__init__(mean=mean, std=std)\n\n self.transform = transforms.Compose(\n [\n transforms.Resize(\n (image_size,image_size),\n interpolation=InterpolationMode.BICUBIC,\n ),\n transforms.ToTensor(),\n self.normalize,\n ]\n )\n\n def __call__(self, item):\n return self.transform(item)\n\n @classmethod\n def from_config(cls, cfg=None):\n if cfg is None:\n cfg = OmegaConf.create()\n\n image_size = cfg.get(\"image_size\", 224)\n\n mean = cfg.get(\"mean\", None)\n std = cfg.get(\"std\", None)\n\n min_scale = cfg.get(\"min_scale\", 0.5)\n max_scale = cfg.get(\"max_scale\", 1.0)\n\n return cls(\n image_size=image_size,\n mean=mean,\n std=std,\n min_scale=min_scale,\n max_scale=max_scale,\n )\n\n\n@registry.register_processor(\"blip2_image_eval\")\nclass Blip2ImageEvalProcessor(BlipImageBaseProcessor):\n def __init__(self, image_size=224, mean=None, std=None):\n super().__init__(mean=mean, std=std)\n\n self.transform = transforms.Compose(\n [\n transforms.Resize(\n (image_size, image_size), interpolation=InterpolationMode.BICUBIC\n ),\n transforms.ToTensor(),\n self.normalize,\n ]\n )\n\n def __call__(self, item):\n return self.transform(item)\n\n @classmethod\n def from_config(cls, cfg=None):\n if cfg is None:\n cfg = OmegaConf.create()\n\n image_size = cfg.get(\"image_size\", 224)\n\n mean = cfg.get(\"mean\", None)\n std = cfg.get(\"std\", None)\n\n return cls(image_size=image_size, mean=mean, std=std)\n" }, { "path": "minigpt4/processors/randaugment.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport cv2\nimport numpy as np\n\nimport torch\n\n\n## aug functions\ndef identity_func(img):\n return img\n\n\ndef autocontrast_func(img, cutoff=0):\n \"\"\"\n same output as PIL.ImageOps.autocontrast\n \"\"\"\n n_bins = 256\n\n def tune_channel(ch):\n n = ch.size\n cut = cutoff * n // 100\n if cut == 0:\n high, low = ch.max(), ch.min()\n else:\n hist = cv2.calcHist([ch], [0], None, [n_bins], [0, n_bins])\n low = np.argwhere(np.cumsum(hist) > cut)\n low = 0 if low.shape[0] == 0 else low[0]\n high = np.argwhere(np.cumsum(hist[::-1]) > cut)\n high = n_bins - 1 if high.shape[0] == 0 else n_bins - 1 - high[0]\n if high <= low:\n table = np.arange(n_bins)\n else:\n scale = (n_bins - 1) / (high - low)\n offset = -low * scale\n table = np.arange(n_bins) * scale + offset\n table[table < 0] = 0\n table[table > n_bins - 1] = n_bins - 1\n table = table.clip(0, 255).astype(np.uint8)\n return table[ch]\n\n channels = [tune_channel(ch) for ch in cv2.split(img)]\n out = cv2.merge(channels)\n return out\n\n\ndef equalize_func(img):\n \"\"\"\n same output as PIL.ImageOps.equalize\n PIL's implementation is different from cv2.equalize\n \"\"\"\n n_bins = 256\n\n def tune_channel(ch):\n hist = cv2.calcHist([ch], [0], None, [n_bins], [0, n_bins])\n non_zero_hist = hist[hist != 0].reshape(-1)\n step = np.sum(non_zero_hist[:-1]) // (n_bins - 1)\n if step == 0:\n return ch\n n = np.empty_like(hist)\n n[0] = step // 2\n n[1:] = hist[:-1]\n table = (np.cumsum(n) // step).clip(0, 255).astype(np.uint8)\n return table[ch]\n\n channels = [tune_channel(ch) for ch in cv2.split(img)]\n out = cv2.merge(channels)\n return out\n\n\ndef rotate_func(img, degree, fill=(0, 0, 0)):\n \"\"\"\n like PIL, rotate by degree, not radians\n \"\"\"\n H, W = img.shape[0], img.shape[1]\n center = W / 2, H / 2\n M = cv2.getRotationMatrix2D(center, degree, 1)\n out = cv2.warpAffine(img, M, (W, H), borderValue=fill)\n return out\n\n\ndef solarize_func(img, thresh=128):\n \"\"\"\n same output as PIL.ImageOps.posterize\n \"\"\"\n table = np.array([el if el < thresh else 255 - el for el in range(256)])\n table = table.clip(0, 255).astype(np.uint8)\n out = table[img]\n return out\n\n\ndef color_func(img, factor):\n \"\"\"\n same output as PIL.ImageEnhance.Color\n \"\"\"\n ## implementation according to PIL definition, quite slow\n # degenerate = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)[:, :, np.newaxis]\n # out = blend(degenerate, img, factor)\n # M = (\n # np.eye(3) * factor\n # + np.float32([0.114, 0.587, 0.299]).reshape(3, 1) * (1. - factor)\n # )[np.newaxis, np.newaxis, :]\n M = np.float32(\n [[0.886, -0.114, -0.114], [-0.587, 0.413, -0.587], [-0.299, -0.299, 0.701]]\n ) * factor + np.float32([[0.114], [0.587], [0.299]])\n out = np.matmul(img, M).clip(0, 255).astype(np.uint8)\n return out\n\n\ndef contrast_func(img, factor):\n \"\"\"\n same output as PIL.ImageEnhance.Contrast\n \"\"\"\n mean = np.sum(np.mean(img, axis=(0, 1)) * np.array([0.114, 0.587, 0.299]))\n table = (\n np.array([(el - mean) * factor + mean for el in range(256)])\n .clip(0, 255)\n .astype(np.uint8)\n )\n out = table[img]\n return out\n\n\ndef brightness_func(img, factor):\n \"\"\"\n same output as PIL.ImageEnhance.Contrast\n \"\"\"\n table = (np.arange(256, dtype=np.float32) * factor).clip(0, 255).astype(np.uint8)\n out = table[img]\n return out\n\n\ndef sharpness_func(img, factor):\n \"\"\"\n The differences the this result and PIL are all on the 4 boundaries, the center\n areas are same\n \"\"\"\n kernel = np.ones((3, 3), dtype=np.float32)\n kernel[1][1] = 5\n kernel /= 13\n degenerate = cv2.filter2D(img, -1, kernel)\n if factor == 0.0:\n out = degenerate\n elif factor == 1.0:\n out = img\n else:\n out = img.astype(np.float32)\n degenerate = degenerate.astype(np.float32)[1:-1, 1:-1, :]\n out[1:-1, 1:-1, :] = degenerate + factor * (out[1:-1, 1:-1, :] - degenerate)\n out = out.astype(np.uint8)\n return out\n\n\ndef shear_x_func(img, factor, fill=(0, 0, 0)):\n H, W = img.shape[0], img.shape[1]\n M = np.float32([[1, factor, 0], [0, 1, 0]])\n out = cv2.warpAffine(\n img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR\n ).astype(np.uint8)\n return out\n\n\ndef translate_x_func(img, offset, fill=(0, 0, 0)):\n \"\"\"\n same output as PIL.Image.transform\n \"\"\"\n H, W = img.shape[0], img.shape[1]\n M = np.float32([[1, 0, -offset], [0, 1, 0]])\n out = cv2.warpAffine(\n img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR\n ).astype(np.uint8)\n return out\n\n\ndef translate_y_func(img, offset, fill=(0, 0, 0)):\n \"\"\"\n same output as PIL.Image.transform\n \"\"\"\n H, W = img.shape[0], img.shape[1]\n M = np.float32([[1, 0, 0], [0, 1, -offset]])\n out = cv2.warpAffine(\n img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR\n ).astype(np.uint8)\n return out\n\n\ndef posterize_func(img, bits):\n \"\"\"\n same output as PIL.ImageOps.posterize\n \"\"\"\n out = np.bitwise_and(img, np.uint8(255 << (8 - bits)))\n return out\n\n\ndef shear_y_func(img, factor, fill=(0, 0, 0)):\n H, W = img.shape[0], img.shape[1]\n M = np.float32([[1, 0, 0], [factor, 1, 0]])\n out = cv2.warpAffine(\n img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR\n ).astype(np.uint8)\n return out\n\n\ndef cutout_func(img, pad_size, replace=(0, 0, 0)):\n replace = np.array(replace, dtype=np.uint8)\n H, W = img.shape[0], img.shape[1]\n rh, rw = np.random.random(2)\n pad_size = pad_size // 2\n ch, cw = int(rh * H), int(rw * W)\n x1, x2 = max(ch - pad_size, 0), min(ch + pad_size, H)\n y1, y2 = max(cw - pad_size, 0), min(cw + pad_size, W)\n out = img.copy()\n out[x1:x2, y1:y2, :] = replace\n return out\n\n\n### level to args\ndef enhance_level_to_args(MAX_LEVEL):\n def level_to_args(level):\n return ((level / MAX_LEVEL) * 1.8 + 0.1,)\n\n return level_to_args\n\n\ndef shear_level_to_args(MAX_LEVEL, replace_value):\n def level_to_args(level):\n level = (level / MAX_LEVEL) * 0.3\n if np.random.random() > 0.5:\n level = -level\n return (level, replace_value)\n\n return level_to_args\n\n\ndef translate_level_to_args(translate_const, MAX_LEVEL, replace_value):\n def level_to_args(level):\n level = (level / MAX_LEVEL) * float(translate_const)\n if np.random.random() > 0.5:\n level = -level\n return (level, replace_value)\n\n return level_to_args\n\n\ndef cutout_level_to_args(cutout_const, MAX_LEVEL, replace_value):\n def level_to_args(level):\n level = int((level / MAX_LEVEL) * cutout_const)\n return (level, replace_value)\n\n return level_to_args\n\n\ndef solarize_level_to_args(MAX_LEVEL):\n def level_to_args(level):\n level = int((level / MAX_LEVEL) * 256)\n return (level,)\n\n return level_to_args\n\n\ndef none_level_to_args(level):\n return ()\n\n\ndef posterize_level_to_args(MAX_LEVEL):\n def level_to_args(level):\n level = int((level / MAX_LEVEL) * 4)\n return (level,)\n\n return level_to_args\n\n\ndef rotate_level_to_args(MAX_LEVEL, replace_value):\n def level_to_args(level):\n level = (level / MAX_LEVEL) * 30\n if np.random.random() < 0.5:\n level = -level\n return (level, replace_value)\n\n return level_to_args\n\n\nfunc_dict = {\n \"Identity\": identity_func,\n \"AutoContrast\": autocontrast_func,\n \"Equalize\": equalize_func,\n \"Rotate\": rotate_func,\n \"Solarize\": solarize_func,\n \"Color\": color_func,\n \"Contrast\": contrast_func,\n \"Brightness\": brightness_func,\n \"Sharpness\": sharpness_func,\n \"ShearX\": shear_x_func,\n \"TranslateX\": translate_x_func,\n \"TranslateY\": translate_y_func,\n \"Posterize\": posterize_func,\n \"ShearY\": shear_y_func,\n}\n\ntranslate_const = 10\nMAX_LEVEL = 10\nreplace_value = (128, 128, 128)\narg_dict = {\n \"Identity\": none_level_to_args,\n \"AutoContrast\": none_level_to_args,\n \"Equalize\": none_level_to_args,\n \"Rotate\": rotate_level_to_args(MAX_LEVEL, replace_value),\n \"Solarize\": solarize_level_to_args(MAX_LEVEL),\n \"Color\": enhance_level_to_args(MAX_LEVEL),\n \"Contrast\": enhance_level_to_args(MAX_LEVEL),\n \"Brightness\": enhance_level_to_args(MAX_LEVEL),\n \"Sharpness\": enhance_level_to_args(MAX_LEVEL),\n \"ShearX\": shear_level_to_args(MAX_LEVEL, replace_value),\n \"TranslateX\": translate_level_to_args(translate_const, MAX_LEVEL, replace_value),\n \"TranslateY\": translate_level_to_args(translate_const, MAX_LEVEL, replace_value),\n \"Posterize\": posterize_level_to_args(MAX_LEVEL),\n \"ShearY\": shear_level_to_args(MAX_LEVEL, replace_value),\n}\n\n\nclass RandomAugment(object):\n def __init__(self, N=2, M=10, isPIL=False, augs=[]):\n self.N = N\n self.M = M\n self.isPIL = isPIL\n if augs:\n self.augs = augs\n else:\n self.augs = list(arg_dict.keys())\n\n def get_random_ops(self):\n sampled_ops = np.random.choice(self.augs, self.N)\n return [(op, 0.5, self.M) for op in sampled_ops]\n\n def __call__(self, img):\n if self.isPIL:\n img = np.array(img)\n ops = self.get_random_ops()\n for name, prob, level in ops:\n if np.random.random() > prob:\n continue\n args = arg_dict[name](level)\n img = func_dict[name](img, *args)\n return img\n\n\nclass VideoRandomAugment(object):\n def __init__(self, N=2, M=10, p=0.0, tensor_in_tensor_out=True, augs=[]):\n self.N = N\n self.M = M\n self.p = p\n self.tensor_in_tensor_out = tensor_in_tensor_out\n if augs:\n self.augs = augs\n else:\n self.augs = list(arg_dict.keys())\n\n def get_random_ops(self):\n sampled_ops = np.random.choice(self.augs, self.N, replace=False)\n return [(op, self.M) for op in sampled_ops]\n\n def __call__(self, frames):\n assert (\n frames.shape[-1] == 3\n ), \"Expecting last dimension for 3-channels RGB (b, h, w, c).\"\n\n if self.tensor_in_tensor_out:\n frames = frames.numpy().astype(np.uint8)\n\n num_frames = frames.shape[0]\n\n ops = num_frames * [self.get_random_ops()]\n apply_or_not = num_frames * [np.random.random(size=self.N) > self.p]\n\n frames = torch.stack(\n list(map(self._aug, frames, ops, apply_or_not)), dim=0\n ).float()\n\n return frames\n\n def _aug(self, img, ops, apply_or_not):\n for i, (name, level) in enumerate(ops):\n if not apply_or_not[i]:\n continue\n args = arg_dict[name](level)\n img = func_dict[name](img, *args)\n return torch.from_numpy(img)\n\n\nif __name__ == \"__main__\":\n a = RandomAugment()\n img = np.random.randn(32, 32, 3)\n a(img)\n" }, { "path": "minigpt4/runners/__init__.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nfrom minigpt4.runners.runner_base import RunnerBase\n\n__all__ = [\"RunnerBase\"]\n" }, { "path": "minigpt4/runners/runner_base.