Repository: facebookresearch/detr Branch: main Commit: 29901c51d7fe Files: 42 Total size: 197.3 KB Directory structure: gitextract_f347uvq4/ ├── .github/ │ ├── CODE_OF_CONDUCT.md │ ├── CONTRIBUTING.md │ └── ISSUE_TEMPLATE/ │ ├── bugs.md │ ├── questions-help-support.md │ └── unexpected-problems-bugs.md ├── .gitignore ├── Dockerfile ├── LICENSE ├── README.md ├── d2/ │ ├── README.md │ ├── configs/ │ │ ├── detr_256_6_6_torchvision.yaml │ │ └── detr_segm_256_6_6_torchvision.yaml │ ├── converter.py │ ├── detr/ │ │ ├── __init__.py │ │ ├── config.py │ │ ├── dataset_mapper.py │ │ └── detr.py │ └── train_net.py ├── datasets/ │ ├── __init__.py │ ├── coco.py │ ├── coco_eval.py │ ├── coco_panoptic.py │ ├── panoptic_eval.py │ └── transforms.py ├── engine.py ├── hubconf.py ├── main.py ├── models/ │ ├── __init__.py │ ├── backbone.py │ ├── detr.py │ ├── matcher.py │ ├── position_encoding.py │ ├── segmentation.py │ └── transformer.py ├── requirements.txt ├── run_with_submitit.py ├── test_all.py ├── tox.ini └── util/ ├── __init__.py ├── box_ops.py ├── misc.py └── plot_utils.py ================================================ FILE CONTENTS ================================================ ================================================ FILE: .github/CODE_OF_CONDUCT.md ================================================ # Code of Conduct Facebook has adopted a Code of Conduct that we expect project participants to adhere to. Please read the [full text](https://code.fb.com/codeofconduct/) so that you can understand what actions will and will not be tolerated. ================================================ FILE: .github/CONTRIBUTING.md ================================================ # Contributing to DETR We want to make contributing to this project as easy and transparent as possible. ## Our Development Process Minor changes and improvements will be released on an ongoing basis. Larger changes (e.g., changesets implementing a new paper) will be released on a more periodic basis. ## Pull Requests We actively welcome your pull requests. 1. Fork the repo and create your branch from `master`. 2. If you've added code that should be tested, add tests. 3. If you've changed APIs, update the documentation. 4. Ensure the test suite passes. 5. Make sure your code lints. 6. If you haven't already, complete the Contributor License Agreement ("CLA"). ## Contributor License Agreement ("CLA") In order to accept your pull request, we need you to submit a CLA. You only need to do this once to work on any of Facebook's open source projects. Complete your CLA here: ## Issues We use GitHub issues to track public bugs. Please ensure your description is clear and has sufficient instructions to be able to reproduce the issue. Facebook has a [bounty program](https://www.facebook.com/whitehat/) for the safe disclosure of security bugs. In those cases, please go through the process outlined on that page and do not file a public issue. ## Coding Style * 4 spaces for indentation rather than tabs * 80 character line length * PEP8 formatting following [Black](https://black.readthedocs.io/en/stable/) ## License By contributing to DETR, you agree that your contributions will be licensed under the LICENSE file in the root directory of this source tree. ================================================ FILE: .github/ISSUE_TEMPLATE/bugs.md ================================================ --- name: "🐛 Bugs" about: Report bugs in DETR title: Please read & provide the following --- ## Instructions To Reproduce the 🐛 Bug: 1. what changes you made (`git diff`) or what code you wrote ``` ``` 2. what exact command you run: 3. what you observed (including __full logs__): ``` ``` 4. please simplify the steps as much as possible so they do not require additional resources to run, such as a private dataset. ## Expected behavior: If there are no obvious error in "what you observed" provided above, please tell us the expected behavior. ## Environment: Provide your environment information using the following command: ``` python -m torch.utils.collect_env ``` ================================================ FILE: .github/ISSUE_TEMPLATE/questions-help-support.md ================================================ --- name: "How to do something❓" about: How to do something using DETR? --- ## ❓ How to do something using DETR Describe what you want to do, including: 1. what inputs you will provide, if any: 2. what outputs you are expecting: NOTE: 1. Only general answers are provided. If you want to ask about "why X did not work", please use the [Unexpected behaviors](https://github.com/facebookresearch/detr/issues/new/choose) issue template. 2. About how to implement new models / new dataloader / new training logic, etc., check documentation first. 3. We do not answer general machine learning / computer vision questions that are not specific to DETR, such as how a model works, how to improve your training/make it converge, or what algorithm/methods can be used to achieve X. ================================================ FILE: .github/ISSUE_TEMPLATE/unexpected-problems-bugs.md ================================================ --- name: "Unexpected behaviors" about: Run into unexpected behaviors when using DETR title: Please read & provide the following --- If you do not know the root cause of the problem, and wish someone to help you, please post according to this template: ## Instructions To Reproduce the Issue: 1. what changes you made (`git diff`) or what code you wrote ``` ``` 2. what exact command you run: 3. what you observed (including __full logs__): ``` ``` 4. please simplify the steps as much as possible so they do not require additional resources to run, such as a private dataset. ## Expected behavior: If there are no obvious error in "what you observed" provided above, please tell us the expected behavior. If you expect the model to converge / work better, note that we do not give suggestions on how to train a new model. Only in one of the two conditions we will help with it: (1) You're unable to reproduce the results in DETR model zoo. (2) It indicates a DETR bug. ## Environment: Provide your environment information using the following command: ``` python -m torch.utils.collect_env ``` ================================================ FILE: .gitignore ================================================ .nfs* *.ipynb *.pyc .dumbo.json .DS_Store .*.swp *.pth **/__pycache__/** .ipynb_checkpoints/ datasets/data/ experiment-* *.tmp *.pkl **/.mypy_cache/* .mypy_cache/* not_tracked_dir/ .vscode ================================================ FILE: Dockerfile ================================================ FROM pytorch/pytorch:1.5-cuda10.1-cudnn7-runtime ENV DEBIAN_FRONTEND=noninteractive RUN apt-get update -qq && \ apt-get install -y git vim libgtk2.0-dev && \ rm -rf /var/cache/apk/* RUN pip --no-cache-dir install Cython COPY requirements.txt /workspace RUN pip --no-cache-dir install -r /workspace/requirements.txt ================================================ FILE: LICENSE ================================================ Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 1. Definitions. "License" shall mean the terms and conditions for use, reproduction, and distribution as defined by Sections 1 through 9 of this document. "Licensor" shall mean the copyright owner or entity authorized by the copyright owner that is granting the License. "Legal Entity" shall mean the union of the acting entity and all other entities that control, are controlled by, or are under common control with that entity. For the purposes of this definition, "control" means (i) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the outstanding shares, or (iii) beneficial ownership of such entity. "You" (or "Your") shall mean an individual or Legal Entity exercising permissions granted by this License. "Source" form shall mean the preferred form for making modifications, including but not limited to software source code, documentation source, and configuration files. "Object" form shall mean any form resulting from mechanical transformation or translation of a Source form, including but not limited to compiled object code, generated documentation, and conversions to other media types. "Work" shall mean the work of authorship, whether in Source or Object form, made available under the License, as indicated by a copyright notice that is included in or attached to the work (an example is provided in the Appendix below). "Derivative Works" shall mean any work, whether in Source or Object form, that is based on (or derived from) the Work and for which the editorial revisions, annotations, elaborations, or other modifications represent, as a whole, an original work of authorship. For the purposes of this License, Derivative Works shall not include works that remain separable from, or merely link (or bind by name) to the interfaces of, the Work and Derivative Works thereof. "Contribution" shall mean any work of authorship, including the original version of the Work and any modifications or additions to that Work or Derivative Works thereof, that is intentionally submitted to Licensor for inclusion in the Work by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Work, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution." "Contributor" shall mean Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by Licensor and subsequently incorporated within the Work. 2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form. 3. Grant of Patent License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this section) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Work, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Work to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Work or a Contribution incorporated within the Work constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for that Work shall terminate as of the date such litigation is filed. 4. Redistribution. You may reproduce and distribute copies of the Work or Derivative Works thereof in any medium, with or without modifications, and in Source or Object form, provided that You meet the following conditions: (a) You must give any other recipients of the Work or Derivative Works a copy of this License; and (b) You must cause any modified files to carry prominent notices stating that You changed the files; and (c) You must retain, in the Source form of any Derivative Works that You distribute, all copyright, patent, trademark, and attribution notices from the Source form of the Work, excluding those notices that do not pertain to any part of the Derivative Works; and (d) If the Work includes a "NOTICE" text file as part of its distribution, then any Derivative Works that You distribute must include a readable copy of the attribution notices contained within such NOTICE file, excluding those notices that do not pertain to any part of the Derivative Works, in at least one of the following places: within a NOTICE text file distributed as part of the Derivative Works; within the Source form or documentation, if provided along with the Derivative Works; or, within a display generated by the Derivative Works, if and wherever such third-party notices normally appear. The contents of the NOTICE file are for informational purposes only and do not modify the License. You may add Your own attribution notices within Derivative Works that You distribute, alongside or as an addendum to the NOTICE text from the Work, provided that such additional attribution notices cannot be construed as modifying the License. You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of Your modifications, or for any such Derivative Works as a whole, provided Your use, reproduction, and distribution of the Work otherwise complies with the conditions stated in this License. 5. Submission of Contributions. Unless You explicitly state otherwise, any Contribution intentionally submitted for inclusion in the Work by You to the Licensor shall be under the terms and conditions of this License, without any additional terms or conditions. Notwithstanding the above, nothing herein shall supersede or modify the terms of any separate license agreement you may have executed with Licensor regarding such Contributions. 6. Trademarks. This License does not grant permission to use the trade names, trademarks, service marks, or product names of the Licensor, except as required for reasonable and customary use in describing the origin of the Work and reproducing the content of the NOTICE file. 7. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Work and assume any risks associated with Your exercise of permissions under this License. 8. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Work (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages. 9. Accepting Warranty or Additional Liability. While redistributing the Work or Derivative Works thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability. END OF TERMS AND CONDITIONS APPENDIX: How to apply the Apache License to your work. To apply the Apache License to your work, attach the following boilerplate notice, with the fields enclosed by brackets "[]" replaced with your own identifying information. (Don't include the brackets!) The text should be enclosed in the appropriate comment syntax for the file format. We also recommend that a file or class name and description of purpose be included on the same "printed page" as the copyright notice for easier identification within third-party archives. Copyright 2020 - present, Facebook, Inc Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. ================================================ FILE: README.md ================================================ **DE⫶TR**: End-to-End Object Detection with Transformers ======== [![Support Ukraine](https://img.shields.io/badge/Support-Ukraine-FFD500?style=flat&labelColor=005BBB)](https://opensource.fb.com/support-ukraine) PyTorch training code and pretrained models for **DETR** (**DE**tection **TR**ansformer). We replace the full complex hand-crafted object detection pipeline with a Transformer, and match Faster R-CNN with a ResNet-50, obtaining **42 AP** on COCO using half the computation power (FLOPs) and the same number of parameters. Inference in 50 lines of PyTorch. ![DETR](.github/DETR.png) **What it is**. Unlike traditional computer vision techniques, DETR approaches object detection as a direct set prediction problem. It consists of a set-based global loss, which forces unique predictions via bipartite matching, and a Transformer encoder-decoder architecture. Given a fixed small set of learned object queries, DETR reasons about the relations of the objects and the global image context to directly output the final set of predictions in parallel. Due to this parallel nature, DETR is very fast and efficient. **About the code**. We believe that object detection should not be more difficult than classification, and should not require complex libraries for training and inference. DETR is very simple to implement and experiment with, and we provide a [standalone Colab Notebook](https://colab.research.google.com/github/facebookresearch/detr/blob/colab/notebooks/detr_demo.ipynb) showing how to do inference with DETR in only a few lines of PyTorch code. Training code follows this idea - it is not a library, but simply a [main.py](main.py) importing model and criterion definitions with standard training loops. Additionnally, we provide a Detectron2 wrapper in the d2/ folder. See the readme there for more information. For details see [End-to-End Object Detection with Transformers](https://ai.facebook.com/research/publications/end-to-end-object-detection-with-transformers) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. See our [blog post](https://ai.facebook.com/blog/end-to-end-object-detection-with-transformers/) to learn more about end to end object detection with transformers. # Model Zoo We provide baseline DETR and DETR-DC5 models, and plan to include more in future. AP is computed on COCO 2017 val5k, and inference time is over the first 100 val5k COCO images, with torchscript transformer.
name backbone schedule inf_time box AP url size
0 DETR R50 500 0.036 42.0 model | logs 159Mb
1 DETR-DC5 R50 500 0.083 43.3 model | logs 159Mb
2 DETR R101 500 0.050 43.5 model | logs 232Mb
3 DETR-DC5 R101 500 0.097 44.9 model | logs 232Mb
COCO val5k evaluation results can be found in this [gist](https://gist.github.com/szagoruyko/9c9ebb8455610958f7deaa27845d7918). The models are also available via torch hub, to load DETR R50 with pretrained weights simply do: ```python model = torch.hub.load('facebookresearch/detr:main', 'detr_resnet50', pretrained=True) ``` COCO panoptic val5k models:
name backbone box AP segm AP PQ url size
0 DETR R50 38.8 31.1 43.4 download 165Mb
1 DETR-DC5 R50 40.2 31.9 44.6 download 165Mb
2 DETR R101 40.1 33 45.1 download 237Mb
Checkout our [panoptic colab](https://colab.research.google.com/github/facebookresearch/detr/blob/colab/notebooks/DETR_panoptic.ipynb) to see how to use and visualize DETR's panoptic segmentation prediction. # Notebooks We provide a few notebooks in colab to help you get a grasp on DETR: * [DETR's hands on Colab Notebook](https://colab.research.google.com/github/facebookresearch/detr/blob/colab/notebooks/detr_attention.ipynb): Shows how to load a model from hub, generate predictions, then visualize the attention of the model (similar to the figures of the paper) * [Standalone Colab Notebook](https://colab.research.google.com/github/facebookresearch/detr/blob/colab/notebooks/detr_demo.ipynb): In this notebook, we demonstrate how to implement a simplified version of DETR from the grounds up in 50 lines of Python, then visualize the predictions. It is a good starting point if you want to gain better understanding the architecture and poke around before diving in the codebase. * [Panoptic Colab Notebook](https://colab.research.google.com/github/facebookresearch/detr/blob/colab/notebooks/DETR_panoptic.ipynb): Demonstrates how to use DETR for panoptic segmentation and plot the predictions. # Usage - Object detection There are no extra compiled components in DETR and package dependencies are minimal, so the code is very simple to use. We provide instructions how to install dependencies via conda. First, clone the repository locally: ``` git clone https://github.com/facebookresearch/detr.git ``` Then, install PyTorch 1.5+ and torchvision 0.6+: ``` conda install -c pytorch pytorch torchvision ``` Install pycocotools (for evaluation on COCO) and scipy (for training): ``` conda install cython scipy pip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI' ``` That's it, should be good to train and evaluate detection models. (optional) to work with panoptic install panopticapi: ``` pip install git+https://github.com/cocodataset/panopticapi.git ``` ## Data preparation Download and extract COCO 2017 train and val images with annotations from [http://cocodataset.org](http://cocodataset.org/#download). We expect the directory structure to be the following: ``` path/to/coco/ annotations/ # annotation json files train2017/ # train images val2017/ # val images ``` ## Training To train baseline DETR on a single node with 8 gpus for 300 epochs run: ``` python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --coco_path /path/to/coco ``` A single epoch takes 28 minutes, so 300 epoch training takes around 6 days on a single machine with 8 V100 cards. To ease reproduction of our results we provide [results and training logs](https://gist.github.com/szagoruyko/b4c3b2c3627294fc369b899987385a3f) for 150 epoch schedule (3 days on a single machine), achieving 39.5/60.3 AP/AP50. We train DETR with AdamW setting learning rate in the transformer to 1e-4 and 1e-5 in the backbone. Horizontal flips, scales and crops are used for augmentation. Images are rescaled to have min size 800 and max size 1333. The transformer is trained with dropout of 0.1, and the whole model is trained with grad clip of 0.1. ## Evaluation To evaluate DETR R50 on COCO val5k with a single GPU run: ``` python main.py --batch_size 2 --no_aux_loss --eval --resume https://dl.fbaipublicfiles.com/detr/detr-r50-e632da11.pth --coco_path /path/to/coco ``` We provide results for all DETR detection models in this [gist](https://gist.github.com/szagoruyko/9c9ebb8455610958f7deaa27845d7918). Note that numbers vary depending on batch size (number of images) per GPU. Non-DC5 models were trained with batch size 2, and DC5 with 1, so DC5 models show a significant drop in AP if evaluated with more than 1 image per GPU. ## Multinode training Distributed training is available via Slurm and [submitit](https://github.com/facebookincubator/submitit): ``` pip install submitit ``` Train baseline DETR-6-6 model on 4 nodes for 300 epochs: ``` python run_with_submitit.py --timeout 3000 --coco_path /path/to/coco ``` # Usage - Segmentation We show that it is relatively straightforward to extend DETR to predict segmentation masks. We mainly demonstrate strong panoptic segmentation results. ## Data preparation For panoptic