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport datetime\nimport json\nimport logging\nimport os\nimport time\nfrom pathlib import Path\n\nimport torch\nimport torch.distributed as dist\nimport webdataset as wds\nfrom minigpt4.common.dist_utils import (\n download_cached_file,\n get_rank,\n get_world_size,\n is_main_process,\n main_process,\n)\nfrom minigpt4.common.registry import registry\nfrom minigpt4.common.utils import is_url\nfrom minigpt4.datasets.data_utils import concat_datasets, reorg_datasets_by_split, ChainDataset\nfrom minigpt4.datasets.datasets.dataloader_utils import (\n IterLoader,\n MultiIterLoader,\n PrefetchLoader,\n)\nfrom torch.nn.parallel import DistributedDataParallel as DDP\nfrom torch.utils.data import DataLoader, DistributedSampler\n\n\n@registry.register_runner(\"runner_base\")\nclass RunnerBase:\n \"\"\"\n A runner class to train and evaluate a model given a task and datasets.\n\n The runner uses pytorch distributed data parallel by default. Future release\n will support other distributed frameworks.\n \"\"\"\n\n def __init__(self, cfg, task, model, datasets, job_id):\n self.config = cfg\n self.job_id = job_id\n\n self.task = task\n self.datasets = datasets\n\n self._model = model\n\n self._wrapped_model = None\n self._device = None\n self._optimizer = None\n self._scaler = None\n self._dataloaders = None\n self._lr_sched = None\n\n self.start_epoch = 0\n\n # self.setup_seeds()\n self.setup_output_dir()\n\n @property\n def device(self):\n if self._device is None:\n self._device = torch.device(self.config.run_cfg.device)\n\n return self._device\n\n @property\n def use_distributed(self):\n return self.config.run_cfg.distributed\n\n @property\n def model(self):\n \"\"\"\n A property to get the DDP-wrapped model on the device.\n \"\"\"\n # move model to device\n if self._model.device != self.device:\n self._model = self._model.to(self.device)\n\n # distributed training wrapper\n if self.use_distributed:\n if self._wrapped_model is None:\n self._wrapped_model = DDP(\n self._model, device_ids=[self.config.run_cfg.gpu], find_unused_parameters=True\n )\n else:\n self._wrapped_model = self._model\n\n return self._wrapped_model\n\n @property\n def optimizer(self):\n # TODO make optimizer class and configurations\n if self._optimizer is None:\n num_parameters = 0\n p_wd, p_non_wd = [], []\n for n, p in self.model.named_parameters():\n if not p.requires_grad:\n continue # frozen weights\n print(n)\n if p.ndim < 2 or \"bias\" in n or \"ln\" in n or \"bn\" in n:\n p_non_wd.append(p)\n else:\n p_wd.append(p)\n num_parameters += p.data.nelement()\n logging.info(\"number of trainable parameters: %d\" % num_parameters)\n optim_params = [\n {\n \"params\": p_wd,\n \"weight_decay\": float(self.config.run_cfg.weight_decay),\n },\n {\"params\": p_non_wd, \"weight_decay\": 0},\n ]\n beta2 = self.config.run_cfg.get(\"beta2\", 0.999)\n self._optimizer = torch.optim.AdamW(\n optim_params,\n lr=float(self.config.run_cfg.init_lr),\n weight_decay=float(self.config.run_cfg.weight_decay),\n betas=(0.9, beta2),\n )\n\n return self._optimizer\n\n @property\n def scaler(self):\n amp = self.config.run_cfg.get(\"amp\", False)\n\n if amp:\n if self._scaler is None:\n self._scaler = torch.cuda.amp.GradScaler()\n\n return self._scaler\n\n @property\n def lr_scheduler(self):\n \"\"\"\n A property to get and create learning rate scheduler by split just in need.\n \"\"\"\n if self._lr_sched is None:\n lr_sched_cls = registry.get_lr_scheduler_class(self.config.run_cfg.lr_sched)\n\n # max_epoch = self.config.run_cfg.max_epoch\n max_epoch = self.max_epoch\n # min_lr = self.config.run_cfg.min_lr\n min_lr = self.min_lr\n # init_lr = self.config.run_cfg.init_lr\n init_lr = self.init_lr\n\n # optional parameters\n decay_rate = self.config.run_cfg.get(\"lr_decay_rate\", None)\n warmup_start_lr = self.config.run_cfg.get(\"warmup_lr\", -1)\n warmup_steps = self.config.run_cfg.get(\"warmup_steps\", 0)\n iters_per_epoch = self.config.run_cfg.get(\"iters_per_epoch\", None)\n\n if iters_per_epoch is None:\n try:\n iters_per_epoch = len(self.dataloaders['train'])\n except (AttributeError, TypeError):\n iters_per_epoch = 10000\n\n self._lr_sched = lr_sched_cls(\n optimizer=self.optimizer,\n max_epoch=max_epoch,\n iters_per_epoch=iters_per_epoch,\n min_lr=min_lr,\n init_lr=init_lr,\n decay_rate=decay_rate,\n warmup_start_lr=warmup_start_lr,\n warmup_steps=warmup_steps,\n )\n\n return self._lr_sched\n\n @property\n def dataloaders(self) -> dict:\n \"\"\"\n A property to get and create dataloaders by split just in need.\n\n If no train_dataset_ratio is provided, concatenate map-style datasets and\n chain wds.DataPipe datasets separately. Training set becomes a tuple\n (ConcatDataset, ChainDataset), both are optional but at least one of them is\n required. The resultant ConcatDataset and ChainDataset will be sampled evenly.\n\n If train_dataset_ratio is provided, create a MultiIterLoader to sample\n each dataset by ratios during training.\n\n Currently do not support multiple datasets for validation and test.\n\n Returns:\n dict: {split_name: (tuples of) dataloader}\n \"\"\"\n if self._dataloaders is None:\n\n # concatenate map-style datasets and chain wds.DataPipe datasets separately\n # training set becomes a tuple (ConcatDataset, ChainDataset), both are\n # optional but at least one of them is required. The resultant ConcatDataset\n # and ChainDataset will be sampled evenly.\n logging.info(\n \"dataset_ratios not specified, datasets will be concatenated (map-style datasets) or chained (webdataset.DataPipeline).