segmentation, you need the panoptic annotations additionally to the coco dataset (see above for the coco dataset). You need to download and extract the [annotations](http://images.cocodataset.org/annotations/panoptic_annotations_trainval2017.zip). We expect the directory structure to be the following: ``` path/to/coco_panoptic/ annotations/ # annotation json files panoptic_train2017/ # train panoptic annotations panoptic_val2017/ # val panoptic annotations ``` ## Training We recommend training segmentation in two stages: first train DETR to detect all the boxes, and then train the segmentation head. For panoptic segmentation, DETR must learn to detect boxes for both stuff and things classes. You can train it on a single node with 8 gpus for 300 epochs with: ``` python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --coco_path /path/to/coco --coco_panoptic_path /path/to/coco_panoptic --dataset_file coco_panoptic --output_dir /output/path/box_model ``` For instance segmentation, you can simply train a normal box model (or used a pre-trained one we provide). Once you have a box model checkpoint, you need to freeze it, and train the segmentation head in isolation. For panoptic segmentation you can train on a single node with 8 gpus for 25 epochs: ``` python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --masks --epochs 25 --lr_drop 15 --coco_path /path/to/coco --coco_panoptic_path /path/to/coco_panoptic --dataset_file coco_panoptic --frozen_weights /output/path/box_model/checkpoint.pth --output_dir /output/path/segm_model ``` For instance segmentation only, simply remove the `dataset_file` and `coco_panoptic_path` arguments from the above command line. # License DETR is released under the Apache 2.0 license. Please see the [LICENSE](LICENSE) file for more information. # Contributing We actively welcome your pull requests! Please see [CONTRIBUTING.md](.github/CONTRIBUTING.md) and [CODE_OF_CONDUCT.md](.github/CODE_OF_CONDUCT.md) for more info. ================================================ FILE: d2/README.md ================================================ Detectron2 wrapper for DETR ======= We provide a Detectron2 wrapper for DETR, thus providing a way to better integrate it in the existing detection ecosystem. It can be used for example to easily leverage datasets or backbones provided in Detectron2. This wrapper currently supports only box detection, and is intended to be as close as possible to the original implementation, and we checked that it indeed match the results. Some notable facts and caveats: - The data augmentation matches DETR's original data augmentation. This required patching the RandomCrop augmentation from Detectron2, so you'll need a version from the master branch from June 24th 2020 or more recent. - To match DETR's original backbone initialization, we use the weights of a ResNet50 trained on imagenet using torchvision. This network uses a different pixel mean and std than most of the backbones available in Detectron2 by default, so extra care must be taken when switching to another one. Note that no other torchvision models are available in Detectron2 as of now, though it may change in the future. - The gradient clipping mode is "full_model", which is not the default in Detectron2. # Usage To install Detectron2, please follow the [official installation instructions](https://github.com/facebookresearch/detectron2/blob/master/INSTALL.md). ## Evaluating a model For convenience, we provide a conversion script to convert models trained by the main DETR training loop into the format of this wrapper. To download and convert the main Resnet50 model, simply do: ``` python converter.py --source_model https://dl.fbaipublicfiles.com/detr/detr-r50-e632da11.pth --output_model converted_model.pth ``` You can then evaluate it using: ``` python train_net.py --eval-only --config configs/detr_256_6_6_torchvision.yaml MODEL.WEIGHTS "converted_model.pth" ``` ## Training To train DETR on a single node with 8 gpus, simply use: ``` python train_net.py --config configs/detr_256_6_6_torchvision.yaml --num-gpus 8 ``` To fine-tune DETR for instance segmentation on a single node with 8 gpus, simply use: ``` python train_net.py --config configs/detr_segm_256_6_6_torchvision.yaml --num-gpus 8 MODEL.DETR.FROZEN_WEIGHTS ``` ================================================ FILE: d2/configs/detr_256_6_6_torchvision.yaml ================================================ MODEL: META_ARCHITECTURE: "Detr" WEIGHTS: "detectron2://ImageNetPretrained/torchvision/R-50.pkl" PIXEL_MEAN: [123.675, 116.280, 103.530] PIXEL_STD: [58.395, 57.120, 57.375] MASK_ON: False RESNETS: DEPTH: 50 STRIDE_IN_1X1: False OUT_FEATURES: ["res2", "res3", "res4", "res5"] DETR: GIOU_WEIGHT: 2.0 L1_WEIGHT: 5.0 NUM_OBJECT_QUERIES: 100 DATASETS: TRAIN: ("coco_2017_train",) TEST: ("coco_2017_val",) SOLVER: IMS_PER_BATCH: 64 BASE_LR: 0.0001 STEPS: (369600,) MAX_ITER: 554400 WARMUP_FACTOR: 1.0 WARMUP_ITERS: 10 WEIGHT_DECAY: 0.0001 OPTIMIZER: "ADAMW" BACKBONE_MULTIPLIER: 0.1 CLIP_GRADIENTS: ENABLED: True CLIP_TYPE: "full_model" CLIP_VALUE: 0.01 NORM_TYPE: 2.0 INPUT: MIN_SIZE_TRAIN: (480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800) CROP: ENABLED: True TYPE: "absolute_range" SIZE: (384, 600) FORMAT: "RGB" TEST: EVAL_PERIOD: 4000 DATALOADER: FILTER_EMPTY_ANNOTATIONS: False NUM_WORKERS: 4 VERSION: 2 ================================================ FILE: d2/configs/detr_segm_256_6_6_torchvision.yaml ================================================ MODEL: META_ARCHITECTURE: "Detr" # WEIGHTS: "detectron2://ImageNetPretrained/torchvision/R-50.pkl" PIXEL_MEAN: [123.675, 116.280, 103.530] PIXEL_STD: [58.395, 57.120, 57.375] MASK_ON: True RESNETS: DEPTH: 50 STRIDE_IN_1X1: False OUT_FEATURES: ["res2", "res3", "res4", "res5"] DETR: GIOU_WEIGHT: 2.0 L1_WEIGHT: 5.0 NUM_OBJECT_QUERIES: 100 FROZEN_WEIGHTS: '' DATASETS: TRAIN: ("coco_2017_train",) TEST: ("coco_2017_val",) SOLVER: IMS_PER_BATCH: 64 BASE_LR: 0.0001 STEPS: (55440,) MAX_ITER: 92400 WARMUP_FACTOR: 1.0 WARMUP_ITERS: 10 WEIGHT_DECAY: 0.0001 OPTIMIZER: "ADAMW" BACKBONE_MULTIPLIER: 0.1 CLIP_GRADIENTS: ENABLED: True CLIP_TYPE: "full_model" CLIP_VALUE: 0.01 NORM_TYPE: 2.0 INPUT: MIN_SIZE_TRAIN: (480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800) CROP: ENABLED: True TYPE: "absolute_range" SIZE: (384, 600) FORMAT: "RGB" TEST: EVAL_PERIOD: 4000 DATALOADER: FILTER_EMPTY_ANNOTATIONS: False NUM_WORKERS: 4 VERSION: 2 ================================================ FILE: d2/converter.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Helper script to convert models trained with the main version of DETR to be used with the Detectron2 version. """ import json import argparse import numpy as np import torch def parse_args(): parser = argparse.ArgumentParser("D2 model converter") parser.add_argument("--source_model", default="", type=str, help="Path or url to the DETR model to convert") parser.add_argument("--output_model", default="", type=str, help="Path where to save the converted model") return parser.parse_args() def main(): args = parse_args() # D2 expects contiguous classes, so we need to remap the 92 classes from DETR # fmt: off coco_idx = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27, 28, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 67, 70, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 84, 85, 86, 87, 88, 89, 90, 91] # fmt: on coco_idx = np.array(coco_idx) if args.source_model.startswith("https"): checkpoint = torch.hub.load_state_dict_from_url(args.source_model, map_location="cpu", check_hash=True) else: checkpoint = torch.load(args.source_model, map_location="cpu") model_to_convert = checkpoint["model"] model_converted = {} for k in model_to_convert.keys(): old_k = k if "backbone" in k: k = k.replace("backbone.0.body.", "") if "layer" not in k: k = "stem." + k for t in [1, 2, 3, 4]: k = k.replace(f"layer{t}", f"res{t + 1}") for t in [1, 2, 3]: k = k.replace(f"bn{t}", f"conv{t}.norm") k = k.replace("downsample.0", "shortcut") k = k.replace("downsample.1", "shortcut.norm") k = "backbone.0.backbone." + k k = "detr." + k print(old_k, "->", k) if "class_embed" in old_k: v = model_to_convert[old_k].detach() if v.shape[0] == 92: shape_old = v.shape model_converted[k] = v[coco_idx] print("Head conversion: changing shape from {} to {}".format(shape_old, model_converted[k].shape)) continue model_converted[k] = model_to_convert[old_k].detach() model_to_save = {"model": model_converted} torch.save(model_to_save, args.output_model) if __name__ == "__main__": main() ================================================ FILE: d2/detr/__init__.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved from .config import add_detr_config from .detr import Detr from .dataset_mapper import DetrDatasetMapper ================================================ FILE: d2/detr/config.py ================================================ # -*- coding: utf-8 -*- # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved from detectron2.config import CfgNode as CN def add_detr_config(cfg): """ Add config for DETR. """ cfg.MODEL.DETR = CN() cfg.MODEL.DETR.NUM_CLASSES = 80 # For Segmentation cfg.MODEL.DETR.FROZEN_WEIGHTS = '' # LOSS cfg.MODEL.DETR.GIOU_WEIGHT = 2.0 cfg.MODEL.DETR.L1_WEIGHT = 5.0 cfg.MODEL.DETR.DEEP_SUPERVISION = True cfg.MODEL.DETR.NO_OBJECT_WEIGHT = 0.1 # TRANSFORMER cfg.MODEL.DETR.NHEADS = 8 cfg.MODEL.DETR.DROPOUT = 0.1 cfg.MODEL.DETR.DIM_FEEDFORWARD = 2048 cfg.MODEL.DETR.ENC_LAYERS = 6 cfg.MODEL.DETR.DEC_LAYERS = 6 cfg.MODEL.DETR.PRE_NORM = False cfg.MODEL.DETR.HIDDEN_DIM = 256 cfg.MODEL.DETR.NUM_OBJECT_QUERIES = 100 cfg.SOLVER.OPTIMIZER = "ADAMW" cfg.SOLVER.BACKBONE_MULTIPLIER = 0.1 ================================================ FILE: d2/detr/dataset_mapper.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import copy import logging import numpy as np import torch from detectron2.data import detection_utils as utils from detectron2.data import transforms as T from detectron2.data.transforms import TransformGen __all__ = ["DetrDatasetMapper"] def build_transform_gen(cfg, is_train): """ Create a list of :class:`TransformGen` from config. Returns: list[TransformGen] """ if is_train: min_size = cfg.INPUT.MIN_SIZE_TRAIN max_size = cfg.INPUT.MAX_SIZE_TRAIN sample_style = cfg.INPUT.MIN_SIZE_TRAIN_SAMPLING else: min_size = cfg.INPUT.MIN_SIZE_TEST max_size = cfg.INPUT.MAX_SIZE_TEST sample_style = "choice" if sample_style == "range": assert len(min_size) == 2, "more than 2 ({}) min_size(s) are provided for ranges".format(len(min_size)) logger = logging.getLogger(__name__) tfm_gens = [] if is_train: tfm_gens.append(T.RandomFlip()) tfm_gens.append(T.ResizeShortestEdge(min_size, max_size, sample_style)) if is_train: logger.info("TransformGens used in training: " + str(tfm_gens)) return tfm_gens class DetrDatasetMapper: """ A callable which takes a dataset dict in Detectron2 Dataset format, and map it into a format used by DETR. The callable currently does the following: 1. Read the image from "file_name" 2. Applies geometric transforms to the image and annotation 3. Find and applies suitable cropping to the image and annotation 4. Prepare image and annotation to Tensors """ def __init__(self, cfg, is_train=True): if cfg.INPUT.CROP.ENABLED and is_train: self.crop_gen = [ T.ResizeShortestEdge([400, 500, 600], sample_style="choice"), T.RandomCrop(cfg.INPUT.CROP.TYPE, cfg.INPUT.CROP.SIZE), ] else: self.crop_gen = None self.mask_on = cfg.MODEL.MASK_ON self.tfm_gens = build_transform_gen(cfg, is_train) logging.getLogger(__name__).info( "Full TransformGens used in training: {}, crop: {}".format(str(self.tfm_gens), str(self.crop_gen)) ) self.img_format = cfg.INPUT.FORMAT self.is_train = is_train def __call__(self, dataset_dict): """ Args: dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format. Returns: dict: a format that builtin models in detectron2 accept """ dataset_dict = copy.deepcopy(dataset_dict) # it will be modified by code below image = utils.read_image(dataset_dict["file_name"], format=self.img_format) utils.check_image_size(dataset_dict, image) if self.crop_gen is None: image, transforms = T.apply_transform_gens(self.tfm_gens, image) else: if np.random.rand() > 0.5: image, transforms = T.apply_transform_gens(self.tfm_gens, image) else: image, transforms = T.apply_transform_gens( self.tfm_gens[:-1] + self.crop_gen + self.tfm_gens[-1:], image ) image_shape = image.shape[:2] # h, w # Pytorch's dataloader is efficient on torch.Tensor due to shared-memory, # but not efficient on large generic data structures due to the use of pickle & mp.Queue. # Therefore it's important to use torch.Tensor. dataset_dict["image"] = torch.as_tensor(np.ascontiguousarray(image.transpose(2, 0, 1))) if not self.is_train: # USER: Modify this if you want to keep them for some reason. dataset_dict.pop("annotations", None) return dataset_dict if "annotations" in dataset_dict: # USER: Modify this if you want to keep them for some reason. for anno in dataset_dict["annotations"]: if not self.mask_on: anno.pop("segmentation", None) anno.pop("keypoints", None) # USER: Implement additional transformations if you have other types of data annos = [ utils.transform_instance_annotations(obj, transforms, image_shape) for obj in dataset_dict.pop("annotations") if obj.get("iscrowd", 0) == 0 ] instances = utils.annotations_to_instances(annos, image_shape) dataset_dict["instances"] = utils.filter_empty_instances(instances) return dataset_dict ================================================ FILE: d2/detr/detr.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import logging import math from typing import List import numpy as np import torch import torch.distributed as dist import torch.nn.functional as F from scipy.optimize import linear_sum_assignment from torch import nn from detectron2.layers import ShapeSpec from detectron2.modeling import META_ARCH_REGISTRY, build_backbone, detector_postprocess from detectron2.structures import Boxes, ImageList, Instances, BitMasks, PolygonMasks from detectron2.utils.logger import log_first_n from fvcore.nn import giou_loss, smooth_l1_loss from models.backbone import Joiner from models.detr import DETR, SetCriterion from models.matcher import HungarianMatcher from models.position_encoding import PositionEmbeddingSine from models.transformer import Transformer from models.segmentation import DETRsegm, PostProcessPanoptic, PostProcessSegm from util.box_ops import box_cxcywh_to_xyxy, box_xyxy_to_cxcywh from util.misc import NestedTensor from datasets.coco import convert_coco_poly_to_mask __all__ = ["Detr"] class MaskedBackbone(nn.Module): """ This is a thin wrapper around D2's backbone to provide padding masking""" def __init__(self, cfg): super().__init__() self.backbone = build_backbone(cfg) backbone_shape = self.backbone.output_shape() self.feature_strides = [backbone_shape[f].stride for f in backbone_shape.keys()] self.num_channels = backbone_shape[list(backbone_shape.keys())[-1]].channels def forward(self, images): features = self.backbone(images.tensor) masks = self.mask_out_padding( [features_per_level.shape for features_per_level in features.values()], images.image_sizes, images.tensor.device, ) assert len(features) == len(masks) for i, k in enumerate(features.keys()): features[k] = NestedTensor(features[k], masks[i]) return features def mask_out_padding(self, feature_shapes, image_sizes, device): masks = [] assert len(feature_shapes) == len(self.feature_strides) for idx, shape in enumerate(feature_shapes): N, _, H, W = shape masks_per_feature_level = torch.ones((N, H, W), dtype=torch.bool, device=device) for img_idx, (h, w) in enumerate(image_sizes): masks_per_feature_level[ img_idx, : int(np.ceil(float(h) / self.feature_strides[idx])), : int(np.ceil(float(w) / self.feature_strides[idx])), ] = 0 masks.append(masks_per_feature_level) return masks @META_ARCH_REGISTRY.register() class Detr(nn.Module): """ Implement Detr """ def __init__(self, cfg): super().__init__() self.device = torch.device(cfg.MODEL.DEVICE) self.num_classes = cfg.MODEL.DETR.NUM_CLASSES self.mask_on = cfg.MODEL.MASK_ON hidden_dim = cfg.MODEL.DETR.HIDDEN_DIM num_queries = cfg.MODEL.DETR.NUM_OBJECT_QUERIES # Transformer parameters: nheads = cfg.MODEL.DETR.NHEADS dropout = cfg.MODEL.DETR.DROPOUT dim_feedforward = cfg.MODEL.DETR.DIM_FEEDFORWARD enc_layers = cfg.MODEL.DETR.ENC_LAYERS dec_layers = cfg.MODEL.DETR.DEC_LAYERS pre_norm = cfg.MODEL.DETR.PRE_NORM # Loss parameters: giou_weight = cfg.MODEL.DETR.GIOU_WEIGHT l1_weight = cfg.MODEL.DETR.L1_WEIGHT deep_supervision = cfg.MODEL.DETR.DEEP_SUPERVISION no_object_weight = cfg.MODEL.DETR.NO_OBJECT_WEIGHT N_steps = hidden_dim // 2 d2_backbone = MaskedBackbone(cfg) backbone = Joiner(d2_backbone, PositionEmbeddingSine(N_steps, normalize=True)) backbone.num_channels = d2_backbone.num_channels transformer = Transformer( d_model=hidden_dim, dropout=dropout, nhead=nheads, dim_feedforward=dim_feedforward, num_encoder_layers=enc_layers, num_decoder_layers=dec_layers, normalize_before=pre_norm, return_intermediate_dec=deep_supervision, ) self.detr = DETR( backbone, transformer, num_classes=self.num_classes, num_queries=num_queries, aux_loss=deep_supervision ) if self.mask_on: frozen_weights = cfg.MODEL.DETR.FROZEN_WEIGHTS if frozen_weights != '': print("LOAD pre-trained weights") weight = torch.load(frozen_weights, map_location=lambda storage, loc: storage)['model'] new_weight = {} for k, v in weight.items(): if 'detr.' in k: new_weight[k.replace('detr.', '')] = v else: print(f"Skipping loading weight {k} from frozen model") del weight self.detr.load_state_dict(new_weight) del new_weight self.detr = DETRsegm(self.detr, freeze_detr=(frozen_weights != '')) self.seg_postprocess = PostProcessSegm self.detr.to(self.device) # building criterion matcher = HungarianMatcher(cost_class=1, cost_bbox=l1_weight, cost_giou=giou_weight) weight_dict = {"loss_ce": 1, "loss_bbox": l1_weight} weight_dict["loss_giou"] = giou_weight if deep_supervision: aux_weight_dict = {} for i in range(dec_layers - 1): aux_weight_dict.update({k + f"_{i}": v for k, v in weight_dict.items()}) weight_dict.update(aux_weight_dict) losses = ["labels", "boxes", "cardinality"] if self.mask_on: losses += ["masks"] self.criterion = SetCriterion( self.num_classes, matcher=matcher, weight_dict=weight_dict, eos_coef=no_object_weight, losses=losses, ) self.criterion.to(self.device) pixel_mean = torch.Tensor(cfg.MODEL.PIXEL_MEAN).to(self.device).view(3, 1, 1) pixel_std = torch.Tensor(cfg.MODEL.PIXEL_STD).to(self.device).view(3, 1, 1) self.normalizer = lambda x: (x - pixel_mean) / pixel_std self.to(self.device) def forward(self, batched_inputs): """ Args: batched_inputs: a list, batched outputs of :class:`DatasetMapper` . Each item in the list contains the inputs for one image. For now, each item in the list is a dict that contains: * image: Tensor, image in (C, H, W) format. * instances: Instances Other information that's included in the original dicts, such as: * "height", "width" (int): the output resolution of the model, used in inference. See :meth:`postprocess` for details. Returns: dict[str: Tensor]: mapping from a named loss to a tensor storing the loss. Used during training only. """ images = self.preprocess_image(batched_inputs) output = self.detr(images) if self.training: gt_instances = [x["instances"].to(self.device) for x in batched_inputs] targets = self.prepare_targets(gt_instances) loss_dict = self.criterion(output, targets) weight_dict = self.criterion.weight_dict for k in loss_dict.keys(): if k in weight_dict: loss_dict[k] *= weight_dict[k] return loss_dict else: box_cls = output["pred_logits"] box_pred = output["pred_boxes"] mask_pred = output["pred_masks"] if self.mask_on else None results = self.inference(box_cls, box_pred, mask_pred, images.image_sizes) processed_results = [] for results_per_image, input_per_image, image_size in zip(results, batched_inputs, images.image_sizes): height = input_per_image.get("height", image_size[0]) width = input_per_image.get("width", image_size[1]) r = detector_postprocess(results_per_image, height, width) processed_results.append({"instances": r}) return processed_results def prepare_targets(self, targets): new_targets = [] for targets_per_image in targets: h, w = targets_per_image.image_size image_size_xyxy = torch.as_tensor([w, h, w, h], dtype=torch.float, device=self.device) gt_classes = targets_per_image.gt_classes gt_boxes = targets_per_image.gt_boxes.tensor / image_size_xyxy gt_boxes = box_xyxy_to_cxcywh(gt_boxes) new_targets.append({"labels": gt_classes, "boxes": gt_boxes}) if self.mask_on and hasattr(targets_per_image, 'gt_masks'): gt_masks = targets_per_image.gt_masks gt_masks = convert_coco_poly_to_mask(gt_masks.polygons, h, w) new_targets[-1].update({'masks': gt_masks}) return new_targets def inference(self, box_cls, box_pred, mask_pred, image_sizes): """ Arguments: box_cls (Tensor): tensor of shape (batch_size, num_queries, K). The tensor predicts the classification probability for each query. box_pred (Tensor): tensors of shape (batch_size, num_queries, 4). The tensor predicts 4-vector (x,y,w,h) box regression values for every queryx image_sizes (List[torch.Size]): the input image sizes Returns: results (List[Instances]): a list of #images elements. """ assert len(box_cls) == len(image_sizes) results = [] # For each box we assign the best class or the second best if the best on is `no_object`. scores, labels = F.softmax(box_cls, dim=-1)[:, :, :-1].max(-1) for i, (scores_per_image, labels_per_image, box_pred_per_image, image_size) in enumerate(zip( scores, labels, box_pred, image_sizes )): result = Instances(image_size) result.pred_boxes = Boxes(box_cxcywh_to_xyxy(box_pred_per_image)) result.pred_boxes.scale(scale_x=image_size[1], scale_y=image_size[0]) if self.mask_on: mask = F.interpolate(mask_pred[i].unsqueeze(0), size=image_size, mode='bilinear', align_corners=False) mask = mask[0].sigmoid() > 0.5 B, N, H, W = mask_pred.shape mask = BitMasks(mask.cpu()).crop_and_resize(result.pred_boxes.tensor.cpu(), 32) result.pred_masks = mask.unsqueeze(1).to(mask_pred[0].device) result.scores = scores_per_image result.pred_classes = labels_per_image results.append(result) return results def preprocess_image(self, batched_inputs): """ Normalize, pad and batch the input images. """ images = [self.normalizer(x["image"].to(self.device)) for x in batched_inputs] images = ImageList.from_tensors(images) return images ================================================ FILE: d2/train_net.