\"\n )\n\n batch_sizes = {dataset_name: getattr(self.config.datasets_cfg, dataset_name).batch_size\n for dataset_name in self.datasets.keys()}\n datasets, batch_sizes = reorg_datasets_by_split(self.datasets, batch_sizes)\n self.datasets = datasets\n # self.datasets = concat_datasets(datasets)\n\n # print dataset statistics after concatenation/chaining\n for split_name in self.datasets:\n if isinstance(self.datasets[split_name], tuple) or isinstance(\n self.datasets[split_name], list\n ):\n # mixed wds.DataPipeline and torch.utils.data.Dataset\n num_records = sum(\n [\n len(d)\n if not type(d) in [wds.DataPipeline, ChainDataset]\n else 0\n for d in self.datasets[split_name]\n ]\n )\n\n else:\n if hasattr(self.datasets[split_name], \"__len__\"):\n # a single map-style dataset\n num_records = len(self.datasets[split_name])\n else:\n # a single wds.DataPipeline\n num_records = -1\n logging.info(\n \"Only a single wds.DataPipeline dataset, no __len__ attribute.\"\n )\n\n if num_records >= 0:\n logging.info(\n \"Loaded {} records for {} split from the dataset.\".format(\n num_records, split_name\n )\n )\n\n # create dataloaders\n split_names = sorted(self.datasets.keys())\n\n datasets = [self.datasets[split] for split in split_names]\n batch_sizes = [batch_sizes[split] for split in split_names]\n is_trains = [split in self.train_splits for split in split_names]\n\n print(\"batch sizes\", batch_sizes)\n\n collate_fns = []\n for dataset in datasets:\n if isinstance(dataset, tuple) or isinstance(dataset, list):\n collate_fns.append([getattr(d, \"collater\", None) for d in dataset])\n else:\n collate_fns.append(getattr(dataset, \"collater\", None))\n\n dataloaders = self.create_loaders(\n datasets=datasets,\n num_workers=self.config.run_cfg.num_workers,\n batch_sizes=batch_sizes,\n is_trains=is_trains,\n collate_fns=collate_fns,\n )\n\n self._dataloaders = {k: v for k, v in zip(split_names, dataloaders)}\n\n return self._dataloaders\n\n @property\n def cuda_enabled(self):\n return self.device.type == \"cuda\"\n\n @property\n def max_epoch(self):\n return int(self.config.run_cfg.max_epoch)\n\n @property\n def log_freq(self):\n log_freq = self.config.run_cfg.get(\"log_freq\", 50)\n return int(log_freq)\n\n @property\n def init_lr(self):\n return float(self.config.run_cfg.init_lr)\n\n @property\n def min_lr(self):\n return float(self.config.run_cfg.min_lr)\n\n @property\n def accum_grad_iters(self):\n return int(self.config.run_cfg.get(\"accum_grad_iters\", 1))\n\n @property\n def valid_splits(self):\n valid_splits = self.config.run_cfg.get(\"valid_splits\", [])\n\n if len(valid_splits) == 0:\n logging.info(\"No validation splits found.\")\n\n return valid_splits\n\n @property\n def test_splits(self):\n test_splits = self.config.run_cfg.get(\"test_splits\", [])\n\n return test_splits\n\n @property\n def train_splits(self):\n train_splits = self.config.run_cfg.get(\"train_splits\", [])\n\n if len(train_splits) == 0:\n logging.info(\"Empty train splits.\")\n\n return train_splits\n\n @property\n def evaluate_only(self):\n \"\"\"\n Set to True to skip training.\n \"\"\"\n return self.config.run_cfg.evaluate\n\n @property\n def use_dist_eval_sampler(self):\n return self.config.run_cfg.get(\"use_dist_eval_sampler\", True)\n\n @property\n def resume_ckpt_path(self):\n return self.config.run_cfg.get(\"resume_ckpt_path\", None)\n\n @property\n def train_loader(self):\n train_dataloader = self.dataloaders[\"train\"]\n\n return train_dataloader\n\n def setup_output_dir(self):\n lib_root = Path(registry.get_path(\"library_root\"))\n\n output_dir = lib_root / self.config.run_cfg.output_dir / self.job_id\n # output_dir = lib_root / self.config.run_cfg.output_dir\n result_dir = output_dir / \"result\"\n\n output_dir.mkdir(parents=True, exist_ok=True)\n result_dir.mkdir(parents=True, exist_ok=True)\n\n registry.register_path(\"result_dir\", str(result_dir))\n registry.register_path(\"output_dir\", str(output_dir))\n\n self.result_dir = result_dir\n self.output_dir = output_dir\n\n def train(self):\n start_time = time.time()\n best_agg_metric = 0\n best_epoch = 0\n\n self.log_config()\n\n # resume from checkpoint if specified\n if not self.evaluate_only and self.resume_ckpt_path is not None:\n self._load_checkpoint(self.resume_ckpt_path)\n\n for cur_epoch in range(self.start_epoch, self.max_epoch):\n # training phase\n if not self.evaluate_only:\n logging.info(\"Start training\")\n train_stats = self.train_epoch(cur_epoch)\n self.log_stats(split_name=\"train\", stats=train_stats)\n\n # evaluation phase\n if len(self.valid_splits) > 0:\n for split_name in self.valid_splits:\n logging.info(\"Evaluating on {}.\".format(split_name))\n\n val_log = self.eval_epoch(\n split_name=split_name, cur_epoch=cur_epoch\n )\n if val_log is not None:\n if is_main_process():\n assert (\n \"agg_metrics\" in val_log\n ), \"No agg_metrics found in validation log.\"\n\n agg_metrics = val_log[\"agg_metrics\"]\n if agg_metrics > best_agg_metric and split_name == \"val\":\n best_epoch, best_agg_metric = cur_epoch, agg_metrics\n\n self._save_checkpoint(cur_epoch, is_best=True)\n\n val_log.update({\"best_epoch\": best_epoch})\n self.log_stats(val_log, split_name)\n\n else:\n # if no validation split is provided, we just save the checkpoint at the end of each epoch.\n if not self.evaluate_only:\n self._save_checkpoint(cur_epoch, is_best=False)\n\n if self.evaluate_only:\n break\n\n if self.config.run_cfg.distributed:\n dist.barrier()\n\n # testing phase\n test_epoch = \"best\" if len(self.valid_splits) > 0 else cur_epoch\n self.evaluate(cur_epoch=test_epoch, skip_reload=self.evaluate_only)\n\n total_time = time.time() - start_time\n total_time_str = str(datetime.timedelta(seconds=int(total_time)))\n logging.info(\"Training time {}\".format(total_time_str))\n\n def evaluate(self, cur_epoch=\"best\", skip_reload=False):\n test_logs = dict()\n\n if len(self.test_splits) > 0:\n for split_name in self.test_splits:\n test_logs[split_name] = self.eval_epoch(\n split_name=split_name, cur_epoch=cur_epoch, skip_reload=skip_reload\n )\n\n return test_logs\n\n def train_epoch(self, epoch):\n # train\n self.model.train()\n\n return self.task.train_epoch(\n epoch=epoch,\n model=self.model,\n data_loader=self.train_loader,\n optimizer=self.optimizer,\n scaler=self.scaler,\n lr_scheduler=self.lr_scheduler,\n cuda_enabled=self.cuda_enabled,\n log_freq=self.log_freq,\n accum_grad_iters=self.accum_grad_iters,\n )\n\n @torch.no_grad()\n def eval_epoch(self, split_name, cur_epoch, skip_reload=False):\n \"\"\"\n Evaluate the model on a given split.