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ DETR Training Script. This script is a simplified version of the training script in detectron2/tools. """ import os import sys import itertools # fmt: off sys.path.insert(1, os.path.join(sys.path[0], '..')) # fmt: on import time from typing import Any, Dict, List, Set import torch import detectron2.utils.comm as comm from d2.detr import DetrDatasetMapper, add_detr_config from detectron2.checkpoint import DetectionCheckpointer from detectron2.config import get_cfg from detectron2.data import MetadataCatalog, build_detection_train_loader from detectron2.engine import DefaultTrainer, default_argument_parser, default_setup, launch from detectron2.evaluation import COCOEvaluator, verify_results from detectron2.solver.build import maybe_add_gradient_clipping class Trainer(DefaultTrainer): """ Extension of the Trainer class adapted to DETR. """ @classmethod def build_evaluator(cls, cfg, dataset_name, output_folder=None): """ Create evaluator(s) for a given dataset. This uses the special metadata "evaluator_type" associated with each builtin dataset. For your own dataset, you can simply create an evaluator manually in your script and do not have to worry about the hacky if-else logic here. """ if output_folder is None: output_folder = os.path.join(cfg.OUTPUT_DIR, "inference") return COCOEvaluator(dataset_name, cfg, True, output_folder) @classmethod def build_train_loader(cls, cfg): if "Detr" == cfg.MODEL.META_ARCHITECTURE: mapper = DetrDatasetMapper(cfg, True) else: mapper = None return build_detection_train_loader(cfg, mapper=mapper) @classmethod def build_optimizer(cls, cfg, model): params: List[Dict[str, Any]] = [] memo: Set[torch.nn.parameter.Parameter] = set() for key, value in model.named_parameters(recurse=True): if not value.requires_grad: continue # Avoid duplicating parameters if value in memo: continue memo.add(value) lr = cfg.SOLVER.BASE_LR weight_decay = cfg.SOLVER.WEIGHT_DECAY if "backbone" in key: lr = lr * cfg.SOLVER.BACKBONE_MULTIPLIER params += [{"params": [value], "lr": lr, "weight_decay": weight_decay}] def maybe_add_full_model_gradient_clipping(optim): # optim: the optimizer class # detectron2 doesn't have full model gradient clipping now clip_norm_val = cfg.SOLVER.CLIP_GRADIENTS.CLIP_VALUE enable = ( cfg.SOLVER.CLIP_GRADIENTS.ENABLED and cfg.SOLVER.CLIP_GRADIENTS.CLIP_TYPE == "full_model" and clip_norm_val > 0.0 ) class FullModelGradientClippingOptimizer(optim): def step(self, closure=None): all_params = itertools.chain(*[x["params"] for x in self.param_groups]) torch.nn.utils.clip_grad_norm_(all_params, clip_norm_val) super().step(closure=closure) return FullModelGradientClippingOptimizer if enable else optim optimizer_type = cfg.SOLVER.OPTIMIZER if optimizer_type == "SGD": optimizer = maybe_add_full_model_gradient_clipping(torch.optim.SGD)( params, cfg.SOLVER.BASE_LR, momentum=cfg.SOLVER.MOMENTUM ) elif optimizer_type == "ADAMW": optimizer = maybe_add_full_model_gradient_clipping(torch.optim.AdamW)( params, cfg.SOLVER.BASE_LR ) else: raise NotImplementedError(f"no optimizer type {optimizer_type}") if not cfg.SOLVER.CLIP_GRADIENTS.CLIP_TYPE == "full_model": optimizer = maybe_add_gradient_clipping(cfg, optimizer) return optimizer def setup(args): """ Create configs and perform basic setups. """ cfg = get_cfg() add_detr_config(cfg) cfg.merge_from_file(args.config_file) cfg.merge_from_list(args.opts) cfg.freeze() default_setup(cfg, args) return cfg def main(args): cfg = setup(args) if args.eval_only: model = Trainer.build_model(cfg) DetectionCheckpointer(model, save_dir=cfg.OUTPUT_DIR).resume_or_load(cfg.MODEL.WEIGHTS, resume=args.resume) res = Trainer.test(cfg, model) if comm.is_main_process(): verify_results(cfg, res) return res trainer = Trainer(cfg) trainer.resume_or_load(resume=args.resume) return trainer.train() if __name__ == "__main__": args = default_argument_parser().parse_args() print("Command Line Args:", args) launch( main, args.num_gpus, num_machines=args.num_machines, machine_rank=args.machine_rank, dist_url=args.dist_url, args=(args,), ) ================================================ FILE: datasets/__init__.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import torch.utils.data import torchvision from .coco import build as build_coco def get_coco_api_from_dataset(dataset): for _ in range(10): # if isinstance(dataset, torchvision.datasets.CocoDetection): # break if isinstance(dataset, torch.utils.data.Subset): dataset = dataset.dataset if isinstance(dataset, torchvision.datasets.CocoDetection): return dataset.coco def build_dataset(image_set, args): if args.dataset_file == 'coco': return build_coco(image_set, args) if args.dataset_file == 'coco_panoptic': # to avoid making panopticapi required for coco from .coco_panoptic import build as build_coco_panoptic return build_coco_panoptic(image_set, args) raise ValueError(f'dataset {args.dataset_file} not supported') ================================================ FILE: datasets/coco.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ COCO dataset which returns image_id for evaluation. Mostly copy-paste from https://github.com/pytorch/vision/blob/13b35ff/references/detection/coco_utils.py """ from pathlib import Path import torch import torch.utils.data import torchvision from pycocotools import mask as coco_mask import datasets.transforms as T class CocoDetection(torchvision.datasets.CocoDetection): def __init__(self, img_folder, ann_file, transforms, return_masks): super(CocoDetection, self).__init__(img_folder, ann_file) self._transforms = transforms self.prepare = ConvertCocoPolysToMask(return_masks) def __getitem__(self, idx): img, target = super(CocoDetection, self).__getitem__(idx) image_id = self.ids[idx] target = {'image_id': image_id, 'annotations': target} img, target = self.prepare(img, target) if self._transforms is not None: img, target = self._transforms(img, target) return img, target def convert_coco_poly_to_mask(segmentations, height, width): masks = [] for polygons in segmentations: rles = coco_mask.frPyObjects(polygons, height, width) mask = coco_mask.decode(rles) if len(mask.shape) < 3: mask = mask[..., None] mask = torch.as_tensor(mask, dtype=torch.uint8) mask = mask.any(dim=2) masks.append(mask) if masks: masks = torch.stack(masks, dim=0) else: masks = torch.zeros((0, height, width), dtype=torch.uint8) return masks class ConvertCocoPolysToMask(object): def __init__(self, return_masks=False): self.return_masks = return_masks def __call__(self, image, target): w, h = image.size image_id = target["image_id"] image_id = torch.tensor([image_id]) anno = target["annotations"] anno = [obj for obj in anno if 'iscrowd' not in obj or obj['iscrowd'] == 0] boxes = [obj["bbox"] for obj in anno] # guard against no boxes via resizing boxes = torch.as_tensor(boxes, dtype=torch.float32).reshape(-1, 4) boxes[:, 2:] += boxes[:, :2] boxes[:, 0::2].clamp_(min=0, max=w) boxes[:, 1::2].clamp_(min=0, max=h) classes = [obj["category_id"] for obj in anno] classes = torch.tensor(classes, dtype=torch.int64) if self.return_masks: segmentations = [obj["segmentation"] for obj in anno] masks = convert_coco_poly_to_mask(segmentations, h, w) keypoints = None if anno and "keypoints" in anno[0]: keypoints = [obj["keypoints"] for obj in anno] keypoints = torch.as_tensor(keypoints, dtype=torch.float32) num_keypoints = keypoints.shape[0] if num_keypoints: keypoints = keypoints.view(num_keypoints, -1, 3) keep = (boxes[:, 3] > boxes[:, 1]) & (boxes[:, 2] > boxes[:, 0]) boxes = boxes[keep] classes = classes[keep] if self.return_masks: masks = masks[keep] if keypoints is not None: keypoints = keypoints[keep] target = {} target["boxes"] = boxes target["labels"] = classes if self.return_masks: target["masks"] = masks target["image_id"] = image_id if keypoints is not None: target["keypoints"] = keypoints # for conversion to coco api area = torch.tensor([obj["area"] for obj in anno]) iscrowd = torch.tensor([obj["iscrowd"] if "iscrowd" in obj else 0 for obj in anno]) target["area"] = area[keep] target["iscrowd"] = iscrowd[keep] target["orig_size"] = torch.as_tensor([int(h), int(w)]) target["size"] = torch.as_tensor([int(h), int(w)]) return image, target def make_coco_transforms(image_set): normalize = T.Compose([ T.ToTensor(), T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ]) scales = [480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800] if image_set == 'train': return T.Compose([ T.RandomHorizontalFlip(), T.RandomSelect( T.RandomResize(scales, max_size=1333), T.Compose([ T.RandomResize([400, 500, 600]), T.RandomSizeCrop(384, 600), T.RandomResize(scales, max_size=1333), ]) ), normalize, ]) if image_set == 'val': return T.Compose([ T.RandomResize([800], max_size=1333), normalize, ]) raise ValueError(f'unknown {image_set}') def build(image_set, args): root = Path(args.coco_path) assert root.exists(), f'provided COCO path {root} does not exist' mode = 'instances' PATHS = { "train": (root / "train2017", root / "annotations" / f'{mode}_train2017.json'), "val": (root / "val2017", root / "annotations" / f'{mode}_val2017.json'), } img_folder, ann_file = PATHS[image_set] dataset = CocoDetection(img_folder, ann_file, transforms=make_coco_transforms(image_set), return_masks=args.masks) return dataset ================================================ FILE: datasets/coco_eval.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ COCO evaluator that works in distributed mode. Mostly copy-paste from https://github.com/pytorch/vision/blob/edfd5a7/references/detection/coco_eval.py The difference is that there is less copy-pasting from pycocotools in the end of the file, as python3 can suppress prints with contextlib """ import os import contextlib import copy import numpy as np import torch from pycocotools.cocoeval import COCOeval from pycocotools.coco import COCO import pycocotools.mask as mask_util from util.misc import all_gather class CocoEvaluator(object): def __init__(self, coco_gt, iou_types): assert isinstance(iou_types, (list, tuple)) coco_gt = copy.deepcopy(coco_gt) self.coco_gt = coco_gt self.iou_types = iou_types self.coco_eval = {} for iou_type in iou_types: self.coco_eval[iou_type] = COCOeval(coco_gt, iouType=iou_type) self.img_ids = [] self.eval_imgs = {k: [] for k in iou_types} def update(self, predictions): img_ids = list(np.unique(list(predictions.keys()))) self.img_ids.extend(img_ids) for iou_type in self.iou_types: results = self.prepare(predictions, iou_type) # suppress pycocotools prints with open(os.devnull, 'w') as devnull: with contextlib.redirect_stdout(devnull): coco_dt = COCO.loadRes(self.coco_gt, results) if results else COCO() coco_eval = self.coco_eval[iou_type] coco_eval.cocoDt = coco_dt coco_eval.params.imgIds = list(img_ids) img_ids, eval_imgs = evaluate(coco_eval) self.eval_imgs[iou_type].append(eval_imgs) def synchronize_between_processes(self): for iou_type in self.iou_types: self.eval_imgs[iou_type] = np.concatenate(self.eval_imgs[iou_type], 2) create_common_coco_eval(self.coco_eval[iou_type], self.img_ids, self.eval_imgs[iou_type]) def accumulate(self): for coco_eval in self.coco_eval.values(): coco_eval.accumulate() def summarize(self): for iou_type, coco_eval in self.coco_eval.items(): print("IoU metric: {}".format(iou_type)) coco_eval.summarize() def prepare(self, predictions, iou_type): if iou_type == "bbox": return self.prepare_for_coco_detection(predictions) elif iou_type == "segm": return self.prepare_for_coco_segmentation(predictions) elif iou_type == "keypoints": return self.prepare_for_coco_keypoint(predictions) else: raise ValueError("Unknown iou type {}".format(iou_type)) def prepare_for_coco_detection(self, predictions): coco_results = [] for original_id, prediction in predictions.items(): if len(prediction) == 0: continue boxes = prediction["boxes"] boxes = convert_to_xywh(boxes).tolist() scores = prediction["scores"].tolist() labels = prediction["labels"].tolist() coco_results.extend( [ { "image_id": original_id, "category_id": labels[k], "bbox": box, "score": scores[k], } for k, box in enumerate(boxes) ] ) return coco_results def prepare_for_coco_segmentation(self, predictions): coco_results = [] for original_id, prediction in predictions.items(): if len(prediction) == 0: continue scores = prediction["scores"] labels = prediction["labels"] masks = prediction["masks"] masks = masks > 0.5 scores = prediction["scores"].tolist() labels = prediction["labels"].tolist() rles = [ mask_util.encode(np.array(mask[0, :, :, np.newaxis], dtype=np.uint8, order="F"))[0] for mask in masks ] for rle in rles: rle["counts"] = rle["counts"].decode("utf-8") coco_results.extend( [ { "image_id": original_id, "category_id": labels[k], "segmentation": rle, "score": scores[k], } for k, rle in enumerate(rles) ] ) return coco_results def prepare_for_coco_keypoint(self, predictions): coco_results = [] for original_id, prediction in predictions.items(): if len(prediction) == 0: continue boxes = prediction["boxes"] boxes = convert_to_xywh(boxes).tolist() scores = prediction["scores"].tolist() labels = prediction["labels"].tolist() keypoints = prediction["keypoints"] keypoints = keypoints.flatten(start_dim=1).tolist() coco_results.extend( [ { "image_id": original_id, "category_id": labels[k], 'keypoints': keypoint, "score": scores[k], } for k, keypoint in enumerate(keypoints) ] ) return coco_results def convert_to_xywh(boxes): xmin, ymin, xmax, ymax = boxes.unbind(1) return torch.stack((xmin, ymin, xmax - xmin, ymax - ymin), dim=1) def merge(img_ids, eval_imgs): all_img_ids = all_gather(img_ids) all_eval_imgs = all_gather(eval_imgs) merged_img_ids = [] for p in all_img_ids: merged_img_ids.extend(p) merged_eval_imgs = [] for p in all_eval_imgs: merged_eval_imgs.append(p) merged_img_ids = np.array(merged_img_ids) merged_eval_imgs = np.concatenate(merged_eval_imgs, 2) # keep only unique (and in sorted order) images merged_img_ids, idx = np.unique(merged_img_ids, return_index=True) merged_eval_imgs = merged_eval_imgs[..., idx] return merged_img_ids, merged_eval_imgs def create_common_coco_eval(coco_eval, img_ids, eval_imgs): img_ids, eval_imgs = merge(img_ids, eval_imgs) img_ids = list(img_ids) eval_imgs = list(eval_imgs.flatten()) coco_eval.evalImgs = eval_imgs coco_eval.params.imgIds = img_ids coco_eval._paramsEval = copy.deepcopy(coco_eval.params) ################################################################# # From pycocotools, just removed the prints and fixed # a Python3 bug about unicode not defined ################################################################# def evaluate(self): ''' Run per image evaluation on given images and store results (a list of dict) in self.evalImgs :return: None ''' # tic = time.time() # print('Running per image evaluation...') p = self.params # add backward compatibility if useSegm is specified in params if p.useSegm is not None: p.iouType = 'segm' if p.useSegm == 1 else 'bbox' print('useSegm (deprecated) is not None. Running {} evaluation'.format(p.iouType)) # print('Evaluate annotation type *{}*'.format(p.iouType)) p.imgIds = list(np.unique(p.imgIds)) if p.useCats: p.catIds = list(np.unique(p.catIds)) p.maxDets = sorted(p.maxDets) self.params = p self._prepare() # loop through images, area range, max detection number catIds = p.catIds if p.useCats else [-1] if p.iouType == 'segm' or p.iouType == 'bbox': computeIoU = self.computeIoU elif p.iouType == 'keypoints': computeIoU = self.computeOks self.ious = { (imgId, catId): computeIoU(imgId, catId) for imgId in p.imgIds for catId in catIds} evaluateImg = self.evaluateImg maxDet = p.maxDets[-1] evalImgs = [ evaluateImg(imgId, catId, areaRng, maxDet) for catId in catIds for areaRng in p.areaRng for imgId in p.imgIds ] # this is NOT in the pycocotools code, but could be done outside evalImgs = np.asarray(evalImgs).reshape(len(catIds), len(p.areaRng), len(p.imgIds)) self._paramsEval = copy.deepcopy(self.params) # toc = time.time() # print('DONE (t={:0.2f}s).'.format(toc-tic)) return p.imgIds, evalImgs ################################################################# # end of straight copy from pycocotools, just removing the prints ################################################################# ================================================ FILE: datasets/coco_panoptic.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import json from pathlib import Path import numpy as np import torch from PIL import Image from panopticapi.utils import rgb2id from util.box_ops import masks_to_boxes from .coco import make_coco_transforms class CocoPanoptic: def __init__(self, img_folder, ann_folder, ann_file, transforms=None, return_masks=True): with open(ann_file, 'r') as f: self.coco = json.load(f) # sort 'images' field so that they are aligned with 'annotations' # i.e., in alphabetical order self.coco['images'] = sorted(self.coco['images'], key=lambda x: x['id']) # sanity check if "annotations" in self.coco: for img, ann in zip(self.coco['images'], self.coco['annotations']): assert img['file_name'][:-4] == ann['file_name'][:-4] self.img_folder = img_folder self.ann_folder = ann_folder self.ann_file = ann_file self.transforms = transforms self.return_masks = return_masks def __getitem__(self, idx): ann_info = self.coco['annotations'][idx] if "annotations" in self.coco else self.coco['images'][idx] img_path = Path(self.img_folder) / ann_info['file_name'].replace('.png', '.jpg') ann_path = Path(self.ann_folder) / ann_info['file_name'] img = Image.open(img_path).convert('RGB') w, h = img.size if "segments_info" in ann_info: masks = np.asarray(Image.open(ann_path), dtype=np.uint32) masks = rgb2id(masks) ids = np.array([ann['id'] for ann in ann_info['segments_info']]) masks = masks == ids[:, None, None] masks = torch.as_tensor(masks, dtype=torch.uint8) labels = torch.tensor([ann['category_id'] for ann in ann_info['segments_info']], dtype=torch.int64) target = {} target['image_id'] = torch.tensor([ann_info['image_id'] if "image_id" in ann_info else ann_info["id"]]) if self.return_masks: target['masks'] = masks target['labels'] = labels target["boxes"] = masks_to_boxes(masks) target['size'] = torch.as_tensor([int(h), int(w)]) target['orig_size'] = torch.as_tensor([int(h), int(w)]) if "segments_info" in ann_info: for name in ['iscrowd', 'area']: target[name] = torch.tensor([ann[name] for ann in ann_info['segments_info']]) if self.transforms is not None: img, target = self.transforms(img, target) return img, target def __len__(self): return len(self.coco['images']) def get_height_and_width(self, idx): img_info = self.coco['images'][idx] height = img_info['height'] width = img_info['width'] return height, width def build(image_set, args): img_folder_root = Path(args.coco_path) ann_folder_root = Path(args.coco_panoptic_path) assert img_folder_root.exists(), f'provided COCO path {img_folder_root} does not exist' assert ann_folder_root.exists(), f'provided COCO path {ann_folder_root} does not exist' mode = 'panoptic' PATHS = { "train": ("train2017", Path("annotations") / f'{mode}_train2017.json'), "val": ("val2017", Path("annotations") / f'{mode}_val2017.json'), } img_folder, ann_file = PATHS[image_set] img_folder_path = img_folder_root / img_folder ann_folder = ann_folder_root / f'{mode}_{img_folder}' ann_file = ann_folder_root / ann_file dataset = CocoPanoptic(img_folder_path, ann_folder, ann_file, transforms=make_coco_transforms(image_set), return_masks=args.masks) return dataset ================================================ FILE: datasets/panoptic_eval.