\n\n Args:\n split_name (str): name of the split to evaluate on.\n cur_epoch (int): current epoch.\n skip_reload_best (bool): whether to skip reloading the best checkpoint.\n During training, we will reload the best checkpoint for validation.\n During testing, we will use provided weights and skip reloading the best checkpoint .\n \"\"\"\n data_loader = self.dataloaders.get(split_name, None)\n assert data_loader, \"data_loader for split {} is None.\".format(split_name)\n\n # TODO In validation, you need to compute loss as well as metrics\n # TODO consider moving to model.before_evaluation()\n model = self.unwrap_dist_model(self.model)\n if not skip_reload and cur_epoch == \"best\":\n model = self._reload_best_model(model)\n model.eval()\n\n self.task.before_evaluation(\n model=model,\n dataset=self.datasets[split_name],\n )\n results = self.task.evaluation(model, data_loader)\n\n if results is not None:\n return self.task.after_evaluation(\n val_result=results,\n split_name=split_name,\n epoch=cur_epoch,\n )\n\n def unwrap_dist_model(self, model):\n if self.use_distributed:\n return model.module\n else:\n return model\n\n def create_loaders(\n self,\n datasets,\n num_workers,\n batch_sizes,\n is_trains,\n collate_fns,\n dataset_ratios=None,\n ):\n \"\"\"\n Create dataloaders for training and validation.\n \"\"\"\n\n def _create_loader(dataset, num_workers, bsz, is_train, collate_fn):\n # create a single dataloader for each split\n if isinstance(dataset, ChainDataset) or isinstance(\n dataset, wds.DataPipeline\n ):\n # wds.WebdDataset instance are chained together\n # webdataset.DataPipeline has its own sampler and collate_fn\n loader = iter(\n DataLoader(\n dataset,\n batch_size=bsz,\n num_workers=num_workers,\n pin_memory=True,\n )\n )\n else:\n # map-style dataset are concatenated together\n # setup distributed sampler\n\n if self.use_distributed:\n sampler = DistributedSampler(\n dataset,\n shuffle=is_train,\n num_replicas=get_world_size(),\n rank=get_rank(),\n )\n if not self.use_dist_eval_sampler:\n # e.g. retrieval evaluation\n sampler = sampler if is_train else None\n else:\n sampler = None\n\n loader = DataLoader(\n dataset,\n batch_size=bsz,\n num_workers=num_workers,\n pin_memory=True,\n sampler=sampler,\n shuffle=sampler is None and is_train,\n collate_fn=collate_fn,\n drop_last=True if is_train else False,\n )\n loader = PrefetchLoader(loader)\n\n if is_train:\n loader = IterLoader(loader, use_distributed=self.use_distributed)\n\n return loader\n\n loaders = []\n\n for dataset, bsz, is_train, collate_fn in zip(\n datasets, batch_sizes, is_trains, collate_fns\n ):\n if isinstance(dataset, list) or isinstance(dataset, tuple):\n if hasattr(dataset[0], 'sample_ratio') and dataset_ratios is None:\n dataset_ratios = [d.sample_ratio for d in dataset]\n loader = MultiIterLoader(\n loaders=[\n _create_loader(d, num_workers, bsz[i], is_train, collate_fn[i])\n for i, d in enumerate(dataset)\n ],\n ratios=dataset_ratios,\n )\n else:\n loader = _create_loader(dataset, num_workers, bsz, is_train, collate_fn)\n\n loaders.append(loader)\n\n return loaders\n\n @main_process\n def _save_checkpoint(self, cur_epoch, is_best=False):\n \"\"\"\n Save the checkpoint at the current epoch.\n \"\"\"\n model_no_ddp = self.unwrap_dist_model(self.model)\n param_grad_dic = {\n k: v.requires_grad for (k, v) in model_no_ddp.named_parameters()\n }\n state_dict = model_no_ddp.state_dict()\n for k in list(state_dict.keys()):\n if k in param_grad_dic.keys() and not param_grad_dic[k]:\n # delete parameters that do not require gradient\n del state_dict[k]\n save_obj = {\n \"model\": state_dict,\n \"optimizer\": self.optimizer.state_dict(),\n \"config\": self.config.to_dict(),\n \"scaler\": self.scaler.state_dict() if self.scaler else None,\n \"epoch\": cur_epoch,\n }\n save_to = os.path.join(\n self.output_dir,\n \"checkpoint_{}.pth\".format(\"best\" if is_best else cur_epoch),\n )\n logging.info(\"Saving checkpoint at epoch {} to {}.\".format(cur_epoch, save_to))\n torch.save(save_obj, save_to)\n\n def _reload_best_model(self, model):\n \"\"\"\n Load the best checkpoint for evaluation.\n \"\"\"\n checkpoint_path = os.path.join(self.output_dir, \"checkpoint_best.pth\")\n\n logging.info(\"Loading checkpoint from {}.\".format(checkpoint_path))\n checkpoint = torch.load(checkpoint_path, map_location=\"cpu\")\n try:\n model.load_state_dict(checkpoint[\"model\"])\n except RuntimeError as e:\n logging.warning(\n \"\"\"\n Key mismatch when loading checkpoint. This is expected if only part of the model is saved.\n Trying to load the model with strict=False.\n \"\"\"\n )\n model.load_state_dict(checkpoint[\"model\"], strict=False)\n return model\n\n def _load_checkpoint(self, url_or_filename):\n \"\"\"\n Resume from a checkpoint.\n \"\"\"\n if is_url(url_or_filename):\n cached_file = download_cached_file(\n url_or_filename, check_hash=False, progress=True\n )\n checkpoint = torch.load(cached_file, map_location=self.device)\n elif os.path.isfile(url_or_filename):\n checkpoint = torch.load(url_or_filename, map_location=self.device)\n else:\n raise RuntimeError(\"checkpoint url or path is invalid\")\n\n state_dict = checkpoint[\"model\"]\n message = self.unwrap_dist_model(self.model).load_state_dict(state_dict,strict=False)\n\n self.optimizer.load_state_dict(checkpoint[\"optimizer\"])\n if self.scaler and \"scaler\" in checkpoint:\n self.scaler.load_state_dict(checkpoint[\"scaler\"])\n\n self.start_epoch = checkpoint[\"epoch\"] + 1\n print(\"resume the checkpoint\")\n logging.info(\"Resume checkpoint from {}\".format(url_or_filename))\n\n @main_process\n def log_stats(self, stats, split_name):\n if isinstance(stats, dict):\n log_stats = {**{f\"{split_name}_{k}\": v for k, v in stats.items()}}\n with open(os.path.join(self.output_dir, \"log.txt\"), \"a\") as f:\n f.write(json.dumps(log_stats) + \"\\n\")\n elif isinstance(stats, list):\n pass\n\n @main_process\n def log_config(self):\n with open(os.path.join(self.output_dir, \"log.txt\"), \"a\") as f:\n f.write(json.dumps(self.config.to_dict(), indent=4) + \"\\n\")\n" }, { "path": "minigpt4/tasks/__init__.