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import json import os import util.misc as utils try: from panopticapi.evaluation import pq_compute except ImportError: pass class PanopticEvaluator(object): def __init__(self, ann_file, ann_folder, output_dir="panoptic_eval"): self.gt_json = ann_file self.gt_folder = ann_folder if utils.is_main_process(): if not os.path.exists(output_dir): os.mkdir(output_dir) self.output_dir = output_dir self.predictions = [] def update(self, predictions): for p in predictions: with open(os.path.join(self.output_dir, p["file_name"]), "wb") as f: f.write(p.pop("png_string")) self.predictions += predictions def synchronize_between_processes(self): all_predictions = utils.all_gather(self.predictions) merged_predictions = [] for p in all_predictions: merged_predictions += p self.predictions = merged_predictions def summarize(self): if utils.is_main_process(): json_data = {"annotations": self.predictions} predictions_json = os.path.join(self.output_dir, "predictions.json") with open(predictions_json, "w") as f: f.write(json.dumps(json_data)) return pq_compute(self.gt_json, predictions_json, gt_folder=self.gt_folder, pred_folder=self.output_dir) return None ================================================ FILE: datasets/transforms.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Transforms and data augmentation for both image + bbox. """ import random import PIL import torch import torchvision.transforms as T import torchvision.transforms.functional as F from util.box_ops import box_xyxy_to_cxcywh from util.misc import interpolate def crop(image, target, region): cropped_image = F.crop(image, *region) target = target.copy() i, j, h, w = region # should we do something wrt the original size? target["size"] = torch.tensor([h, w]) fields = ["labels", "area", "iscrowd"] if "boxes" in target: boxes = target["boxes"] max_size = torch.as_tensor([w, h], dtype=torch.float32) cropped_boxes = boxes - torch.as_tensor([j, i, j, i]) cropped_boxes = torch.min(cropped_boxes.reshape(-1, 2, 2), max_size) cropped_boxes = cropped_boxes.clamp(min=0) area = (cropped_boxes[:, 1, :] - cropped_boxes[:, 0, :]).prod(dim=1) target["boxes"] = cropped_boxes.reshape(-1, 4) target["area"] = area fields.append("boxes") if "masks" in target: # FIXME should we update the area here if there are no boxes? target['masks'] = target['masks'][:, i:i + h, j:j + w] fields.append("masks") # remove elements for which the boxes or masks that have zero area if "boxes" in target or "masks" in target: # favor boxes selection when defining which elements to keep # this is compatible with previous implementation if "boxes" in target: cropped_boxes = target['boxes'].reshape(-1, 2, 2) keep = torch.all(cropped_boxes[:, 1, :] > cropped_boxes[:, 0, :], dim=1) else: keep = target['masks'].flatten(1).any(1) for field in fields: target[field] = target[field][keep] return cropped_image, target def hflip(image, target): flipped_image = F.hflip(image) w, h = image.size target = target.copy() if "boxes" in target: boxes = target["boxes"] boxes = boxes[:, [2, 1, 0, 3]] * torch.as_tensor([-1, 1, -1, 1]) + torch.as_tensor([w, 0, w, 0]) target["boxes"] = boxes if "masks" in target: target['masks'] = target['masks'].flip(-1) return flipped_image, target def resize(image, target, size, max_size=None): # size can be min_size (scalar) or (w, h) tuple def get_size_with_aspect_ratio(image_size, size, max_size=None): w, h = image_size if max_size is not None: min_original_size = float(min((w, h))) max_original_size = float(max((w, h))) if max_original_size / min_original_size * size > max_size: size = int(round(max_size * min_original_size / max_original_size)) if (w <= h and w == size) or (h <= w and h == size): return (h, w) if w < h: ow = size oh = int(size * h / w) else: oh = size ow = int(size * w / h) return (oh, ow) def get_size(image_size, size, max_size=None): if isinstance(size, (list, tuple)): return size[::-1] else: return get_size_with_aspect_ratio(image_size, size, max_size) size = get_size(image.size, size, max_size) rescaled_image = F.resize(image, size) if target is None: return rescaled_image, None ratios = tuple(float(s) / float(s_orig) for s, s_orig in zip(rescaled_image.size, image.size)) ratio_width, ratio_height = ratios target = target.copy() if "boxes" in target: boxes = target["boxes"] scaled_boxes = boxes * torch.as_tensor([ratio_width, ratio_height, ratio_width, ratio_height]) target["boxes"] = scaled_boxes if "area" in target: area = target["area"] scaled_area = area * (ratio_width * ratio_height) target["area"] = scaled_area h, w = size target["size"] = torch.tensor([h, w]) if "masks" in target: target['masks'] = interpolate( target['masks'][:, None].float(), size, mode="nearest")[:, 0] > 0.5 return rescaled_image, target def pad(image, target, padding): # assumes that we only pad on the bottom right corners padded_image = F.pad(image, (0, 0, padding[0], padding[1])) if target is None: return padded_image, None target = target.copy() # should we do something wrt the original size? target["size"] = torch.tensor(padded_image.size[::-1]) if "masks" in target: target['masks'] = torch.nn.functional.pad(target['masks'], (0, padding[0], 0, padding[1])) return padded_image, target class RandomCrop(object): def __init__(self, size): self.size = size def __call__(self, img, target): region = T.RandomCrop.get_params(img, self.size) return crop(img, target, region) class RandomSizeCrop(object): def __init__(self, min_size: int, max_size: int): self.min_size = min_size self.max_size = max_size def __call__(self, img: PIL.Image.Image, target: dict): w = random.randint(self.min_size, min(img.width, self.max_size)) h = random.randint(self.min_size, min(img.height, self.max_size)) region = T.RandomCrop.get_params(img, [h, w]) return crop(img, target, region) class CenterCrop(object): def __init__(self, size): self.size = size def __call__(self, img, target): image_width, image_height = img.size crop_height, crop_width = self.size crop_top = int(round((image_height - crop_height) / 2.)) crop_left = int(round((image_width - crop_width) / 2.)) return crop(img, target, (crop_top, crop_left, crop_height, crop_width)) class RandomHorizontalFlip(object): def __init__(self, p=0.5): self.p = p def __call__(self, img, target): if random.random() < self.p: return hflip(img, target) return img, target class RandomResize(object): def __init__(self, sizes, max_size=None): assert isinstance(sizes, (list, tuple)) self.sizes = sizes self.max_size = max_size def __call__(self, img, target=None): size = random.choice(self.sizes) return resize(img, target, size, self.max_size) class RandomPad(object): def __init__(self, max_pad): self.max_pad = max_pad def __call__(self, img, target): pad_x = random.randint(0, self.max_pad) pad_y = random.randint(0, self.max_pad) return pad(img, target, (pad_x, pad_y)) class RandomSelect(object): """ Randomly selects between transforms1 and transforms2, with probability p for transforms1 and (1 - p) for transforms2 """ def __init__(self, transforms1, transforms2, p=0.5): self.transforms1 = transforms1 self.transforms2 = transforms2 self.p = p def __call__(self, img, target): if random.random() < self.p: return self.transforms1(img, target) return self.transforms2(img, target) class ToTensor(object): def __call__(self, img, target): return F.to_tensor(img), target class RandomErasing(object): def __init__(self, *args, **kwargs): self.eraser = T.RandomErasing(*args, **kwargs) def __call__(self, img, target): return self.eraser(img), target class Normalize(object): def __init__(self, mean, std): self.mean = mean self.std = std def __call__(self, image, target=None): image = F.normalize(image, mean=self.mean, std=self.std) if target is None: return image, None target = target.copy() h, w = image.shape[-2:] if "boxes" in target: boxes = target["boxes"] boxes = box_xyxy_to_cxcywh(boxes) boxes = boxes / torch.tensor([w, h, w, h], dtype=torch.float32) target["boxes"] = boxes return image, target class Compose(object): def __init__(self, transforms): self.transforms = transforms def __call__(self, image, target): for t in self.transforms: image, target = t(image, target) return image, target def __repr__(self): format_string = self.__class__.__name__ + "(" for t in self.transforms: format_string += "\n" format_string += " {0}".format(t) format_string += "\n)" return format_string ================================================ FILE: engine.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Train and eval functions used in main.py """ import math import os import sys from typing import Iterable import torch import util.misc as utils from datasets.coco_eval import CocoEvaluator from datasets.panoptic_eval import PanopticEvaluator def train_one_epoch(model: torch.nn.Module, criterion: torch.nn.Module, data_loader: Iterable, optimizer: torch.optim.Optimizer, device: torch.device, epoch: int, max_norm: float = 0): model.train() criterion.train() metric_logger = utils.MetricLogger(delimiter=" ") metric_logger.add_meter('lr', utils.SmoothedValue(window_size=1, fmt='{value:.6f}')) metric_logger.add_meter('class_error', utils.SmoothedValue(window_size=1, fmt='{value:.2f}')) header = 'Epoch: [{}]'.format(epoch) print_freq = 10 for samples, targets in metric_logger.log_every(data_loader, print_freq, header): samples = samples.to(device) targets = [{k: v.to(device) for k, v in t.items()} for t in targets] outputs = model(samples) loss_dict = criterion(outputs, targets) weight_dict = criterion.weight_dict losses = sum(loss_dict[k] * weight_dict[k] for k in loss_dict.keys() if k in weight_dict) # reduce losses over all GPUs for logging purposes loss_dict_reduced = utils.reduce_dict(loss_dict) loss_dict_reduced_unscaled = {f'{k}_unscaled': v for k, v in loss_dict_reduced.items()} loss_dict_reduced_scaled = {k: v * weight_dict[k] for k, v in loss_dict_reduced.items() if k in weight_dict} losses_reduced_scaled = sum(loss_dict_reduced_scaled.values()) loss_value = losses_reduced_scaled.item() if not math.isfinite(loss_value): print("Loss is {}, stopping training".format(loss_value)) print(loss_dict_reduced) sys.exit(1) optimizer.zero_grad() losses.backward() if max_norm > 0: torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm) optimizer.step() metric_logger.update(loss=loss_value, **loss_dict_reduced_scaled, **loss_dict_reduced_unscaled) metric_logger.update(class_error=loss_dict_reduced['class_error']) metric_logger.update(lr=optimizer.param_groups[0]["lr"]) # gather the stats from all processes metric_logger.synchronize_between_processes() print("Averaged stats:", metric_logger) return {k: meter.global_avg for k, meter in metric_logger.meters.items()} @torch.no_grad() def evaluate(model, criterion, postprocessors, data_loader, base_ds, device, output_dir): model.eval() criterion.eval() metric_logger = utils.MetricLogger(delimiter=" ") metric_logger.add_meter('class_error', utils.SmoothedValue(window_size=1, fmt='{value:.2f}')) header = 'Test:' iou_types = tuple(k for k in ('segm', 'bbox') if k in postprocessors.keys()) coco_evaluator = CocoEvaluator(base_ds, iou_types) # coco_evaluator.coco_eval[iou_types[0]].params.iouThrs = [0, 0.1, 0.5, 0.75] panoptic_evaluator = None if 'panoptic' in postprocessors.keys(): panoptic_evaluator = PanopticEvaluator( data_loader.dataset.ann_file, data_loader.dataset.ann_folder, output_dir=os.path.join(output_dir, "panoptic_eval"), ) for samples, targets in metric_logger.log_every(data_loader, 10, header): samples = samples.to(device) targets = [{k: v.to(device) for k, v in t.items()} for t in targets] outputs = model(samples) loss_dict = criterion(outputs, targets) weight_dict = criterion.weight_dict # reduce losses over all GPUs for logging purposes loss_dict_reduced = utils.reduce_dict(loss_dict) loss_dict_reduced_scaled = {k: v * weight_dict[k] for k, v in loss_dict_reduced.items() if k in weight_dict} loss_dict_reduced_unscaled = {f'{k}_unscaled': v for k, v in loss_dict_reduced.items()} metric_logger.update(loss=sum(loss_dict_reduced_scaled.values()), **loss_dict_reduced_scaled, **loss_dict_reduced_unscaled) metric_logger.update(class_error=loss_dict_reduced['class_error']) orig_target_sizes = torch.stack([t["orig_size"] for t in targets], dim=0) results = postprocessors['bbox'](outputs, orig_target_sizes) if 'segm' in postprocessors.keys(): target_sizes = torch.stack([t["size"] for t in targets], dim=0) results = postprocessors['segm'](results, outputs, orig_target_sizes, target_sizes) res = {target['image_id'].item(): output for target, output in zip(targets, results)} if coco_evaluator is not None: coco_evaluator.update(res) if panoptic_evaluator is not None: res_pano = postprocessors["panoptic"](outputs, target_sizes, orig_target_sizes) for i, target in enumerate(targets): image_id = target["image_id"].item() file_name = f"{image_id:012d}.png" res_pano[i]["image_id"] = image_id res_pano[i]["file_name"] = file_name panoptic_evaluator.update(res_pano) # gather the stats from all processes metric_logger.synchronize_between_processes() print("Averaged stats:", metric_logger) if coco_evaluator is not None: coco_evaluator.synchronize_between_processes() if panoptic_evaluator is not None: panoptic_evaluator.synchronize_between_processes() # accumulate predictions from all images if coco_evaluator is not None: coco_evaluator.accumulate() coco_evaluator.summarize() panoptic_res = None if panoptic_evaluator is not None: panoptic_res = panoptic_evaluator.summarize() stats = {k: meter.global_avg for k, meter in metric_logger.meters.items()} if coco_evaluator is not None: if 'bbox' in postprocessors.keys(): stats['coco_eval_bbox'] = coco_evaluator.coco_eval['bbox'].stats.tolist() if 'segm' in postprocessors.keys(): stats['coco_eval_masks'] = coco_evaluator.coco_eval['segm'].stats.tolist() if panoptic_res is not None: stats['PQ_all'] = panoptic_res["All"] stats['PQ_th'] = panoptic_res["Things"] stats['PQ_st'] = panoptic_res["Stuff"] return stats, coco_evaluator ================================================ FILE: hubconf.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import torch from models.backbone import Backbone, Joiner from models.detr import DETR, PostProcess from models.position_encoding import PositionEmbeddingSine from models.segmentation import DETRsegm, PostProcessPanoptic from models.transformer import Transformer dependencies = ["torch", "torchvision"] def _make_detr(backbone_name: str, dilation=False, num_classes=91, mask=False): hidden_dim = 256 backbone = Backbone(backbone_name, train_backbone=True, return_interm_layers=mask, dilation=dilation) pos_enc = PositionEmbeddingSine(hidden_dim // 2, normalize=True) backbone_with_pos_enc = Joiner(backbone, pos_enc) backbone_with_pos_enc.num_channels = backbone.num_channels transformer = Transformer(d_model=hidden_dim, return_intermediate_dec=True) detr = DETR(backbone_with_pos_enc, transformer, num_classes=num_classes, num_queries=100) if mask: return DETRsegm(detr) return detr def detr_resnet50(pretrained=False, num_classes=91, return_postprocessor=False): """ DETR R50 with 6 encoder and 6 decoder layers. Achieves 42/62.4 AP/AP50 on COCO val5k. """ model = _make_detr("resnet50", dilation=False, num_classes=num_classes) if pretrained: checkpoint = torch.hub.load_state_dict_from_url( url="https://dl.fbaipublicfiles.com/detr/detr-r50-e632da11.pth", map_location="cpu", check_hash=True ) model.load_state_dict(checkpoint["model"]) if return_postprocessor: return model, PostProcess() return model def detr_resnet50_dc5(pretrained=False, num_classes=91, return_postprocessor=False): """ DETR-DC5 R50 with 6 encoder and 6 decoder layers. The last block of ResNet-50 has dilation to increase output resolution. Achieves 43.3/63.1 AP/AP50 on COCO val5k. """ model = _make_detr("resnet50", dilation=True, num_classes=num_classes) if pretrained: checkpoint = torch.hub.load_state_dict_from_url( url="https://dl.fbaipublicfiles.com/detr/detr-r50-dc5-f0fb7ef5.pth", map_location="cpu", check_hash=True ) model.load_state_dict(checkpoint["model"]) if return_postprocessor: return model, PostProcess() return model def detr_resnet101(pretrained=False, num_classes=91, return_postprocessor=False): """ DETR-DC5 R101 with 6 encoder and 6 decoder layers. Achieves 43.5/63.8 AP/AP50 on COCO val5k. """ model = _make_detr("resnet101", dilation=False, num_classes=num_classes) if pretrained: checkpoint = torch.hub.load_state_dict_from_url( url="https://dl.fbaipublicfiles.com/detr/detr-r101-2c7b67e5.pth", map_location="cpu", check_hash=True ) model.load_state_dict(checkpoint["model"]) if return_postprocessor: return model, PostProcess() return model def detr_resnet101_dc5(pretrained=False, num_classes=91, return_postprocessor=False): """ DETR-DC5 R101 with 6 encoder and 6 decoder layers. The last block of ResNet-101 has dilation to increase output resolution. Achieves 44.9/64.7 AP/AP50 on COCO val5k. """ model = _make_detr("resnet101", dilation=True, num_classes=num_classes) if pretrained: checkpoint = torch.hub.load_state_dict_from_url( url="https://dl.fbaipublicfiles.com/detr/detr-r101-dc5-a2e86def.pth", map_location="cpu", check_hash=True ) model.load_state_dict(checkpoint["model"]) if return_postprocessor: return model, PostProcess() return model def detr_resnet50_panoptic( pretrained=False, num_classes=250, threshold=0.85, return_postprocessor=False ): """ DETR R50 with 6 encoder and 6 decoder layers. Achieves 43.4 PQ on COCO val5k. threshold is the minimum confidence required for keeping segments in the prediction """ model = _make_detr("resnet50", dilation=False, num_classes=num_classes, mask=True) is_thing_map = {i: i <= 90 for i in range(250)} if pretrained: checkpoint = torch.hub.load_state_dict_from_url( url="https://dl.fbaipublicfiles.com/detr/detr-r50-panoptic-00ce5173.pth", map_location="cpu", check_hash=True, ) model.load_state_dict(checkpoint["model"]) if return_postprocessor: return model, PostProcessPanoptic(is_thing_map, threshold=threshold) return model def detr_resnet50_dc5_panoptic( pretrained=False, num_classes=250, threshold=0.85, return_postprocessor=False ): """ DETR-DC5 R50 with 6 encoder and 6 decoder layers. The last block of ResNet-50 has dilation to increase output resolution. Achieves 44.6 on COCO val5k. threshold is the minimum confidence required for keeping segments in the prediction """ model = _make_detr("resnet50", dilation=True, num_classes=num_classes, mask=True) is_thing_map = {i: i <= 90 for i in range(250)} if pretrained: checkpoint = torch.hub.load_state_dict_from_url( url="https://dl.fbaipublicfiles.com/detr/detr-r50-dc5-panoptic-da08f1b1.pth", map_location="cpu", check_hash=True, ) model.load_state_dict(checkpoint["model"]) if return_postprocessor: return model, PostProcessPanoptic(is_thing_map, threshold=threshold) return model def detr_resnet101_panoptic( pretrained=False, num_classes=250, threshold=0.85, return_postprocessor=False ): """ DETR-DC5 R101 with 6 encoder and 6 decoder layers. Achieves 45.1 PQ on COCO val5k. threshold is the minimum confidence required for keeping segments in the prediction """ model = _make_detr("resnet101", dilation=False, num_classes=num_classes, mask=True) is_thing_map = {i: i <= 90 for i in range(250)} if pretrained: checkpoint = torch.hub.load_state_dict_from_url( url="https://dl.fbaipublicfiles.com/detr/detr-r101-panoptic-40021d53.pth", map_location="cpu", check_hash=True, ) model.load_state_dict(checkpoint["model"]) if return_postprocessor: return model, PostProcessPanoptic(is_thing_map, threshold=threshold) return model ================================================ FILE: main.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import argparse import datetime import json import random import time from pathlib import Path import numpy as np import torch from torch.utils.data import DataLoader, DistributedSampler import datasets import util.misc as utils from datasets import build_dataset, get_coco_api_from_dataset from engine import evaluate, train_one_epoch from models import build_model def get_args_parser(): parser = argparse.ArgumentParser('Set transformer detector', add_help=False) parser.add_argument('--lr', default=1e-4, type=float) parser.add_argument('--lr_backbone', default=1e-5, type=float) parser.add_argument('--batch_size', default=2, type=int) parser.add_argument('--weight_decay', default=1e-4, type=float) parser.add_argument('--epochs', default=300, type=int) parser.add_argument('--lr_drop', default=200, type=int) parser.add_argument('--clip_max_norm', default=0.1, type=float, help='gradient clipping max norm') # Model parameters parser.add_argument('--frozen_weights', type=str, default=None, help="Path to the pretrained model. If set, only the mask head will be trained") # * Backbone parser.add_argument('--backbone', default='resnet50', type=str, help="Name of the convolutional backbone to use") parser.add_argument('--dilation', action='store_true', help="If true, we replace stride with dilation in the last convolutional block (DC5)") parser.add_argument('--position_embedding', default='sine', type=str, choices=('sine', 'learned'), help="Type of positional embedding to use on top of the image features") # * Transformer parser.add_argument('--enc_layers', default=6, type=int, help="Number of encoding layers in the transformer") parser.add_argument('--dec_layers', default=6, type=int, help="Number of decoding layers in the transformer") parser.add_argument('--dim_feedforward', default=2048, type=int, help="Intermediate size of the feedforward layers in the transformer blocks") parser.add_argument('--hidden_dim', default=256, type=int, help="Size of the embeddings (dimension of the transformer)") parser.add_argument('--dropout', default=0.1, type=float, help="Dropout applied in the transformer") parser.add_argument('--nheads', default=8, type=int, help="Number of attention heads inside the transformer's attentions") parser.add_argument('--num_queries', default=100, type=int, help="Number of query slots") parser.add_argument('--pre_norm', action='store_true') # * Segmentation parser.add_argument('--masks', action='store_true', help="Train segmentation head if the flag is provided") # Loss parser.add_argument('--no_aux_loss', dest='aux_loss', action='store_false', help="Disables auxiliary decoding losses (loss at each layer)") # * Matcher parser.add_argument('--set_cost_class', default=1, type=float, help="Class coefficient in the matching cost") parser.add_argument('--set_cost_bbox', default=5, type=float, help="L1 box coefficient in the matching cost") parser.add_argument('--set_cost_giou', default=2, type=float, help="giou box coefficient in the matching cost") # * Loss coefficients parser.add_argument('--mask_loss_coef', default=1, type=float) parser.add_argument('--dice_loss_coef', default=1, type=float) parser.add_argument('--bbox_loss_coef', default=5, type=float) parser.add_argument('--giou_loss_coef', default=2, type=float) parser.add_argument('--eos_coef', default=0.1, type=float, help="Relative classification weight of the no-object class") # dataset parameters parser.add_argument('--dataset_file', default='coco') parser.add_argument('--coco_path', type=str) parser.add_argument('--coco_panoptic_path', type=str) parser.add_argument('--remove_difficult', action='store_true') parser.add_argument('--output_dir', default='', help='path where to save, empty for no saving') parser.add_argument('--device', default='cuda', help='device to use for training / testing') parser.add_argument('--seed', default=42, type=int) parser.add_argument('--resume', default='', help='resume from checkpoint') parser.add_argument('--start_epoch', default=0, type=int, metavar='N', help='start epoch') parser.add_argument('--eval', action='store_true') parser.add_argument('--num_workers', default=2, type=int) # distributed training parameters parser.add_argument('--world_size', default=1, type=int, help='number of distributed processes') parser.add_argument('--dist_url', default='env://', help='url used to set up distributed training') return parser def main(args): utils.init_distributed_mode(args) print("git:\n {}\n".format(utils.get_sha())) if args.frozen_weights is not None: assert args.masks, "Frozen training is meant for segmentation only" print(args) device = torch.device(args.device) # fix the seed for reproducibility seed = args.seed + utils.get_rank() torch.manual_seed(seed) np.random.seed(seed) random.seed(seed) model, criterion, postprocessors = build_model(args) model.to(device) model_without_ddp = model if args.distributed: model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu]) model_without_ddp = model.module n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad) print('number of params:', n_parameters) param_dicts = [ {"params": [p for n, p in model_without_ddp.named_parameters() if "backbone" not in n and p.requires_grad]}, { "params": [p for n, p in model_without_ddp.named_parameters() if "backbone" in n and p.requires_grad], "lr": args.lr_backbone, }, ] optimizer = torch.optim.AdamW(param_dicts, lr=args.lr, weight_decay=args.weight_decay) lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, args.lr_drop) dataset_train = build_dataset(image_set='train', args=args) dataset_val = build_dataset(image_set='val', args=args) if args.distributed: sampler_train = DistributedSampler(dataset_train) sampler_val = DistributedSampler(dataset_val, shuffle=False) else: sampler_train = torch.utils.data.RandomSampler(dataset_train) sampler_val = torch.utils.data.SequentialSampler(dataset_val) batch_sampler_train = torch.utils.data.BatchSampler( sampler_train, args.batch_size, drop_last=True) data_loader_train = DataLoader(dataset_train, batch_sampler=batch_sampler_train, collate_fn=utils.collate_fn, num_workers=args.num_workers) data_loader_val = DataLoader(dataset_val, args.batch_size, sampler=sampler_val, drop_last=False, collate_fn=utils.collate_fn, num_workers=args.num_workers) if args.dataset_file == "coco_panoptic": # We also evaluate AP during panoptic training, on original coco DS coco_val = datasets.coco.build("val", args) base_ds = get_coco_api_from_dataset(coco_val) else: base_ds = get_coco_api_from_dataset(dataset_val) if args.frozen_weights is not None: checkpoint = torch.load(args.frozen_weights, map_location='cpu') model_without_ddp.detr.load_state_dict(checkpoint['model']) output_dir = Path(args.output_dir) if args.resume: if args.resume.startswith('https'): checkpoint = torch.hub.load_state_dict_from_url( args.resume, map_location='cpu', check_hash=True) else: checkpoint = torch.load(args.resume, map_location='cpu') model_without_ddp.load_state_dict(checkpoint['model']) if not args.eval and 'optimizer' in checkpoint and 'lr_scheduler' in checkpoint and 'epoch' in checkpoint: optimizer.load_state_dict(checkpoint['optimizer']) lr_scheduler.load_state_dict(checkpoint['lr_scheduler']) args.start_epoch = checkpoint['epoch'] + 1 if args.eval: test_stats, coco_evaluator = evaluate(model, criterion, postprocessors, data_loader_val, base_ds, device, args.output_dir) if args.output_dir: utils.save_on_master(coco_evaluator.coco_eval["bbox"].eval, output_dir / "eval.pth") return print("Start training") start_time = time.time() for epoch in range(args.start_epoch, args.epochs): if args.distributed: sampler_train.set_epoch(epoch) train_stats = train_one_epoch( model, criterion, data_loader_train, optimizer, device, epoch, args.clip_max_norm) lr_scheduler.step() if args.output_dir: checkpoint_paths = [output_dir / 'checkpoint.pth'] # extra checkpoint before LR drop and every 100 epochs if (epoch + 1) % args.lr_drop == 0 or (epoch + 1) % 100 == 0: checkpoint_paths.append(output_dir / f'checkpoint{epoch:04}.pth') for checkpoint_path in checkpoint_paths: utils.save_on_master({ 'model': model_without_ddp.state_dict(), 'optimizer': optimizer.state_dict(), 'lr_scheduler': lr_scheduler.state_dict(), 'epoch': epoch, 'args': args, }, checkpoint_path) test_stats, coco_evaluator = evaluate( model, criterion, postprocessors, data_loader_val, base_ds, device, args.output_dir ) log_stats = {**{f'train_{k}': v for k, v in train_stats.items()}, **{f'test_{k}': v for k, v in test_stats.items()}, 'epoch': epoch, 'n_parameters': n_parameters} if args.output_dir and utils.is_main_process(): with (output_dir / "log.txt").open("a") as f: f.write(json.dumps(log_stats) + "\n") # for evaluation logs if coco_evaluator is not None: (output_dir / 'eval').mkdir(exist_ok=True) if "bbox" in coco_evaluator.coco_eval: filenames = ['latest.pth'] if epoch % 50 == 0: filenames.append(f'{epoch:03}.pth') for name in filenames: torch.save(coco_evaluator.coco_eval["bbox"].eval, output_dir / "eval" / name) total_time = time.time() - start_time total_time_str = str(datetime.timedelta(seconds=int(total_time))) print('Training time {}'.format(total_time_str)) if __name__ == '__main__': parser = argparse.ArgumentParser('DETR training and evaluation script', parents=[get_args_parser()]) args = parser.parse_args() if args.output_dir: Path(args.output_dir).mkdir(parents=True, exist_ok=True) main(args) ================================================ FILE: models/__init__.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved from .detr import build def build_model(args): return build(args) ================================================ FILE: models/backbone.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Backbone modules. """ from collections import OrderedDict import torch import torch.nn.functional as F import torchvision from torch import nn from torchvision.models._utils import IntermediateLayerGetter from typing import Dict, List from util.misc import NestedTensor, is_main_process from .position_encoding import build_position_encoding class FrozenBatchNorm2d(torch.nn.Module): """ BatchNorm2d where the batch statistics and the affine parameters are fixed. Copy-paste from torchvision.misc.ops with added eps before rqsrt, without which any other models than torchvision.models.resnet[18,34,50,101] produce nans. """ def __init__(self, n): super(FrozenBatchNorm2d, self).__init__() self.register_buffer("weight", torch.ones(n)) self.register_buffer("bias", torch.zeros(n)) self.register_buffer("running_mean", torch.zeros(n)) self.register_buffer("running_var", torch.ones(n)) def _load_from_state_dict(self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs): num_batches_tracked_key = prefix + 'num_batches_tracked' if num_batches_tracked_key in state_dict: del state_dict[num_batches_tracked_key] super(FrozenBatchNorm2d, self)._load_from_state_dict( state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) def forward(self, x): # move reshapes to the beginning # to make it fuser-friendly w = self.weight.reshape(1, -1, 1, 1) b = self.bias.reshape(1, -1, 1, 1) rv = self.running_var.reshape(1, -1, 1, 1) rm = self.running_mean.reshape(1, -1, 1, 1) eps = 1e-5 scale = w * (rv + eps).rsqrt() bias = b - rm * scale return x * scale + bias class BackboneBase(nn.Module): def __init__(self, backbone: nn.Module, train_backbone: bool, num_channels: int, return_interm_layers: bool): super().__init__() for name, parameter in backbone.named_parameters(): if not train_backbone or 'layer2' not in name and 'layer3' not in name and 'layer4' not in name: parameter.requires_grad_(False) if return_interm_layers: return_layers = {"layer1": "0", "layer2": "1", "layer3": "2", "layer4": "3"} else: return_layers = {'layer4': "0"} self.body = IntermediateLayerGetter(backbone, return_layers=return_layers) self.num_channels = num_channels def forward(self, tensor_list: NestedTensor): xs = self.body(tensor_list.tensors) out: Dict[str, NestedTensor] = {} for name, x in xs.items(): m = tensor_list.mask assert m is not None mask = F.interpolate(m[None].float(), size=x.shape[-2:]).to(torch.bool)[0] out[name] = NestedTensor(x, mask) return out class Backbone(BackboneBase): """ResNet backbone with frozen BatchNorm.""" def __init__(self, name: str, train_backbone: bool, return_interm_layers: bool, dilation: bool): backbone = getattr(torchvision.models, name)( replace_stride_with_dilation=[False, False, dilation], pretrained=is_main_process(), norm_layer=FrozenBatchNorm2d) num_channels = 512 if name in ('resnet18', 'resnet34') else 2048 super().__init__(backbone, train_backbone, num_channels, return_interm_layers) class Joiner(nn.Sequential): def __init__(self, backbone, position_embedding): super().__init__(backbone, position_embedding) def forward(self, tensor_list: NestedTensor): xs = self[0](tensor_list) out: List[NestedTensor] = [] pos = [] for name, x in xs.items(): out.append(x) # position encoding pos.append(self[1](x).to(x.tensors.dtype)) return out, pos def build_backbone(args): position_embedding = build_position_encoding(args) train_backbone = args.lr_backbone > 0 return_interm_layers = args.masks backbone = Backbone(args.backbone, train_backbone, return_interm_layers, args.dilation) model = Joiner(backbone, position_embedding) model.num_channels = backbone.num_channels return model ================================================ FILE: models/detr.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ DETR model and criterion classes. """ import torch import torch.nn.functional as F from torch import nn from util import box_ops from util.misc import (NestedTensor, nested_tensor_from_tensor_list, accuracy, get_world_size, interpolate, is_dist_avail_and_initialized) from .backbone import build_backbone from .matcher import build_matcher from .segmentation import (DETRsegm, PostProcessPanoptic, PostProcessSegm, dice_loss, sigmoid_focal_loss) from .transformer import build_transformer class DETR(nn.Module): """ This is the DETR module that performs object detection """ def __init__(self, backbone, transformer, num_classes, num_queries, aux_loss=False): """ Initializes the model. Parameters: backbone: torch module of the backbone to be used. See backbone.py transformer: torch module of the transformer architecture. See transformer.py num_classes: number of object classes num_queries: number of object queries, ie detection slot. This is the maximal number of objects DETR can detect in a single image. For COCO, we recommend 100 queries. aux_loss: True if auxiliary decoding losses (loss at each decoder layer) are to be used. """ super().__init__() self.num_queries = num_queries self.transformer = transformer hidden_dim = transformer.d_model self.class_embed = nn.Linear(hidden_dim, num_classes + 1) self.bbox_embed = MLP(hidden_dim, hidden_dim, 4, 3) self.query_embed = nn.Embedding(num_queries, hidden_dim) self.input_proj = nn.Conv2d(backbone.num_channels, hidden_dim, kernel_size=1) self.backbone = backbone self.aux_loss = aux_loss def forward(self, samples: NestedTensor): """ The forward expects a NestedTensor, which consists of: - samples.tensor: batched images, of shape [batch_size x 3 x H x W] - samples.mask: a binary mask of shape [batch_size x H x W], containing 1 on padded pixels It returns a dict with the following elements: - "pred_logits": the classification logits (including no-object) for all queries. Shape= [batch_size x num_queries x (num_classes + 1)] - "pred_boxes": The normalized boxes coordinates for all queries, represented as (center_x, center_y, height, width). These values are normalized in [0, 1], relative to the size of each individual image (disregarding possible padding). See PostProcess for information on how to retrieve the unnormalized bounding box. - "aux_outputs": Optional, only returned when auxilary losses are activated. It is a list of dictionnaries containing the two above keys for each decoder layer. """ if isinstance(samples, (list, torch.Tensor)): samples = nested_tensor_from_tensor_list(samples) features, pos = self.backbone(samples) src, mask = features[-1].decompose() assert mask is not None hs = self.transformer(self.input_proj(src), mask, self.query_embed.weight, pos[-1])[0] outputs_class = self.class_embed(hs) outputs_coord = self.bbox_embed(hs).sigmoid() out = {'pred_logits': outputs_class[-1], 'pred_boxes': outputs_coord[-1]} if self.aux_loss: out['aux_outputs'] = self._set_aux_loss(outputs_class, outputs_coord) return out @torch.jit.unused def _set_aux_loss(self, outputs_class, outputs_coord): # this is a workaround to make torchscript happy, as torchscript # doesn't support dictionary with non-homogeneous values, such # as a dict having both a Tensor and a list. return [{'pred_logits': a, 'pred_boxes': b} for a, b in zip(outputs_class[:-1], outputs_coord[:-1])] class SetCriterion(nn.Module): """ This class computes the loss for DETR. The process happens in two steps: 1) we compute hungarian assignment between ground truth boxes and the outputs of the model 2) we supervise each pair of matched ground-truth / prediction (supervise class and box) """ def __init__(self, num_classes, matcher, weight_dict, eos_coef, losses): """ Create the criterion. Parameters: num_classes: number of object categories, omitting the special no-object category matcher: module able to compute a matching between targets and proposals weight_dict: dict containing as key the names of the losses and as values their relative weight. eos_coef: relative classification weight applied to the no-object category losses: list of all the losses to be applied. See get_loss for list of available losses. """ super().__init__() self.num_classes = num_classes self.matcher = matcher self.weight_dict = weight_dict self.eos_coef = eos_coef self.losses = losses empty_weight = torch.ones(self.num_classes + 1) empty_weight[-1] = self.eos_coef self.register_buffer('empty_weight', empty_weight) def loss_labels(self, outputs, targets, indices, num_boxes, log=True): """Classification loss (NLL) targets dicts must contain the key "labels" containing a tensor of dim [nb_target_boxes] """ assert 'pred_logits' in outputs src_logits = outputs['pred_logits'] idx = self._get_src_permutation_idx(indices) target_classes_o = torch.cat([t["labels"][J] for t, (_, J) in zip(targets, indices)]) target_classes = torch.full(src_logits.shape[:2], self.num_classes, dtype=torch.int64, device=src_logits.device) target_classes[idx] = target_classes_o loss_ce = F.cross_entropy(src_logits.transpose(1, 2), target_classes, self.empty_weight) losses = {'loss_ce': loss_ce} if log: # TODO this should probably be a separate loss, not hacked in this one here losses['class_error'] = 100 - accuracy(src_logits[idx], target_classes_o)[0] return losses @torch.no_grad() def loss_cardinality(self, outputs, targets, indices, num_boxes): """ Compute the cardinality error, ie the absolute error in the number of predicted non-empty boxes This is not really a loss, it is intended for logging purposes only. It doesn't propagate gradients """ pred_logits = outputs['pred_logits'] device = pred_logits.device tgt_lengths = torch.as_tensor([len(v["labels"]) for v in targets], device=device) # Count the number of predictions that are NOT "no-object" (which is the last class) card_pred = (pred_logits.argmax(-1) != pred_logits.shape[-1] - 1).sum(1) card_err = F.l1_loss(card_pred.float(), tgt_lengths.float()) losses = {'cardinality_error': card_err} return losses def loss_boxes(self, outputs, targets, indices, num_boxes): """Compute the losses related to the bounding boxes, the L1 regression loss and the GIoU loss targets dicts must contain the key "boxes" containing a tensor of dim [nb_target_boxes, 4] The target boxes are expected in format (center_x, center_y, w, h), normalized by the image size. """ assert 'pred_boxes' in outputs idx = self._get_src_permutation_idx(indices) src_boxes = outputs['pred_boxes'][idx] target_boxes = torch.cat([t['boxes'][i] for t, (_, i) in zip(targets, indices)], dim=0) loss_bbox = F.l1_loss(src_boxes, target_boxes, reduction='none') losses = {} losses['loss_bbox'] = loss_bbox.sum() / num_boxes loss_giou = 1 - torch.diag(box_ops.generalized_box_iou( box_ops.box_cxcywh_to_xyxy(src_boxes), box_ops.box_cxcywh_to_xyxy(target_boxes))) losses['loss_giou'] = loss_giou.sum() / num_boxes return losses def loss_masks(self, outputs, targets, indices, num_boxes): """Compute the losses related to the masks: the focal loss and the dice loss. targets dicts must contain the key "masks" containing a tensor of dim [nb_target_boxes, h, w] """ assert "pred_masks" in outputs src_idx = self._get_src_permutation_idx(indices) tgt_idx = self._get_tgt_permutation_idx(indices) src_masks = outputs["pred_masks"] src_masks = src_masks[src_idx] masks = [t["masks"] for t in targets] # TODO use valid to mask invalid areas due to padding in loss target_masks, valid = nested_tensor_from_tensor_list(masks).decompose() target_masks = target_masks.to(src_masks) target_masks = target_masks[tgt_idx] # upsample predictions to the target size src_masks = interpolate(src_masks[:, None], size=target_masks.shape[-2:], mode="bilinear", align_corners=False) src_masks = src_masks[:, 0].flatten(1) target_masks = target_masks.flatten(1) target_masks = target_masks.view(src_masks.shape) losses = { "loss_mask": sigmoid_focal_loss(src_masks, target_masks, num_boxes), "loss_dice": dice_loss(src_masks, target_masks, num_boxes), } return losses def _get_src_permutation_idx(self, indices): # permute predictions following indices batch_idx = torch.cat([torch.full_like(src, i) for i, (src, _) in enumerate(indices)]) src_idx = torch.cat([src for (src, _) in indices]) return batch_idx, src_idx def _get_tgt_permutation_idx(self, indices): # permute targets following indices batch_idx = torch.cat([torch.full_like(tgt, i) for i, (_, tgt) in enumerate(indices)]) tgt_idx = torch.cat([tgt for (_, tgt) in indices]) return batch_idx, tgt_idx def get_loss(self, loss, outputs, targets, indices, num_boxes, **kwargs): loss_map = { 'labels': self.loss_labels, 'cardinality': self.loss_cardinality, 'boxes': self.loss_boxes, 'masks': self.loss_masks } assert loss in loss_map, f'do you really want to compute {loss} loss?' return loss_map[loss](outputs, targets, indices, num_boxes, **kwargs) def forward(self, outputs, targets): """ This performs the loss computation. Parameters: outputs: dict of tensors, see the output specification of the model for the format targets: list of dicts, such that len(targets) == batch_size. The expected keys in each dict depends on the losses applied, see each loss' doc """ outputs_without_aux = {k: v for k, v in outputs.items() if k != 'aux_outputs'} # Retrieve the matching between the outputs of the last layer and the targets indices = self.matcher(outputs_without_aux, targets) # Compute the average number of target boxes accross all nodes, for normalization purposes num_boxes = sum(len(t["labels"]) for t in targets) num_boxes = torch.as_tensor([num_boxes], dtype=torch.float, device=next(iter(outputs.values())).device) if is_dist_avail_and_initialized(): torch.distributed.all_reduce(num_boxes) num_boxes = torch.clamp(num_boxes / get_world_size(), min=1).item() # Compute all the requested losses losses = {} for loss in self.losses: losses.update(self.get_loss(loss, outputs, targets, indices, num_boxes)) # In case of auxiliary losses, we repeat this process with the output of each intermediate layer. if 'aux_outputs' in outputs: for i, aux_outputs in enumerate(outputs['aux_outputs']): indices = self.matcher(aux_outputs, targets) for loss in self.losses: if loss == 'masks': # Intermediate masks losses are too costly to compute, we ignore them. continue kwargs = {} if loss == 'labels': # Logging is enabled only for the last layer kwargs = {'log': False} l_dict = self.get_loss(loss, aux_outputs, targets, indices, num_boxes, **kwargs) l_dict = {k + f'_{i}': v for k, v in l_dict.items()} losses.update(l_dict) return losses class PostProcess(nn.Module): """ This module converts the model's output into the format expected by the coco api""" @torch.no_grad() def forward(self, outputs, target_sizes): """ Perform the computation Parameters: outputs: raw outputs of the model target_sizes: tensor of dimension [batch_size x 2] containing the size of each images of the batch For evaluation, this must be the original image size (before any data augmentation) For visualization, this should be the image size after data augment, but before padding """ out_logits, out_bbox = outputs['pred_logits'], outputs['pred_boxes'] assert len(out_logits) == len(target_sizes) assert target_sizes.shape[1] == 2 prob = F.softmax(out_logits, -1) scores, labels = prob[..., :-1].max(-1) # convert to [x0, y0, x1, y1] format boxes = box_ops.box_cxcywh_to_xyxy(out_bbox) # and from relative [0, 1] to absolute [0, height] coordinates img_h, img_w = target_sizes.unbind(1) scale_fct = torch.stack([img_w, img_h, img_w, img_h], dim=1) boxes = boxes * scale_fct[:, None, :] results = [{'scores': s, 'labels': l, 'boxes': b} for s, l, b in zip(scores, labels, boxes)] return results class MLP(nn.Module): """ Very simple multi-layer perceptron (also called FFN)""" def __init__(self, input_dim, hidden_dim, output_dim, num_layers): super().