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.tasks.base_task import BaseTask\nfrom minigpt4.tasks.image_text_pretrain import ImageTextPretrainTask\n\n\ndef setup_task(cfg):\n assert \"task\" in cfg.run_cfg, \"Task name must be provided.\"\n\n task_name = cfg.run_cfg.task\n task = registry.get_task_class(task_name).setup_task(cfg=cfg)\n assert task is not None, \"Task {} not properly registered.\".format(task_name)\n\n return task\n\n\n__all__ = [\n \"BaseTask\",\n \"ImageTextPretrainTask\",\n]\n" }, { "path": "minigpt4/tasks/base_task.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport logging\nimport os\n\nimport torch\nimport torch.distributed as dist\nfrom minigpt4.common.dist_utils import get_rank, get_world_size, is_main_process, is_dist_avail_and_initialized\nfrom minigpt4.common.logger import MetricLogger, SmoothedValue\nfrom minigpt4.common.registry import registry\nfrom minigpt4.datasets.data_utils import prepare_sample\nimport wandb\n\nclass BaseTask:\n def __init__(self, **kwargs):\n super().__init__()\n\n self.inst_id_key = \"instance_id\"\n self.cfg = \"\"\n\n @classmethod\n def setup_task(cls, **kwargs):\n return cls()\n\n def build_model(self, cfg):\n self.cfg = cfg\n model_config = cfg.model_cfg\n\n model_cls = registry.get_model_class(model_config.arch)\n return model_cls.from_config(model_config)\n\n def build_datasets(self, cfg):\n \"\"\"\n Build a dictionary of datasets, keyed by split 'train', 'valid', 'test'.\n Download dataset and annotations automatically if not exist.\n\n Args:\n cfg (common.config.Config): _description_\n\n Returns:\n dict: Dictionary of torch.utils.data.Dataset objects by split.\n \"\"\"\n\n datasets = dict()\n\n datasets_config = cfg.datasets_cfg\n\n assert len(datasets_config) > 0, \"At least one dataset has to be specified.\"\n\n for name in datasets_config:\n dataset_config = datasets_config[name]\n\n builder = registry.get_builder_class(name)(dataset_config)\n dataset = builder.build_datasets()\n\n dataset['train'].name = name\n if 'sample_ratio' in dataset_config:\n dataset['train'].sample_ratio = dataset_config.sample_ratio\n\n datasets[name] = dataset\n\n return datasets\n\n def train_step(self, model, samples):\n loss = model(samples)[\"loss\"]\n return loss\n\n def valid_step(self, model, samples):\n raise NotImplementedError\n\n def before_evaluation(self, model, dataset, **kwargs):\n model.before_evaluation(dataset=dataset, task_type=type(self))\n\n def after_evaluation(self, **kwargs):\n pass\n\n def inference_step(self):\n raise NotImplementedError\n\n def evaluation(self, model, data_loader, cuda_enabled=True):\n metric_logger = MetricLogger(delimiter=\" \")\n header = \"Evaluation\"\n # TODO make it configurable\n print_freq = 10\n\n results = []\n\n for samples in metric_logger.log_every(data_loader, print_freq, header):\n samples = prepare_sample(samples, cuda_enabled=cuda_enabled)\n\n eval_output = self.valid_step(model=model, samples=samples)\n results.extend(eval_output)\n\n if is_dist_avail_and_initialized():\n dist.barrier()\n\n return results\n\n def train_epoch(\n self,\n epoch,\n model,\n data_loader,\n optimizer,\n lr_scheduler,\n scaler=None,\n cuda_enabled=False,\n log_freq=50,\n accum_grad_iters=1,\n ):\n return self._train_inner_loop(\n epoch=epoch,\n iters_per_epoch=lr_scheduler.iters_per_epoch,\n model=model,\n data_loader=data_loader,\n optimizer=optimizer,\n scaler=scaler,\n lr_scheduler=lr_scheduler,\n log_freq=log_freq,\n cuda_enabled=cuda_enabled,\n accum_grad_iters=accum_grad_iters,\n )\n\n def train_iters(\n self,\n epoch,\n start_iters,\n iters_per_inner_epoch,\n model,\n data_loader,\n optimizer,\n lr_scheduler,\n scaler=None,\n cuda_enabled=False,\n log_freq=50,\n accum_grad_iters=1,\n ):\n return self._train_inner_loop(\n epoch=epoch,\n start_iters=start_iters,\n iters_per_epoch=iters_per_inner_epoch,\n model=model,\n data_loader=data_loader,\n optimizer=optimizer,\n scaler=scaler,\n lr_scheduler=lr_scheduler,\n log_freq=log_freq,\n cuda_enabled=cuda_enabled,\n accum_grad_iters=accum_grad_iters,\n )\n\n def _train_inner_loop(\n self,\n epoch,\n iters_per_epoch,\n model,\n data_loader,\n optimizer,\n lr_scheduler,\n scaler=None,\n start_iters=None,\n log_freq=50,\n cuda_enabled=False,\n accum_grad_iters=1,\n ):\n \"\"\"\n An inner training loop compatible with both epoch-based and iter-based training.\n\n When using epoch-based, training stops after one epoch; when using iter-based,\n training stops after #iters_per_epoch iterations.\n \"\"\"\n use_amp = scaler is not None\n\n if not hasattr(data_loader, \"__next__\"):\n # convert to iterator if not already\n data_loader = iter(data_loader)\n\n metric_logger = MetricLogger(delimiter=\" \")\n metric_logger.add_meter(\"lr\", SmoothedValue(window_size=1, fmt=\"{value:.6f}\"))\n metric_logger.add_meter(\"loss\", SmoothedValue(window_size=1, fmt=\"{value:.4f}\"))\n\n # if iter-based runner, schedule lr based on inner epoch.\n logging.info(\n \"Start training epoch {}, {} iters per inner epoch.\".format(\n epoch, iters_per_epoch\n )\n )\n header = \"Train: data epoch: [{}]\".format(epoch)\n if start_iters is None:\n # epoch-based runner\n inner_epoch = epoch\n else:\n # In iter-based runner, we schedule the learning rate based on iterations.