__init__() self.num_layers = num_layers h = [hidden_dim] * (num_layers - 1) self.layers = nn.ModuleList(nn.Linear(n, k) for n, k in zip([input_dim] + h, h + [output_dim])) def forward(self, x): for i, layer in enumerate(self.layers): x = F.relu(layer(x)) if i < self.num_layers - 1 else layer(x) return x def build(args): # the `num_classes` naming here is somewhat misleading. # it indeed corresponds to `max_obj_id + 1`, where max_obj_id # is the maximum id for a class in your dataset. For example, # COCO has a max_obj_id of 90, so we pass `num_classes` to be 91. # As another example, for a dataset that has a single class with id 1, # you should pass `num_classes` to be 2 (max_obj_id + 1). # For more details on this, check the following discussion # https://github.com/facebookresearch/detr/issues/108#issuecomment-650269223 num_classes = 20 if args.dataset_file != 'coco' else 91 if args.dataset_file == "coco_panoptic": # for panoptic, we just add a num_classes that is large enough to hold # max_obj_id + 1, but the exact value doesn't really matter num_classes = 250 device = torch.device(args.device) backbone = build_backbone(args) transformer = build_transformer(args) model = DETR( backbone, transformer, num_classes=num_classes, num_queries=args.num_queries, aux_loss=args.aux_loss, ) if args.masks: model = DETRsegm(model, freeze_detr=(args.frozen_weights is not None)) matcher = build_matcher(args) weight_dict = {'loss_ce': 1, 'loss_bbox': args.bbox_loss_coef} weight_dict['loss_giou'] = args.giou_loss_coef if args.masks: weight_dict["loss_mask"] = args.mask_loss_coef weight_dict["loss_dice"] = args.dice_loss_coef # TODO this is a hack if args.aux_loss: aux_weight_dict = {} for i in range(args.dec_layers - 1): aux_weight_dict.update({k + f'_{i}': v for k, v in weight_dict.items()}) weight_dict.update(aux_weight_dict) losses = ['labels', 'boxes', 'cardinality'] if args.masks: losses += ["masks"] criterion = SetCriterion(num_classes, matcher=matcher, weight_dict=weight_dict, eos_coef=args.eos_coef, losses=losses) criterion.to(device) postprocessors = {'bbox': PostProcess()} if args.masks: postprocessors['segm'] = PostProcessSegm() if args.dataset_file == "coco_panoptic": is_thing_map = {i: i <= 90 for i in range(201)} postprocessors["panoptic"] = PostProcessPanoptic(is_thing_map, threshold=0.85) return model, criterion, postprocessors ================================================ FILE: models/matcher.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Modules to compute the matching cost and solve the corresponding LSAP. """ import torch from scipy.optimize import linear_sum_assignment from torch import nn from util.box_ops import box_cxcywh_to_xyxy, generalized_box_iou class HungarianMatcher(nn.Module): """This class computes an assignment between the targets and the predictions of the network For efficiency reasons, the targets don't include the no_object. Because of this, in general, there are more predictions than targets. In this case, we do a 1-to-1 matching of the best predictions, while the others are un-matched (and thus treated as non-objects). """ def __init__(self, cost_class: float = 1, cost_bbox: float = 1, cost_giou: float = 1): """Creates the matcher Params: cost_class: This is the relative weight of the classification error in the matching cost cost_bbox: This is the relative weight of the L1 error of the bounding box coordinates in the matching cost cost_giou: This is the relative weight of the giou loss of the bounding box in the matching cost """ super().__init__() self.cost_class = cost_class self.cost_bbox = cost_bbox self.cost_giou = cost_giou assert cost_class != 0 or cost_bbox != 0 or cost_giou != 0, "all costs cant be 0" @torch.no_grad() def forward(self, outputs, targets): """ Performs the matching Params: outputs: This is a dict that contains at least these entries: "pred_logits": Tensor of dim [batch_size, num_queries, num_classes] with the classification logits "pred_boxes": Tensor of dim [batch_size, num_queries, 4] with the predicted box coordinates targets: This is a list of targets (len(targets) = batch_size), where each target is a dict containing: "labels": Tensor of dim [num_target_boxes] (where num_target_boxes is the number of ground-truth objects in the target) containing the class labels "boxes": Tensor of dim [num_target_boxes, 4] containing the target box coordinates Returns: A list of size batch_size, containing tuples of (index_i, index_j) where: - index_i is the indices of the selected predictions (in order) - index_j is the indices of the corresponding selected targets (in order) For each batch element, it holds: len(index_i) = len(index_j) = min(num_queries, num_target_boxes) """ bs, num_queries = outputs["pred_logits"].shape[:2] # We flatten to compute the cost matrices in a batch out_prob = outputs["pred_logits"].flatten(0, 1).softmax(-1) # [batch_size * num_queries, num_classes] out_bbox = outputs["pred_boxes"].flatten(0, 1) # [batch_size * num_queries, 4] # Also concat the target labels and boxes tgt_ids = torch.cat([v["labels"] for v in targets]) tgt_bbox = torch.cat([v["boxes"] for v in targets]) # Compute the classification cost. Contrary to the loss, we don't use the NLL, # but approximate it in 1 - proba[target class]. # The 1 is a constant that doesn't change the matching, it can be ommitted. cost_class = -out_prob[:, tgt_ids] # Compute the L1 cost between boxes cost_bbox = torch.cdist(out_bbox, tgt_bbox, p=1) # Compute the giou cost betwen boxes cost_giou = -generalized_box_iou(box_cxcywh_to_xyxy(out_bbox), box_cxcywh_to_xyxy(tgt_bbox)) # Final cost matrix C = self.cost_bbox * cost_bbox + self.cost_class * cost_class + self.cost_giou * cost_giou C = C.view(bs, num_queries, -1).cpu() sizes = [len(v["boxes"]) for v in targets] indices = [linear_sum_assignment(c[i]) for i, c in enumerate(C.split(sizes, -1))] return [(torch.as_tensor(i, dtype=torch.int64), torch.as_tensor(j, dtype=torch.int64)) for i, j in indices] def build_matcher(args): return HungarianMatcher(cost_class=args.set_cost_class, cost_bbox=args.set_cost_bbox, cost_giou=args.set_cost_giou) ================================================ FILE: models/position_encoding.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Various positional encodings for the transformer. """ import math import torch from torch import nn from util.misc import NestedTensor class PositionEmbeddingSine(nn.Module): """ This is a more standard version of the position embedding, very similar to the one used by the Attention is all you need paper, generalized to work on images. """ def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None): super().__init__() self.num_pos_feats = num_pos_feats self.temperature = temperature self.normalize = normalize if scale is not None and normalize is False: raise ValueError("normalize should be True if scale is passed") if scale is None: scale = 2 * math.pi self.scale = scale def forward(self, tensor_list: NestedTensor): x = tensor_list.tensors mask = tensor_list.mask assert mask is not None not_mask = ~mask y_embed = not_mask.cumsum(1, dtype=torch.float32) x_embed = not_mask.cumsum(2, dtype=torch.float32) if self.normalize: eps = 1e-6 y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale dim_t = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device) dim_t = self.temperature ** (2 * (dim_t // 2) / self.num_pos_feats) pos_x = x_embed[:, :, :, None] / dim_t pos_y = y_embed[:, :, :, None] / dim_t pos_x = torch.stack((pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4).flatten(3) pos_y = torch.stack((pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4).flatten(3) pos = torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2) return pos class PositionEmbeddingLearned(nn.Module): """ Absolute pos embedding, learned. """ def __init__(self, num_pos_feats=256): super().__init__() self.row_embed = nn.Embedding(50, num_pos_feats) self.col_embed = nn.Embedding(50, num_pos_feats) self.reset_parameters() def reset_parameters(self): nn.init.uniform_(self.row_embed.weight) nn.init.uniform_(self.col_embed.weight) def forward(self, tensor_list: NestedTensor): x = tensor_list.tensors h, w = x.shape[-2:] i = torch.arange(w, device=x.device) j = torch.arange(h, device=x.device) x_emb = self.col_embed(i) y_emb = self.row_embed(j) pos = torch.cat([ x_emb.unsqueeze(0).repeat(h, 1, 1), y_emb.unsqueeze(1).repeat(1, w, 1), ], dim=-1).permute(2, 0, 1).unsqueeze(0).repeat(x.shape[0], 1, 1, 1) return pos def build_position_encoding(args): N_steps = args.hidden_dim // 2 if args.position_embedding in ('v2', 'sine'): # TODO find a better way of exposing other arguments position_embedding = PositionEmbeddingSine(N_steps, normalize=True) elif args.position_embedding in ('v3', 'learned'): position_embedding = PositionEmbeddingLearned(N_steps) else: raise ValueError(f"not supported {args.position_embedding}") return position_embedding ================================================ FILE: models/segmentation.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ This file provides the definition of the convolutional heads used to predict masks, as well as the losses """ import io from collections import defaultdict from typing import List, Optional import torch import torch.nn as nn import torch.nn.functional as F from torch import Tensor from PIL import Image import util.box_ops as box_ops from util.misc import NestedTensor, interpolate, nested_tensor_from_tensor_list try: from panopticapi.utils import id2rgb, rgb2id except ImportError: pass class DETRsegm(nn.Module): def __init__(self, detr, freeze_detr=False): super().__init__() self.detr = detr if freeze_detr: for p in self.parameters(): p.requires_grad_(False) hidden_dim, nheads = detr.transformer.d_model, detr.transformer.nhead self.bbox_attention = MHAttentionMap(hidden_dim, hidden_dim, nheads, dropout=0.0) self.mask_head = MaskHeadSmallConv(hidden_dim + nheads, [1024, 512, 256], hidden_dim) def forward(self, samples: NestedTensor): if isinstance(samples, (list, torch.Tensor)): samples = nested_tensor_from_tensor_list(samples) features, pos = self.detr.backbone(samples) bs = features[-1].tensors.shape[0] src, mask = features[-1].decompose() assert mask is not None src_proj = self.detr.input_proj(src) hs, memory = self.detr.transformer(src_proj, mask, self.detr.query_embed.weight, pos[-1]) outputs_class = self.detr.class_embed(hs) outputs_coord = self.detr.bbox_embed(hs).sigmoid() out = {"pred_logits": outputs_class[-1], "pred_boxes": outputs_coord[-1]} if self.detr.aux_loss: out['aux_outputs'] = self.detr._set_aux_loss(outputs_class, outputs_coord) # FIXME h_boxes takes the last one computed, keep this in mind bbox_mask = self.bbox_attention(hs[-1], memory, mask=mask) seg_masks = self.mask_head(src_proj, bbox_mask, [features[2].tensors, features[1].tensors, features[0].tensors]) outputs_seg_masks = seg_masks.view(bs, self.detr.num_queries, seg_masks.shape[-2], seg_masks.shape[-1]) out["pred_masks"] = outputs_seg_masks return out def _expand(tensor, length: int): return tensor.unsqueeze(1).repeat(1, int(length), 1, 1, 1).flatten(0, 1) class MaskHeadSmallConv(nn.Module): """ Simple convolutional head, using group norm. Upsampling is done using a FPN approach """ def __init__(self, dim, fpn_dims, context_dim): super().__init__() inter_dims = [dim, context_dim // 2, context_dim // 4, context_dim // 8, context_dim // 16, context_dim // 64] self.lay1 = torch.nn.Conv2d(dim, dim, 3, padding=1) self.gn1 = torch.nn.GroupNorm(8, dim) self.lay2 = torch.nn.Conv2d(dim, inter_dims[1], 3, padding=1) self.gn2 = torch.nn.GroupNorm(8, inter_dims[1]) self.lay3 = torch.nn.Conv2d(inter_dims[1], inter_dims[2], 3, padding=1) self.gn3 = torch.nn.GroupNorm(8, inter_dims[2]) self.lay4 = torch.nn.Conv2d(inter_dims[2], inter_dims[3], 3, padding=1) self.gn4 = torch.nn.GroupNorm(8, inter_dims[3]) self.lay5 = torch.nn.Conv2d(inter_dims[3], inter_dims[4], 3, padding=1) self.gn5 = torch.nn.GroupNorm(8, inter_dims[4]) self.out_lay = torch.nn.Conv2d(inter_dims[4], 1, 3, padding=1) self.dim = dim self.adapter1 = torch.nn.Conv2d(fpn_dims[0], inter_dims[1], 1) self.adapter2 = torch.nn.Conv2d(fpn_dims[1], inter_dims[2], 1) self.adapter3 = torch.nn.Conv2d(fpn_dims[2], inter_dims[3], 1) for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_uniform_(m.weight, a=1) nn.init.constant_(m.bias, 0) def forward(self, x: Tensor, bbox_mask: Tensor, fpns: List[Tensor]): x = torch.cat([_expand(x, bbox_mask.shape[1]), bbox_mask.flatten(0, 1)], 1) x = self.lay1(x) x = self.gn1(x) x = F.relu(x) x = self.lay2(x) x = self.gn2(x) x = F.relu(x) cur_fpn = self.adapter1(fpns[0]) if cur_fpn.size(0) != x.size(0): cur_fpn = _expand(cur_fpn, x.size(0) // cur_fpn.size(0)) x = cur_fpn + F.interpolate(x, size=cur_fpn.shape[-2:], mode="nearest") x = self.lay3(x) x = self.gn3(x) x = F.relu(x) cur_fpn = self.adapter2(fpns[1]) if cur_fpn.size(0) != x.size(0): cur_fpn = _expand(cur_fpn, x.size(0) // cur_fpn.size(0)) x = cur_fpn + F.interpolate(x, size=cur_fpn.shape[-2:], mode="nearest") x = self.lay4(x) x = self.gn4(x) x = F.relu(x) cur_fpn = self.adapter3(fpns[2]) if cur_fpn.size(0) != x.size(0): cur_fpn = _expand(cur_fpn, x.size(0) // cur_fpn.size(0)) x = cur_fpn + F.interpolate(x, size=cur_fpn.shape[-2:], mode="nearest") x = self.lay5(x) x = self.gn5(x) x = F.relu(x) x = self.out_lay(x) return x class MHAttentionMap(nn.Module): """This is a 2D attention module, which only returns the attention softmax (no multiplication by value)""" def __init__(self, query_dim, hidden_dim, num_heads, dropout=0.0, bias=True): super().__init__() self.num_heads = num_heads self.hidden_dim = hidden_dim self.dropout = nn.Dropout(dropout) self.q_linear = nn.Linear(query_dim, hidden_dim, bias=bias) self.k_linear = nn.Linear(query_dim, hidden_dim, bias=bias) nn.init.zeros_(self.k_linear.bias) nn.init.zeros_(self.q_linear.bias) nn.init.xavier_uniform_(self.k_linear.weight) nn.init.xavier_uniform_(self.q_linear.weight) self.normalize_fact = float(hidden_dim / self.num_heads) ** -0.5 def forward(self, q, k, mask: Optional[Tensor] = None): q = self.q_linear(q) k = F.conv2d(k, self.k_linear.weight.unsqueeze(-1).unsqueeze(-1), self.k_linear.bias) qh = q.view(q.shape[0], q.shape[1], self.num_heads, self.hidden_dim // self.num_heads) kh = k.view(k.shape[0], self.num_heads, self.hidden_dim // self.num_heads, k.shape[-2], k.shape[-1]) weights = torch.einsum("bqnc,bnchw->bqnhw", qh * self.normalize_fact, kh) if mask is not None: weights.masked_fill_(mask.unsqueeze(1).unsqueeze(1), float("-inf")) weights = F.softmax(weights.flatten(2), dim=-1).view(weights.size()) weights = self.dropout(weights) return weights def dice_loss(inputs, targets, num_boxes): """ Compute the DICE loss, similar to generalized IOU for masks Args: inputs: A float tensor of arbitrary shape. The predictions for each example. targets: A float tensor with the same shape as inputs. Stores the binary classification label for each element in inputs (0 for the negative class and 1 for the positive class). """ inputs = inputs.sigmoid() inputs = inputs.flatten(1) numerator = 2 * (inputs * targets).sum(1) denominator = inputs.sum(-1) + targets.sum(-1) loss = 1 - (numerator + 1) / (denominator + 1) return loss.sum() / num_boxes def sigmoid_focal_loss(inputs, targets, num_boxes, alpha: float = 0.25, gamma: float = 2): """ Loss used in RetinaNet for dense detection: https://arxiv.org/abs/1708.02002. Args: inputs: A float tensor of arbitrary shape. The predictions for each example. targets: A float tensor with the same shape as inputs. Stores the binary classification label for each element in inputs (0 for the negative class and 1 for the positive class). alpha: (optional) Weighting factor in range (0,1) to balance positive vs negative examples. Default = -1 (no weighting). gamma: Exponent of the modulating factor (1 - p_t) to balance easy vs hard examples. Returns: Loss tensor """ prob = inputs.sigmoid() ce_loss = F.binary_cross_entropy_with_logits(inputs, targets, reduction="none") p_t = prob * targets + (1 - prob) * (1 - targets) loss = ce_loss * ((1 - p_t) ** gamma) if alpha >= 0: alpha_t = alpha * targets + (1 - alpha) * (1 - targets) loss = alpha_t * loss return loss.mean(1).sum() / num_boxes class PostProcessSegm(nn.Module): def __init__(self, threshold=0.5): super().__init__() self.threshold = threshold @torch.no_grad() def forward(self, results, outputs, orig_target_sizes, max_target_sizes): assert len(orig_target_sizes) == len(max_target_sizes) max_h, max_w = max_target_sizes.max(0)[0].tolist() outputs_masks = outputs["pred_masks"].squeeze(2) outputs_masks = F.interpolate(outputs_masks, size=(max_h, max_w), mode="bilinear", align_corners=False) outputs_masks = (outputs_masks.sigmoid() > self.threshold).cpu() for i, (cur_mask, t, tt) in enumerate(zip(outputs_masks, max_target_sizes, orig_target_sizes)): img_h, img_w = t[0], t[1] results[i]["masks"] = cur_mask[:, :img_h, :img_w].unsqueeze(1) results[i]["masks"] = F.interpolate( results[i]["masks"].float(), size=tuple(tt.tolist()), mode="nearest" ).byte() return results class PostProcessPanoptic(nn.Module): """This class converts the output of the model to the final panoptic result, in the format expected by the coco panoptic API """ def __init__(self, is_thing_map, threshold=0.85): """ Parameters: is_thing_map: This is a whose keys are the class ids, and the values a boolean indicating whether the class is a thing (True) or a stuff (False) class threshold: confidence threshold: segments with confidence lower than this will be deleted """ super().