\n inner_epoch = start_iters // iters_per_epoch\n header = header + \"; inner epoch [{}]\".format(inner_epoch)\n\n for i in metric_logger.log_every(range(iters_per_epoch), log_freq, header):\n # if using iter-based runner, we stop after iters_per_epoch iterations.\n if i >= iters_per_epoch:\n break\n\n samples = next(data_loader)\n\n samples = prepare_sample(samples, cuda_enabled=cuda_enabled)\n samples.update(\n {\n \"epoch\": inner_epoch,\n \"num_iters_per_epoch\": iters_per_epoch,\n \"iters\": i,\n }\n )\n\n lr_scheduler.step(cur_epoch=inner_epoch, cur_step=i)\n\n with torch.cuda.amp.autocast(enabled=use_amp):\n loss = self.train_step(model=model, samples=samples)\n\n # after_train_step()\n if use_amp:\n scaler.scale(loss).backward()\n else:\n loss.backward()\n\n # update gradients every accum_grad_iters iterations\n if (i + 1) % accum_grad_iters == 0:\n if use_amp:\n scaler.step(optimizer)\n scaler.update() \n else: \n optimizer.step()\n optimizer.zero_grad()\n # if self.cfg.wandb_log:\n if self.cfg.run_cfg.wandb_log:\n wandb.log({\"epoch\": inner_epoch, \"loss\": loss})\n metric_logger.update(loss=loss.item())\n metric_logger.update(lr=optimizer.param_groups[0][\"lr\"])\n\n # after train_epoch()\n # gather the stats from all processes\n metric_logger.synchronize_between_processes()\n logging.info(\"Averaged stats: \" + str(metric_logger.global_avg()))\n return {\n k: \"{:.3f}\".format(meter.global_avg)\n for k, meter in metric_logger.meters.items()\n }\n\n @staticmethod\n def save_result(result, result_dir, filename, remove_duplicate=\"\"):\n import json\n\n result_file = os.path.join(\n result_dir, \"%s_rank%d.json\" % (filename, get_rank())\n )\n final_result_file = os.path.join(result_dir, \"%s.json\" % filename)\n\n json.dump(result, open(result_file, \"w\"))\n\n if is_dist_avail_and_initialized():\n dist.barrier()\n\n if is_main_process():\n logging.warning(\"rank %d starts merging results.\" % get_rank())\n # combine results from all processes\n result = []\n\n for rank in range(get_world_size()):\n result_file = os.path.join(\n result_dir, \"%s_rank%d.json\" % (filename, rank)\n )\n res = json.load(open(result_file, \"r\"))\n result += res\n\n if remove_duplicate:\n result_new = []\n id_list = []\n for res in result:\n if res[remove_duplicate] not in id_list:\n id_list.append(res[remove_duplicate])\n result_new.append(res)\n result = result_new\n\n json.dump(result, open(final_result_file, \"w\"))\n print(\"result file saved to %s\" % final_result_file)\n\n return final_result_file\n" }, { "path": "minigpt4/tasks/image_text_pretrain.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nfrom minigpt4.common.registry import registry\nfrom minigpt4.tasks.base_task import BaseTask\n\n\n@registry.register_task(\"image_text_pretrain\")\nclass ImageTextPretrainTask(BaseTask):\n def __init__(self):\n super().__init__()\n\n def evaluation(self, model, data_loader, cuda_enabled=True):\n pass\n" }, { "path": "train.py", "content": "\"\"\"\n Copyright (c) 2022, salesforce.com, inc.\n All rights reserved.\n SPDX-License-Identifier: BSD-3-Clause\n For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause\n\"\"\"\n\nimport argparse\nimport os\nimport random\n\nimport numpy as np\nimport torch\nimport torch.backends.cudnn as cudnn\nimport wandb\n\nimport minigpt4.tasks as tasks\nfrom minigpt4.common.config import Config\nfrom minigpt4.common.dist_utils import get_rank, init_distributed_mode\nfrom minigpt4.common.logger import setup_logger\nfrom minigpt4.common.optims import (\n LinearWarmupCosineLRScheduler,\n LinearWarmupStepLRScheduler,\n)\nfrom minigpt4.common.registry import registry\nfrom minigpt4.common.utils import now\n\n# imports modules for registration\nfrom minigpt4.datasets.builders import *\nfrom minigpt4.models import *\nfrom minigpt4.processors import *\nfrom minigpt4.runners import *\nfrom minigpt4.tasks import *\n\n\ndef parse_args():\n parser = argparse.ArgumentParser(description=\"Training\")\n\n parser.add_argument(\"--cfg-path\", required=True, help=\"path to configuration file.\")\n parser.add_argument(\n \"--options\",\n nargs=\"+\",\n help=\"override some settings in the used config, the key-value pair \"\n \"in xxx=yyy format will be merged into config file (deprecate), \"\n \"change to --cfg-options instead.\",\n )\n args = parser.parse_args()\n\n return args\n\n\ndef setup_seeds(config):\n seed = config.run_cfg.seed + get_rank()\n\n random.seed(seed)\n np.random.seed(seed)\n torch.manual_seed(seed)\n\n cudnn.benchmark = False\n cudnn.deterministic = True\n\n\ndef get_runner_class(cfg):\n \"\"\"\n Get runner class from config. Default to epoch-based runner.\n \"\"\"\n runner_cls = registry.get_runner_class(cfg.run_cfg.get(\"runner\", \"runner_base\"))\n\n return runner_cls\n\n\ndef main():\n # allow auto-dl completes on main process without timeout when using NCCL backend.\n # os.environ[\"NCCL_BLOCKING_WAIT\"] = \"1\"\n\n # set before init_distributed_mode() to ensure the same job_id shared across all ranks.\n job_id = now()\n args = parse_args()\n cfg = Config(args)\n\n init_distributed_mode(cfg.run_cfg)\n setup_seeds(cfg)\n\n # set after init_distributed_mode() to only log on master.