__init__() self.threshold = threshold self.is_thing_map = is_thing_map def forward(self, outputs, processed_sizes, target_sizes=None): """ This function computes the panoptic prediction from the model's predictions. Parameters: outputs: This is a dict coming directly from the model. See the model doc for the content. processed_sizes: This is a list of tuples (or torch tensors) of sizes of the images that were passed to the model, ie the size after data augmentation but before batching. target_sizes: This is a list of tuples (or torch tensors) corresponding to the requested final size of each prediction. If left to None, it will default to the processed_sizes """ if target_sizes is None: target_sizes = processed_sizes assert len(processed_sizes) == len(target_sizes) out_logits, raw_masks, raw_boxes = outputs["pred_logits"], outputs["pred_masks"], outputs["pred_boxes"] assert len(out_logits) == len(raw_masks) == len(target_sizes) preds = [] def to_tuple(tup): if isinstance(tup, tuple): return tup return tuple(tup.cpu().tolist()) for cur_logits, cur_masks, cur_boxes, size, target_size in zip( out_logits, raw_masks, raw_boxes, processed_sizes, target_sizes ): # we filter empty queries and detection below threshold scores, labels = cur_logits.softmax(-1).max(-1) keep = labels.ne(outputs["pred_logits"].shape[-1] - 1) & (scores > self.threshold) cur_scores, cur_classes = cur_logits.softmax(-1).max(-1) cur_scores = cur_scores[keep] cur_classes = cur_classes[keep] cur_masks = cur_masks[keep] cur_masks = interpolate(cur_masks[:, None], to_tuple(size), mode="bilinear").squeeze(1) cur_boxes = box_ops.box_cxcywh_to_xyxy(cur_boxes[keep]) h, w = cur_masks.shape[-2:] assert len(cur_boxes) == len(cur_classes) # It may be that we have several predicted masks for the same stuff class. # In the following, we track the list of masks ids for each stuff class (they are merged later on) cur_masks = cur_masks.flatten(1) stuff_equiv_classes = defaultdict(lambda: []) for k, label in enumerate(cur_classes): if not self.is_thing_map[label.item()]: stuff_equiv_classes[label.item()].append(k) def get_ids_area(masks, scores, dedup=False): # This helper function creates the final panoptic segmentation image # It also returns the area of the masks that appears on the image m_id = masks.transpose(0, 1).softmax(-1) if m_id.shape[-1] == 0: # We didn't detect any mask :( m_id = torch.zeros((h, w), dtype=torch.long, device=m_id.device) else: m_id = m_id.argmax(-1).view(h, w) if dedup: # Merge the masks corresponding to the same stuff class for equiv in stuff_equiv_classes.values(): if len(equiv) > 1: for eq_id in equiv: m_id.masked_fill_(m_id.eq(eq_id), equiv[0]) final_h, final_w = to_tuple(target_size) seg_img = Image.fromarray(id2rgb(m_id.view(h, w).cpu().numpy())) seg_img = seg_img.resize(size=(final_w, final_h), resample=Image.NEAREST) np_seg_img = ( torch.ByteTensor(torch.ByteStorage.from_buffer(seg_img.tobytes())).view(final_h, final_w, 3).numpy() ) m_id = torch.from_numpy(rgb2id(np_seg_img)) area = [] for i in range(len(scores)): area.append(m_id.eq(i).sum().item()) return area, seg_img area, seg_img = get_ids_area(cur_masks, cur_scores, dedup=True) if cur_classes.numel() > 0: # We know filter empty masks as long as we find some while True: filtered_small = torch.as_tensor( [area[i] <= 4 for i, c in enumerate(cur_classes)], dtype=torch.bool, device=keep.device ) if filtered_small.any().item(): cur_scores = cur_scores[~filtered_small] cur_classes = cur_classes[~filtered_small] cur_masks = cur_masks[~filtered_small] area, seg_img = get_ids_area(cur_masks, cur_scores) else: break else: cur_classes = torch.ones(1, dtype=torch.long, device=cur_classes.device) segments_info = [] for i, a in enumerate(area): cat = cur_classes[i].item() segments_info.append({"id": i, "isthing": self.is_thing_map[cat], "category_id": cat, "area": a}) del cur_classes with io.BytesIO() as out: seg_img.save(out, format="PNG") predictions = {"png_string": out.getvalue(), "segments_info": segments_info} preds.append(predictions) return preds ================================================ FILE: models/transformer.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ DETR Transformer class. Copy-paste from torch.nn.Transformer with modifications: * positional encodings are passed in MHattention * extra LN at the end of encoder is removed * decoder returns a stack of activations from all decoding layers """ import copy from typing import Optional, List import torch import torch.nn.functional as F from torch import nn, Tensor class Transformer(nn.Module): def __init__(self, d_model=512, nhead=8, num_encoder_layers=6, num_decoder_layers=6, dim_feedforward=2048, dropout=0.1, activation="relu", normalize_before=False, return_intermediate_dec=False): super().__init__() encoder_layer = TransformerEncoderLayer(d_model, nhead, dim_feedforward, dropout, activation, normalize_before) encoder_norm = nn.LayerNorm(d_model) if normalize_before else None self.encoder = TransformerEncoder(encoder_layer, num_encoder_layers, encoder_norm) decoder_layer = TransformerDecoderLayer(d_model, nhead, dim_feedforward, dropout, activation, normalize_before) decoder_norm = nn.LayerNorm(d_model) self.decoder = TransformerDecoder(decoder_layer, num_decoder_layers, decoder_norm, return_intermediate=return_intermediate_dec) self._reset_parameters() self.d_model = d_model self.nhead = nhead def _reset_parameters(self): for p in self.parameters(): if p.dim() > 1: nn.init.xavier_uniform_(p) def forward(self, src, mask, query_embed, pos_embed): # flatten NxCxHxW to HWxNxC bs, c, h, w = src.shape src = src.flatten(2).permute(2, 0, 1) pos_embed = pos_embed.flatten(2).permute(2, 0, 1) query_embed = query_embed.unsqueeze(1).repeat(1, bs, 1) mask = mask.flatten(1) tgt = torch.zeros_like(query_embed) memory = self.encoder(src, src_key_padding_mask=mask, pos=pos_embed) hs = self.decoder(tgt, memory, memory_key_padding_mask=mask, pos=pos_embed, query_pos=query_embed) return hs.transpose(1, 2), memory.permute(1, 2, 0).view(bs, c, h, w) class TransformerEncoder(nn.Module): def __init__(self, encoder_layer, num_layers, norm=None): super().__init__() self.layers = _get_clones(encoder_layer, num_layers) self.num_layers = num_layers self.norm = norm def forward(self, src, mask: Optional[Tensor] = None, src_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None): output = src for layer in self.layers: output = layer(output, src_mask=mask, src_key_padding_mask=src_key_padding_mask, pos=pos) if self.norm is not None: output = self.norm(output) return output class TransformerDecoder(nn.Module): def __init__(self, decoder_layer, num_layers, norm=None, return_intermediate=False): super().__init__() self.layers = _get_clones(decoder_layer, num_layers) self.num_layers = num_layers self.norm = norm self.return_intermediate = return_intermediate def forward(self, tgt, memory, tgt_mask: Optional[Tensor] = None, memory_mask: Optional[Tensor] = None, tgt_key_padding_mask: Optional[Tensor] = None, memory_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None, query_pos: Optional[Tensor] = None): output = tgt intermediate = [] for layer in self.layers: output = layer(output, memory, tgt_mask=tgt_mask, memory_mask=memory_mask, tgt_key_padding_mask=tgt_key_padding_mask, memory_key_padding_mask=memory_key_padding_mask, pos=pos, query_pos=query_pos) if self.return_intermediate: intermediate.append(self.norm(output)) if self.norm is not None: output = self.norm(output) if self.return_intermediate: intermediate.pop() intermediate.append(output) if self.return_intermediate: return torch.stack(intermediate) return output.unsqueeze(0) class TransformerEncoderLayer(nn.Module): def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1, activation="relu", normalize_before=False): super().__init__() self.self_attn = nn.MultiheadAttention(d_model, nhead, dropout=dropout) # Implementation of Feedforward model self.linear1 = nn.Linear(d_model, dim_feedforward) self.dropout = nn.Dropout(dropout) self.linear2 = nn.Linear(dim_feedforward, d_model) self.norm1 = nn.LayerNorm(d_model) self.norm2 = nn.LayerNorm(d_model) self.dropout1 = nn.Dropout(dropout) self.dropout2 = nn.Dropout(dropout) self.activation = _get_activation_fn(activation) self.normalize_before = normalize_before def with_pos_embed(self, tensor, pos: Optional[Tensor]): return tensor if pos is None else tensor + pos def forward_post(self, src, src_mask: Optional[Tensor] = None, src_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None): q = k = self.with_pos_embed(src, pos) src2 = self.self_attn(q, k, value=src, attn_mask=src_mask, key_padding_mask=src_key_padding_mask)[0] src = src + self.dropout1(src2) src = self.norm1(src) src2 = self.linear2(self.dropout(self.activation(self.linear1(src)))) src = src + self.dropout2(src2) src = self.norm2(src) return src def forward_pre(self, src, src_mask: Optional[Tensor] = None, src_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None): src2 = self.norm1(src) q = k = self.with_pos_embed(src2, pos) src2 = self.self_attn(q, k, value=src2, attn_mask=src_mask, key_padding_mask=src_key_padding_mask)[0] src = src + self.dropout1(src2) src2 = self.norm2(src) src2 = self.linear2(self.dropout(self.activation(self.linear1(src2)))) src = src + self.dropout2(src2) return src def forward(self, src, src_mask: Optional[Tensor] = None, src_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None): if self.normalize_before: return self.forward_pre(src, src_mask, src_key_padding_mask, pos) return self.forward_post(src, src_mask, src_key_padding_mask, pos) class TransformerDecoderLayer(nn.Module): def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1, activation="relu", normalize_before=False): super().__init__() self.self_attn = nn.MultiheadAttention(d_model, nhead, dropout=dropout) self.multihead_attn = nn.MultiheadAttention(d_model, nhead, dropout=dropout) # Implementation of Feedforward model self.linear1 = nn.Linear(d_model, dim_feedforward) self.dropout = nn.Dropout(dropout) self.linear2 = nn.Linear(dim_feedforward, d_model) self.norm1 = nn.LayerNorm(d_model) self.norm2 = nn.LayerNorm(d_model) self.norm3 = nn.LayerNorm(d_model) self.dropout1 = nn.Dropout(dropout) self.dropout2 = nn.Dropout(dropout) self.dropout3 = nn.Dropout(dropout) self.activation = _get_activation_fn(activation) self.normalize_before = normalize_before def with_pos_embed(self, tensor, pos: Optional[Tensor]): return tensor if pos is None else tensor + pos def forward_post(self, tgt, memory, tgt_mask: Optional[Tensor] = None, memory_mask: Optional[Tensor] = None, tgt_key_padding_mask: Optional[Tensor] = None, memory_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None, query_pos: Optional[Tensor] = None): q = k = self.with_pos_embed(tgt, query_pos) tgt2 = self.self_attn(q, k, value=tgt, attn_mask=tgt_mask, key_padding_mask=tgt_key_padding_mask)[0] tgt = tgt + self.dropout1(tgt2) tgt = self.norm1(tgt) tgt2 = self.multihead_attn(query=self.with_pos_embed(tgt, query_pos), key=self.with_pos_embed(memory, pos), value=memory, attn_mask=memory_mask, key_padding_mask=memory_key_padding_mask)[0] tgt = tgt + self.dropout2(tgt2) tgt = self.norm2(tgt) tgt2 = self.linear2(self.dropout(self.activation(self.linear1(tgt)))) tgt = tgt + self.dropout3(tgt2) tgt = self.norm3(tgt) return tgt def forward_pre(self, tgt, memory, tgt_mask: Optional[Tensor] = None, memory_mask: Optional[Tensor] = None, tgt_key_padding_mask: Optional[Tensor] = None, memory_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None, query_pos: Optional[Tensor] = None): tgt2 = self.norm1(tgt) q = k = self.with_pos_embed(tgt2, query_pos) tgt2 = self.self_attn(q, k, value=tgt2, attn_mask=tgt_mask, key_padding_mask=tgt_key_padding_mask)[0] tgt = tgt + self.dropout1(tgt2) tgt2 = self.norm2(tgt) tgt2 = self.multihead_attn(query=self.with_pos_embed(tgt2, query_pos), key=self.with_pos_embed(memory, pos), value=memory, attn_mask=memory_mask, key_padding_mask=memory_key_padding_mask)[0] tgt = tgt + self.dropout2(tgt2) tgt2 = self.norm3(tgt) tgt2 = self.linear2(self.dropout(self.activation(self.linear1(tgt2)))) tgt = tgt + self.dropout3(tgt2) return tgt def forward(self, tgt, memory, tgt_mask: Optional[Tensor] = None, memory_mask: Optional[Tensor] = None, tgt_key_padding_mask: Optional[Tensor] = None, memory_key_padding_mask: Optional[Tensor] = None, pos: Optional[Tensor] = None, query_pos: Optional[Tensor] = None): if self.normalize_before: return self.forward_pre(tgt, memory, tgt_mask, memory_mask, tgt_key_padding_mask, memory_key_padding_mask, pos, query_pos) return self.forward_post(tgt, memory, tgt_mask, memory_mask, tgt_key_padding_mask, memory_key_padding_mask, pos, query_pos) def _get_clones(module, N): return nn.ModuleList([copy.deepcopy(module) for i in range(N)]) def build_transformer(args): return Transformer( d_model=args.hidden_dim, dropout=args.dropout, nhead=args.nheads, dim_feedforward=args.dim_feedforward, num_encoder_layers=args.enc_layers, num_decoder_layers=args.dec_layers, normalize_before=args.pre_norm, return_intermediate_dec=True, ) def _get_activation_fn(activation): """Return an activation function given a string""" if activation == "relu": return F.relu if activation == "gelu": return F.gelu if activation == "glu": return F.glu raise RuntimeError(F"activation should be relu/gelu, not {activation}.") ================================================ FILE: requirements.txt ================================================ cython git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI&egg=pycocotools submitit torch>=1.5.0 torchvision>=0.6.0 git+https://github.com/cocodataset/panopticapi.git#egg=panopticapi scipy onnx onnxruntime ================================================ FILE: run_with_submitit.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ A script to run multinode training with submitit. """ import argparse import os import uuid from pathlib import Path import main as detection import submitit def parse_args(): detection_parser = detection.get_args_parser() parser = argparse.ArgumentParser("Submitit for detection", parents=[detection_parser]) parser.add_argument("--ngpus", default=8, type=int, help="Number of gpus to request on each node") parser.add_argument("--nodes", default=4, type=int, help="Number of nodes to request") parser.add_argument("--timeout", default=60, type=int, help="Duration of the job") parser.add_argument("--job_dir", default="", type=str, help="Job dir. Leave empty for automatic.") return parser.parse_args() def get_shared_folder() -> Path: user = os.getenv("USER") if Path("/checkpoint/").is_dir(): p = Path(f"/checkpoint/{user}/experiments") p.mkdir(exist_ok=True) return p raise RuntimeError("No shared folder available") def get_init_file(): # Init file must not exist, but it's parent dir must exist. os.makedirs(str(get_shared_folder()), exist_ok=True) init_file = get_shared_folder() / f"{uuid.uuid4().hex}_init" if init_file.exists(): os.remove(str(init_file)) return init_file class Trainer(object): def __init__(self, args): self.args = args def __call__(self): import main as detection self._setup_gpu_args() detection.main(self.args) def checkpoint(self): import os import submitit from pathlib import Path self.args.dist_url = get_init_file().as_uri() checkpoint_file = os.path.join(self.args.output_dir, "checkpoint.pth") if os.path.exists(checkpoint_file): self.args.resume = checkpoint_file print("Requeuing ", self.args) empty_trainer = type(self)(self.args) return submitit.helpers.DelayedSubmission(empty_trainer) def _setup_gpu_args(self): import submitit from pathlib import Path job_env = submitit.JobEnvironment() self.args.output_dir = Path(str(self.args.output_dir).replace("%j", str(job_env.job_id))) self.args.gpu = job_env.local_rank self.args.rank = job_env.global_rank self.args.world_size = job_env.num_tasks print(f"Process group: {job_env.num_tasks} tasks, rank: {job_env.global_rank}") def main(): args = parse_args() if args.job_dir == "": args.job_dir = get_shared_folder() / "%j" # Note that the folder will depend on the job_id, to easily track experiments executor = submitit.AutoExecutor(folder=args.job_dir, slurm_max_num_timeout=30) # cluster setup is defined by environment variables num_gpus_per_node = args.ngpus nodes = args.nodes timeout_min = args.timeout executor.update_parameters( mem_gb=40 * num_gpus_per_node, gpus_per_node=num_gpus_per_node, tasks_per_node=num_gpus_per_node, # one task per GPU cpus_per_task=10, nodes=nodes, timeout_min=timeout_min, # max is 60 * 72 ) executor.update_parameters(name="detr") args.dist_url = get_init_file().as_uri() args.output_dir = args.job_dir trainer = Trainer(args) job = executor.submit(trainer) print("Submitted job_id:", job.job_id) if __name__ == "__main__": main() ================================================ FILE: test_all.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved import io import unittest import torch from torch import nn, Tensor from typing import List from models.matcher import HungarianMatcher from models.position_encoding import PositionEmbeddingSine, PositionEmbeddingLearned from models.backbone import Backbone, Joiner, BackboneBase from util import box_ops from util.misc import nested_tensor_from_tensor_list from hubconf import detr_resnet50, detr_resnet50_panoptic # onnxruntime requires python 3.5 or above try: import onnxruntime except ImportError: onnxruntime = None class Tester(unittest.TestCase): def test_box_cxcywh_to_xyxy(self): t = torch.rand(10, 4) r = box_ops.box_xyxy_to_cxcywh(box_ops.box_cxcywh_to_xyxy(t)) self.assertLess((t - r).abs().max(), 1e-5) @staticmethod def indices_torch2python(indices): return [(i.tolist(), j.tolist()) for i, j in indices] def test_hungarian(self): n_queries, n_targets, n_classes = 100, 15, 91 logits = torch.rand(1, n_queries, n_classes + 1) boxes = torch.rand(1, n_queries, 4) tgt_labels = torch.randint(high=n_classes, size=(n_targets,)) tgt_boxes = torch.rand(n_targets, 4) matcher = HungarianMatcher() targets = [{'labels': tgt_labels, 'boxes': tgt_boxes}] indices_single = matcher({'pred_logits': logits, 'pred_boxes': boxes}, targets) indices_batched = matcher({'pred_logits': logits.repeat(2, 1, 1), 'pred_boxes': boxes.repeat(2, 1, 1)}, targets * 2) self.assertEqual(len(indices_single[0][0]), n_targets) self.assertEqual(len(indices_single[0][1]), n_targets) self.assertEqual(self.indices_torch2python(indices_single), self.indices_torch2python([indices_batched[0]])) self.assertEqual(self.indices_torch2python(indices_single), self.indices_torch2python([indices_batched[1]])) # test with empty targets tgt_labels_empty = torch.randint(high=n_classes, size=(0,)) tgt_boxes_empty = torch.rand(0, 4) targets_empty = [{'labels': tgt_labels_empty, 'boxes': tgt_boxes_empty}] indices = matcher({'pred_logits': logits.repeat(2, 1, 1), 'pred_boxes': boxes.repeat(2, 1, 1)}, targets + targets_empty) self.assertEqual(len(indices[1][0]), 0) indices = matcher({'pred_logits': logits.repeat(2, 1, 1), 'pred_boxes': boxes.repeat(2, 1, 1)}, targets_empty * 2) self.assertEqual(len(indices[0][0]), 0) def test_position_encoding_script(self): m1, m2 = PositionEmbeddingSine(), PositionEmbeddingLearned() mm1, mm2 = torch.jit.script(m1), torch.jit.script(m2) # noqa def test_backbone_script(self): backbone = Backbone('resnet50', True, False, False) torch.jit.script(backbone) # noqa def test_model_script_detection(self): model = detr_resnet50(pretrained=False).eval() scripted_model = torch.jit.script(model) x = nested_tensor_from_tensor_list([torch.rand(3, 200, 200), torch.rand(3, 200, 250)]) out = model(x) out_script = scripted_model(x) self.assertTrue(out["pred_logits"].equal(out_script["pred_logits"])) self.assertTrue(out["pred_boxes"].equal(out_script["pred_boxes"])) def test_model_script_panoptic(self): model = detr_resnet50_panoptic(pretrained=False).eval() scripted_model = torch.jit.script(model) x = nested_tensor_from_tensor_list([torch.rand(3, 200, 200), torch.rand(3, 200, 250)]) out = model(x) out_script = scripted_model(x) self.assertTrue(out["pred_logits"].equal(out_script["pred_logits"])) self.assertTrue(out["pred_boxes"].equal(out_script["pred_boxes"])) self.assertTrue(out["pred_masks"].equal(out_script["pred_masks"])) def test_model_detection_different_inputs(self): model = detr_resnet50(pretrained=False).eval() # support NestedTensor x = nested_tensor_from_tensor_list([torch.rand(3, 200, 200), torch.rand(3, 200, 250)]) out = model(x) self.assertIn('pred_logits', out) # and 4d Tensor x = torch.rand(1, 3, 200, 200) out = model(x) self.assertIn('pred_logits', out) # and List[Tensor[C, H, W]] x = torch.rand(3, 200, 200) out = model([x]) self.assertIn('pred_logits', out) def test_warpped_model_script_detection(self): class WrappedDETR(nn.Module): def __init__(self, model): super().__init__() self.model = model def forward(self, inputs: List[Tensor]): sample = nested_tensor_from_tensor_list(inputs) return self.model(sample) model = detr_resnet50(pretrained=False) wrapped_model = WrappedDETR(model) wrapped_model.eval() scripted_model = torch.jit.script(wrapped_model) x = [torch.rand(3, 200, 200), torch.rand(3, 200, 250)] out = wrapped_model(x) out_script = scripted_model(x) self.assertTrue(out["pred_logits"].equal(out_script["pred_logits"])) self.assertTrue(out["pred_boxes"].equal(out_script["pred_boxes"])) @unittest.skipIf(onnxruntime is None, 'ONNX Runtime unavailable') class ONNXExporterTester(unittest.TestCase): @classmethod def setUpClass(cls): torch.manual_seed(123) def run_model(self, model, inputs_list, tolerate_small_mismatch=False, do_constant_folding=True, dynamic_axes=None, output_names=None, input_names=None): model.eval() onnx_io = io.BytesIO() # export to onnx with the first input torch.onnx.export(model, inputs_list[0], onnx_io, do_constant_folding=do_constant_folding, opset_version=12, dynamic_axes=dynamic_axes, input_names=input_names, output_names=output_names) # validate the exported model with onnx runtime for test_inputs in inputs_list: with torch.no_grad(): if isinstance(test_inputs, torch.Tensor) or isinstance(test_inputs, list): test_inputs = (nested_tensor_from_tensor_list(test_inputs),) test_ouputs = model(*test_inputs) if isinstance(test_ouputs, torch.Tensor): test_ouputs = (test_ouputs,) self.ort_validate(onnx_io, test_inputs, test_ouputs, tolerate_small_mismatch) def ort_validate(self, onnx_io, inputs, outputs, tolerate_small_mismatch=False): inputs, _ = torch.jit._flatten(inputs) outputs, _ = torch.jit._flatten(outputs) def to_numpy(tensor): if tensor.requires_grad: return tensor.detach().cpu().numpy() else: return tensor.cpu().numpy() inputs = list(map(to_numpy, inputs)) outputs = list(map(to_numpy, outputs)) ort_session = onnxruntime.InferenceSession(onnx_io.getvalue()) # compute onnxruntime output prediction ort_inputs = dict((ort_session.get_inputs()[i].name, inpt) for i, inpt in enumerate(inputs)) ort_outs = ort_session.run(None, ort_inputs) for i, element in enumerate(outputs): try: torch.testing.assert_allclose(element, ort_outs[i], rtol=1e-03, atol=1e-05) except AssertionError as error: if tolerate_small_mismatch: self.assertIn("(0.00%)", str(error), str(error)) else: raise def test_model_onnx_detection(self): model = detr_resnet50(pretrained=False).eval() dummy_image = torch.ones(1, 3, 800, 800) * 0.3 model(dummy_image) # Test exported model on images of different size, or dummy input self.run_model( model, [(torch.rand(1, 3, 750, 800),)], input_names=["inputs"], output_names=["pred_logits", "pred_boxes"], tolerate_small_mismatch=True, ) @unittest.skip("CI doesn't have enough memory") def test_model_onnx_detection_panoptic(self): model = detr_resnet50_panoptic(pretrained=False).eval() dummy_image = torch.ones(1, 3, 800, 800) * 0.3 model(dummy_image) # Test exported model on images of different size, or dummy input self.run_model( model, [(torch.rand(1, 3, 750, 800),)], input_names=["inputs"], output_names=["pred_logits", "pred_boxes", "pred_masks"], tolerate_small_mismatch=True, ) if __name__ == '__main__': unittest.main() ================================================ FILE: tox.ini ================================================ [flake8] max-line-length = 120 ignore = F401,E402,F403,W503,W504 ================================================ FILE: util/__init__.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved ================================================ FILE: util/box_ops.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Utilities for bounding box manipulation and GIoU. """ import torch from torchvision.ops.boxes import box_area def box_cxcywh_to_xyxy(x): x_c, y_c, w, h = x.unbind(-1) b = [(x_c - 0.5 * w), (y_c - 0.5 * h), (x_c + 0.5 * w), (y_c + 0.5 * h)] return torch.stack(b, dim=-1) def box_xyxy_to_cxcywh(x): x0, y0, x1, y1 = x.unbind(-1) b = [(x0 + x1) / 2, (y0 + y1) / 2, (x1 - x0), (y1 - y0)] return torch.stack(b, dim=-1) # modified from torchvision to also return the union def box_iou(boxes1, boxes2): area1 = box_area(boxes1) area2 = box_area(boxes2) lt = torch.max(boxes1[:, None, :2], boxes2[:, :2]) # [N,M,2] rb = torch.min(boxes1[:, None, 2:], boxes2[:, 2:]) # [N,M,2] wh = (rb - lt).clamp(min=0) # [N,M,2] inter = wh[:, :, 0] * wh[:, :, 1] # [N,M] union = area1[:, None] + area2 - inter iou = inter / union return iou, union def generalized_box_iou(boxes1, boxes2): """ Generalized IoU from https://giou.stanford.edu/ The boxes should be in [x0, y0, x1, y1] format Returns a [N, M] pairwise matrix, where N = len(boxes1) and M = len(boxes2) """ # degenerate boxes gives inf / nan results # so do an early check assert (boxes1[:, 2:] >= boxes1[:, :2]).all() assert (boxes2[:, 2:] >= boxes2[:, :2]).all() iou, union = box_iou(boxes1, boxes2) lt = torch.min(boxes1[:, None, :2], boxes2[:, :2]) rb = torch.max(boxes1[:, None, 2:], boxes2[:, 2:]) wh = (rb - lt).clamp(min=0) # [N,M,2] area = wh[:, :, 0] * wh[:, :, 1] return iou - (area - union) / area def masks_to_boxes(masks): """Compute the bounding boxes around the provided masks The masks should be in format [N, H, W] where N is the number of masks, (H, W) are the spatial dimensions. Returns a [N, 4] tensors, with the boxes in xyxy format """ if masks.numel() == 0: return torch.zeros((0, 4), device=masks.device) h, w = masks.shape[-2:] y = torch.arange(0, h, dtype=torch.float) x = torch.arange(0, w, dtype=torch.float) y, x = torch.meshgrid(y, x) x_mask = (masks * x.unsqueeze(0)) x_max = x_mask.flatten(1).max(-1)[0] x_min = x_mask.masked_fill(~(masks.bool()), 1e8).flatten(1).min(-1)[0] y_mask = (masks * y.unsqueeze(0)) y_max = y_mask.flatten(1).max(-1)[0] y_min = y_mask.masked_fill(~(masks.bool()), 1e8).flatten(1).min(-1)[0] return torch.stack([x_min, y_min, x_max, y_max], 1) ================================================ FILE: util/misc.py ================================================ # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved """ Misc functions, including distributed helpers. Mostly copy-paste from torchvision references. """ import os import subprocess import time from collections import defaultdict, deque import datetime import pickle from packaging import version from typing import Optional, List import torch import torch.distributed as dist from torch import Tensor # needed due to empty tensor bug in pytorch and torchvision 0.5 import torchvision if version.parse(torchvision.__version__) < version.parse('0.7'): from torchvision.ops import _new_empty_tensor from torchvision.ops.misc import _output_size class SmoothedValue(object): """Track a series of values and provide access to smoothed values over a window or the global series average. """ def __init__(self, window_size=20, fmt=None): if fmt is None: fmt = "{median:.4f} ({global_avg:.4f})" self.deque = deque(maxlen=window_size) self.total = 0.0 self.count = 0 self.fmt = fmt def update(self, value, n=1): self.deque.append(value) self.count += n self.total += value * n def synchronize_between_processes(self): """ Warning: does not synchronize the deque! """ if not is_dist_avail_and_initialized(): return t = torch.tensor([self.count, self.total], dtype=torch.float64, device='cuda') dist.barrier() dist.all_reduce(t) t = t.tolist() self.count = int(t[0]) self.total = t[1] @property def median(self): d = torch.tensor(list(self.deque)) return d.median().item() @property def avg(self): d = torch.tensor(list(self.deque), dtype=torch.float32) return d.mean().item() @property def global_avg(self): return self.total / self.count @property def max(self): return max(self.deque) @property def value(self): return self.deque[-1] def __str__(self): return self.fmt.format( median=self.median, avg=self.avg, global_avg=self.global_avg, max=self.max, value=self.value) def all_gather(data): """ Run all_gather on arbitrary picklable data (not necessarily tensors) Args: data: any picklable object Returns: list[data]: list of data gathered from each rank """ world_size = get_world_size() if world_size == 1: return [data] # serialized to a Tensor buffer = pickle.dumps(data) storage = torch.ByteStorage.from_buffer(buffer) tensor = torch.ByteTensor(storage).to("cuda") # obtain Tensor size of each rank local_size = torch.tensor([tensor.numel()], device="cuda") size_list = [torch.tensor([0], device="cuda") for _ in range(world_size)] dist.all_gather(size_list, local_size) size_list = [int(size.item()) for size in size_list] max_size = max(size_list) # receiving Tensor from all ranks # we pad the tensor because torch all_gather does not support # gathering tensors of different shapes tensor_list = [] for _ in size_list: tensor_list.append(torch.empty((max_size,), dtype=torch.uint8, device="cuda")) if local_size != max_size: padding = torch.empty(size=(max_size - local_size,), dtype=torch.uint8, device="cuda") tensor = torch.cat((tensor, padding), dim=0) dist.all_gather(tensor_list, tensor) data_list = [] for size, tensor in zip(size_list, tensor_list): buffer = tensor.cpu().numpy().tobytes()[:size] data_list.append(pickle.loads(buffer)) return data_list def reduce_dict(input_dict, average=True): """ Args: input_dict (dict): all the values will be reduced average (bool): whether to do average or sum Reduce the values in the dictionary from all processes so that all processes have the averaged results. Returns a dict with the same fields as input_dict, after reduction. """ world_size = get_world_size() if world_size < 2: return input_dict with torch.no_grad(): names = [] values = [] # sort the keys so that they are consistent across processes for k in sorted(input_dict.keys()): names.append(k) values.append(input_dict[k]) values = torch.stack(values, dim=0) dist.all_reduce(values) if average: values /= world_size reduced_dict = {k: v for k, v in zip(names, values)} return reduced_dict class MetricLogger(object): def __init__(self, delimiter="\t"): self.meters = defaultdict(SmoothedValue) self.delimiter = delimiter def update(self, **kwargs): for k, v in kwargs.items(): if isinstance(v, torch.Tensor): v = v.item() assert isinstance(v, (float, int)) self.meters[k].update(v) def __getattr__(self, attr): if attr in self.meters: return self.meters[attr] if attr in self.__dict__: return self.__dict__[attr] raise AttributeError("'{}' object has no attribute '{}'".format( type(self).__name__, attr)) def __str__(self): loss_str = [] for name, meter in self.meters.items(): loss_str.append( "{}: {}".format(name, str(meter)) ) return self.delimiter.join(loss_str) def synchronize_between_processes(self): for meter in self.meters.values(): meter.synchronize_between_processes() def add_meter(self, name, meter): self.meters[name] = meter def log_every(self, iterable, print_freq, header=None): i = 0 if not header: header = '' start_time = time.time() end = time.time() iter_time = SmoothedValue(fmt='{avg:.4f}') data_time = SmoothedValue(fmt='{avg:.4f}') space_fmt = ':' + str(len(str(len(iterable)))) + 'd' if torch.cuda.is_available(): log_msg = self.delimiter.join([ header, '[{0' + space_fmt + '}/{1}]', 'eta: {eta}', '{meters}', 'time: {time}', 'data: {data}', 'max mem: {memory:.0f}' ]) else: log_msg = self.delimiter.join([ header, '[{0' + space_fmt + '}/{1}]', 'eta: {eta}', '{meters}', 'time: {time}', 'data: {data}' ]) MB = 1024.0 * 1024.0 for obj in iterable: data_time.update(time.time() - end) yield obj iter_time.update(time.time() - end) if i % print_freq == 0 or i == len(iterable) - 1: eta_seconds = iter_time.global_avg * (len(iterable) - i) eta_string = str(datetime.timedelta(seconds=int(eta_seconds))) if torch.cuda.is_available(): print(log_msg.format( i, len(iterable), eta=eta_string, meters=str(self), time=str(iter_time), data=str(data_time), memory=torch.cuda.max_memory_allocated() / MB)) else: print(log_msg.format( i, len(iterable), eta=eta_string, meters=str(self), time=str(iter_time), data=str(data_time))) i += 1 end = time.time() total_time = time.time() - start_time total_time_str = str(datetime.timedelta(seconds=int(total_time))) print('{} Total time: {} ({:.4f} s / it)'.format( header, total_time_str, total_time / len(iterable))) def get_sha(): cwd = os.path.dirname(os.path.abspath(__file__)) def _run(command): return subprocess.check_output(command, cwd=cwd).decode('ascii').strip() sha = 'N/A' diff = "clean" branch = 'N/A' try: sha = _run(['git', 'rev-parse', 'HEAD']) subprocess.check_output(['git', 'diff'], cwd=cwd) diff = _run(['git', 'diff-index', 'HEAD']) diff = "has uncommited changes" if diff else "clean" branch = _run(['git', 'rev-parse', '--abbrev-ref', 'HEAD']) except Exception: pass message = f"sha: {sha}, status: {diff}, branch: {branch}" return message def collate_fn(batch): batch = list(zip(*batch)) batch[0] = nested_tensor_from_tensor_list(batch[0]) return tuple(batch) def _max_by_axis(the_list): # type: (List[List[int]]) -> List[int] maxes = the_list[0] for sublist in the_list[1:]: for index, item in enumerate(sublist): maxes[index] = max(maxes[index], item) return maxes class NestedTensor(object): def __init__(self, tensors, mask: Optional[Tensor]): self.tensors = tensors self.mask = mask def to(self, device): # type: (Device) -> NestedTensor # noqa cast_tensor = self.tensors.to(device) mask = self.mask if mask is not None: assert mask is not None cast_mask = mask.to(device) else: cast_mask = None return NestedTensor(cast_tensor, cast_mask) def decompose(self): return self.tensors, self.mask def __repr__(self): return str(self.tensors) def nested_tensor_from_tensor_list(tensor_list: List[Tensor]): # TODO make this more general if tensor_list[0].ndim == 3: if torchvision._is_tracing(): # nested_tensor_from_tensor_list() does not export well to ONNX # call _onnx_nested_tensor_from_tensor_list() instead return _onnx_nested_tensor_from_tensor_list(tensor_list) # TODO make it support different-sized images max_size = _max_by_axis([list(img.shape) for img in tensor_list]) # min_size = tuple(min(s) for s in zip(*[img.shape for img in tensor_list])) batch_shape = [len(tensor_list)] + max_size b, c, h, w = batch_shape dtype = tensor_list[0].dtype device = tensor_list[0].device tensor = torch.zeros(batch_shape, dtype=dtype, device=device) mask = torch.ones((b, h, w), dtype=torch.bool, device=device) for img, pad_img, m in zip(tensor_list, tensor, mask): pad_img[: img.shape[0], : img.shape[1], : img.shape[2]].copy_(img) m[: img.shape[1], :img.shape[2]] = False else: raise ValueError('not supported') return NestedTensor(tensor, mask) # _onnx_nested_tensor_from_tensor_list() is an implementation of # nested_tensor_from_tensor_list() that is supported by ONNX tracing. @torch.jit.unused def _onnx_nested_tensor_from_tensor_list(tensor_list: List[Tensor]) -> NestedTensor: max_size = [] for i in range(tensor_list[0].dim()): max_size_i = torch.max(torch.stack([img.shape[i] for img in tensor_list]).to(torch.float32)).to(torch.int64) max_size.append(max_size_i) max_size = tuple(max_size) # work around for # pad_img[: img.shape[0], : img.shape[1], : img.shape[2]].copy_(img) # m[: img.shape[1], :img.shape[2]] = False # which is not yet supported in onnx padded_imgs = [] padded_masks = [] for img in tensor_list: padding = [(s1 - s2) for s1, s2 in zip(max_size, tuple(img.shape))] padded_img = torch.nn.functional.pad(img, (0, padding[2], 0, padding[1], 0, padding[0])) padded_imgs.append(padded_img) m = torch.zeros_like(img[0], dtype=torch.int, device=img.device) padded_mask = torch.nn.functional.pad(m, (0, padding[2], 0, padding[1]), "constant", 1) padded_masks.append(padded_mask.to(torch.bool)) tensor = torch.stack(padded_imgs) mask = torch.stack(padded_masks) return NestedTensor(tensor, mask=mask) def setup_for_distributed(is_master): """ This function disables printing when not in master process """ import builtins as __builtin__ builtin_print = __builtin__.print def print(*args, **kwargs): force = kwargs.pop('force', False) if is_master or force: builtin_print(*args, **kwargs) __builtin__.print = print def is_dist_avail_and_initialized(): if not dist.is_available(): return False if not dist.is_initialized(): return False return True def get_world_size(): if not is_dist_avail_and_initialized(): return 1 return dist.get_world_size() def get_rank(): if not is_dist_avail_and_initialized(): return 0 return dist.get_rank() def is_main_process(): return get_rank() == 0 def save_on_master(*args, **kwargs): if is_main_process(): torch.save(*args, **kwargs) def init_distributed_mode(args): if 'RANK' in os.environ and 'WORLD_SIZE' in os.environ: args.rank = int(os.environ["RANK"]) args.world_size = int(os.environ['WORLD_SIZE']) args.gpu = int(os.environ['LOCAL_RANK']) elif 'SLURM_PROCID' in os.environ: args.rank = int(os.environ['SLURM_PROCID']) args.gpu = args.rank % torch.cuda.device_count() else: print('Not using distributed mode') args.distributed = False return args.distributed = True torch.cuda.set_device(args.gpu) args.dist_backend = 'nccl' print('| distributed init (rank {}): {}'.format( args.rank, args.dist_url), flush=True) torch.distributed.init_process_group(backend=args.dist_backend, init_method=args.dist_url, world_size=args.world_size, rank=args.rank) torch.distributed.barrier() setup_for_distributed(args.rank == 0) @torch.no_grad() def accuracy(output, target, topk=(1,)): """Computes the precision@k for the specified values of k""" if target.numel() == 0: return [torch.zeros([], device=output.device)] maxk = max(topk) batch_size = target.size(0) _, pred = output.topk(maxk, 1, True, True) pred = pred.t() correct = pred.eq(target.view(1, -1).expand_as(pred)) res = [] for k in topk: correct_k = correct[:k].view(-1).float().sum(0) res.append(correct_k.mul_(100.0 / batch_size)) return res def interpolate(input, size=None, scale_factor=None, mode="nearest", align_corners=None): # type: (Tensor, Optional[List[int]], Optional[float], str, Optional[bool]) -> Tensor """ Equivalent to nn.functional.interpolate, but with support for empty batch sizes. This will eventually be supported natively by PyTorch, and this class can go away. """ if version.parse(torchvision.__version__) < version.parse('0.7'): if input.numel() > 0: return torch.nn.functional.interpolate( input, size, scale_factor, mode, align_corners ) output_shape = _output_size(2, input, size, scale_factor) output_shape = list(input.shape[:-2]) + list(output_shape) return _new_empty_tensor(input, output_shape) else: return torchvision.ops.misc.interpolate(input, size, scale_factor, mode, align_corners) ================================================ FILE: util/plot_utils.py ================================================ """ Plotting utilities to visualize training logs. """ import torch import pandas as pd import numpy as np import seaborn as sns import matplotlib.pyplot as plt from pathlib import Path, PurePath def plot_logs(logs, fields=('class_error', 'loss_bbox_unscaled', 'mAP'), ewm_col=0, log_name='log.txt'): ''' Function to plot specific fields from training log(s). Plots both training and test results. :: Inputs - logs = list containing Path objects, each pointing to individual dir with a log file - fields = which results to plot from each log file - plots both training and test for each field. - ewm_col = optional, which column to use as the exponential weighted smoothing of the plots - log_name = optional, name of log file if different than default 'log.txt'. :: Outputs - matplotlib plots of results in fields, color coded for each log file. - solid lines are training results, dashed lines are test results. ''' func_name = "plot_utils.py::plot_logs" # verify logs is a list of Paths (list[Paths]) or single Pathlib object Path, # convert single Path to list to avoid 'not iterable' error if not isinstance(logs, list): if isinstance(logs, PurePath): logs = [logs] print(f"{func_name} info: logs param expects a list argument, converted to list[Path].") else: raise ValueError(f"{func_name} - invalid argument for logs parameter.\n \ Expect list[Path] or single Path obj, received {type(logs)}") # Quality checks - verify valid dir(s), that every item in list is Path object, and that log_name exists in each dir for i, dir in enumerate(logs): if not isinstance(dir, PurePath): raise ValueError(f"{func_name} - non-Path object in logs argument of {type(dir)}: \n{dir}") if not dir.exists(): raise ValueError(f"{func_name} - invalid directory in logs argument:\n{dir}") # verify log_name exists fn = Path(dir / log_name) if not fn.exists(): print(f"-> missing {log_name}. Have you gotten to Epoch 1 in training?") print(f"--> full path of missing log file: {fn}") return # load log file(s) and plot dfs = [pd.read_json(Path(p) / log_name, lines=True) for p in logs] fig, axs = plt.subplots(ncols=len(fields), figsize=(16, 5)) for df, color in zip(dfs, sns.color_palette(n_colors=len(logs))): for j, field in enumerate(fields): if field == 'mAP': coco_eval = pd.DataFrame( np.stack(df.test_coco_eval_bbox.dropna().values)[:, 1] ).ewm(com=ewm_col).mean() axs[j].plot(coco_eval, c=color) else: df.interpolate().ewm(com=ewm_col).mean().plot( y=[f'train_{field}', f'test_{field}'], ax=axs[j], color=[color] * 2, style=['-', '--'] ) for ax, field in zip(axs, fields): ax.legend([Path(p).name for p in logs]) ax.set_title(field) def plot_precision_recall(files, naming_scheme='iter'): if naming_scheme == 'exp_id': # name becomes exp_id names = [f.parts[-3] for f in files] elif naming_scheme == 'iter': names = [f.stem for f in files] else: raise ValueError(f'not supported {naming_scheme}') fig, axs = plt.subplots(ncols=2, figsize=(16, 5)) for f, color, name in zip(files, sns.color_palette("Blues", n_colors=len(files)), names): data = torch.load(f) # precision is n_iou, n_points, n_cat, n_area, max_det precision = data['precision'] recall = data['params'].recThrs scores = data['scores'] # take precision for all classes, all areas and 100 detections precision = precision[0, :, :, 0, -1].mean(1) scores = scores[0, :, :, 0, -1].mean(1) prec = precision.mean() rec = data['recall'][0, :, 0, -1].mean() print(f'{naming_scheme} {name}: mAP@50={prec * 100: 05.1f}, ' + f'score={scores.mean():0.3f}, ' + f'f1={2 * prec * rec / (prec + rec + 1e-8):0.3f}' ) axs[0].plot(recall, precision, c=color) axs[1].plot(recall, scores, c=color) axs[0].set_title('Precision / Recall') axs[0].legend(names) axs[1].set_title('Scores / Recall') axs[1].legend(names) return fig, axs