\n setup_logger()\n cfg.pretty_print()\n\n task = tasks.setup_task(cfg)\n datasets = task.build_datasets(cfg)\n model = task.build_model(cfg)\n\n if cfg.run_cfg.wandb_log:\n wandb.login()\n wandb.init(project=\"minigptv\", name=cfg.run_cfg.job_name)\n wandb.watch(model)\n\n runner = get_runner_class(cfg)(\n cfg=cfg, job_id=job_id, task=task, model=model, datasets=datasets\n )\n runner.train()\n\n\nif __name__ == \"__main__\":\n main()\n" }, { "path": "train_configs/minigpt4_llama2_stage1_pretrain.yaml", "content": "model:\n arch: minigpt4\n model_type: pretrain_llama2\n\n\ndatasets:\n laion:\n batch_size: 64\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 115\n cc_sbu:\n batch_size: 64\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 14\n\n\nrun:\n task: image_text_pretrain\n # optimizer\n lr_sched: \"linear_warmup_cosine_lr\"\n init_lr: 1e-4\n min_lr: 8e-5\n warmup_lr: 1e-6\n\n weight_decay: 0.05\n max_epoch: 4\n num_workers: 4\n warmup_steps: 5000\n iters_per_epoch: 5000\n\n seed: 42\n output_dir: \"output/minigpt4_stage1_pretrain\"\n\n amp: True\n resume_ckpt_path: null\n\n evaluate: False \n train_splits: [\"train\"]\n\n device: \"cuda\"\n world_size: 1\n dist_url: \"env://\"\n distributed: True\n\n wandb_log: True\n job_name: minigpt4_llama2_pretrain" }, { "path": "train_configs/minigpt4_llama2_stage2_finetune.yaml", "content": "model:\n arch: minigpt4\n model_type: pretrain_llama2\n\n max_txt_len: 160\n end_sym: \"\"\n prompt_path: \"prompts/alignment.txt\"\n prompt_template: '[INST] {} [/INST] '\n ckpt: '/path/to/stage1/checkpoint/'\n\n\ndatasets:\n cc_sbu_align:\n batch_size: 12\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n\nrun:\n task: image_text_pretrain\n # optimizer\n lr_sched: \"linear_warmup_cosine_lr\"\n init_lr: 3e-5\n min_lr: 1e-5\n warmup_lr: 1e-6\n\n weight_decay: 0.05\n max_epoch: 5\n iters_per_epoch: 200\n num_workers: 4\n warmup_steps: 200\n\n seed: 42\n output_dir: \"output/minigpt4_stage2_finetune\"\n\n amp: True\n resume_ckpt_path: null\n\n evaluate: False \n train_splits: [\"train\"]\n\n device: \"cuda\"\n world_size: 1\n dist_url: \"env://\"\n distributed: True\n\n wandb_log: True\n job_name: minigpt4_llama2_finetune" }, { "path": "train_configs/minigpt4_stage1_pretrain.yaml", "content": "model:\n arch: minigpt4\n model_type: pretrain_vicuna0\n\n\ndatasets:\n laion:\n batch_size: 64\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 115\n cc_sbu:\n batch_size: 64\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 14\n\n\nrun:\n task: image_text_pretrain\n # optimizer\n lr_sched: \"linear_warmup_cosine_lr\"\n init_lr: 1e-4\n min_lr: 8e-5\n warmup_lr: 1e-6\n\n weight_decay: 0.05\n max_epoch: 4\n num_workers: 4\n warmup_steps: 5000\n iters_per_epoch: 5000\n\n seed: 42\n output_dir: \"output/minigpt4_stage1_pretrain\"\n\n amp: True\n resume_ckpt_path: null\n\n evaluate: False \n train_splits: [\"train\"]\n\n device: \"cuda\"\n world_size: 1\n dist_url: \"env://\"\n distributed: True\n\n wandb_log: True\n job_name: minigpt4_pretrain" }, { "path": "train_configs/minigpt4_stage2_finetune.yaml", "content": "model:\n arch: minigpt4\n model_type: pretrain_vicuna0\n\n max_txt_len: 160\n end_sym: \"###\"\n prompt_path: \"prompts/alignment.txt\"\n prompt_template: '###Human: {} ###Assistant: '\n ckpt: '/path/to/stage1/checkpoint/'\n\n\ndatasets:\n cc_sbu_align:\n batch_size: 12\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 224\n text_processor:\n train:\n name: \"blip_caption\"\n\nrun:\n task: image_text_pretrain\n # optimizer\n lr_sched: \"linear_warmup_cosine_lr\"\n init_lr: 3e-5\n min_lr: 1e-5\n warmup_lr: 1e-6\n\n weight_decay: 0.05\n max_epoch: 5\n iters_per_epoch: 200\n num_workers: 4\n warmup_steps: 200\n\n seed: 42\n output_dir: \"output/minigpt4_stage2_finetune\"\n\n amp: True\n resume_ckpt_path: null\n\n evaluate: False \n train_splits: [\"train\"]\n\n device: \"cuda\"\n world_size: 1\n dist_url: \"env://\"\n distributed: True\n\n wandb_log: True\n job_name: minigpt4_finetune" }, { "path": "train_configs/minigptv2_finetune.yaml", "content": "model:\n arch: minigpt_v2\n model_type: pretrain\n max_txt_len: 1024\n image_size: 448\n end_sym: \"\"\n llama_model: \"/path/to/llama_checkpoint\"\n ckpt: \"/path/to/pretrained_checkpoint\"\n use_grad_checkpoint: True\n chat_template: True\n lora_r: 64\n lora_alpha: 16\n\ndatasets:\n multitask_conversation:\n batch_size: 2\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 50\n\n llava_conversation: \n batch_size: 2\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 30\n\n unnatural_instruction:\n batch_size: 1\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 10\n\n\n refvg:\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 40\n\n llava_detail:\n batch_size: 4\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 20\n\n llava_reason: \n batch_size: 4\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 80\n \n\n flickr_grounded_caption:\n batch_size: 2\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 80\n\n flickr_CaptionToPhrase:\n batch_size: 2\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 80\n\n flickr_ObjectToPhrase:\n batch_size: 2\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 80\n\n coco_caption:\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 10 \n\n \n textcaps_caption: #\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 30\n\n refcoco: \n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 25\n\n\n refcocop:\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 25\n\n refcocog:\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 25\n\n\n\n invrefcoco:\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 10\n\n invrefcocop:\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 10\n\n invrefcocog:\n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 10\n\n\n coco_vqa: \n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 15\n\n ok_vqa: \n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 8\n\n aok_vqa: \n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 12\n\n gqa: \n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 50\n\n ocrvqa: \n batch_size: 6\n vis_processor:\n train:\n name: \"blip2_image_train\"\n image_size: 448\n text_processor:\n train:\n name: \"blip_caption\"\n sample_ratio: 30\n\n\nrun:\n task: image_text_pretrain\n # optimizer\n lr_sched: \"linear_warmup_cosine_lr\"\n init_lr: 1e-5\n min_lr: 1e-6\n warmup_lr: 1e-6\n\n weight_decay: 0.05\n max_epoch: 50\n num_workers: 6\n warmup_steps: 1000\n iters_per_epoch: 1000\n\n seed: 42\n output_dir: \"/path/to/save_checkpoint\"\n\n amp: True\n resume_ckpt_path: null\n\n evaluate: False \n train_splits: [\"train\"]\n\n device: \"cuda\"\n world_size: 1\n dist_url: \"env://\"\n distributed: True\n\n wandb_log: True\n job_name: minigptv2_finetune\n" } ]