Repository: arimousa/DDAD Branch: main Commit: 4323a150dd63 Files: 16 Total size: 77.1 KB Directory structure: gitextract_ujrg7voc/ ├── LICENSE ├── README.md ├── anomaly_map.py ├── config.yaml ├── dataset.py ├── ddad.py ├── feature_extractor.py ├── loss.py ├── main.py ├── metrics.py ├── reconstruction.py ├── requirements.txt ├── resnet.py ├── train.py ├── unet.py └── visualize.py ================================================ FILE CONTENTS ================================================ ================================================ FILE: LICENSE ================================================ MIT License Copyright (c) 2023 Arian Mousakhan Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. ================================================ FILE: README.md ================================================ # Anomaly Detection with Conditioned Denoising Diffusion Models. Official implementation of [DDAD](https://arxiv.org/abs/2305.15956) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/anomaly-detection-with-conditioned-denoising/anomaly-detection-on-mvtec-ad)](https://paperswithcode.com/sota/anomaly-detection-on-mvtec-ad?p=anomaly-detection-with-conditioned-denoising) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/anomaly-detection-with-conditioned-denoising/anomaly-detection-on-visa)](https://paperswithcode.com/sota/anomaly-detection-on-visa?p=anomaly-detection-with-conditioned-denoising) ![Framework](images/DDAD_Framework.png) ## Requirements This repository is implemented and tested on Python 3.8 and PyTorch 2.1. To install requirements: ```setup pip install -r requirements.txt ``` ## Train and Evaluation of the Model You can download the model checkpoints directly from [Checkpoints](https://drive.google.com/drive/u/0/folders/1FF83llo3a-mN5pJN8-_mw0hL5eZqe9fC) To train the denoising UNet, run: ```train python main.py --train True ``` Modify the settings in the config.yaml file to train the model on different categories. For fine-tuning the feature extractor, use the following command: ```domain_adaptation python main.py --domain_adaptation True ``` To evaluate and test the model, run: ```detection python main.py --detection True ``` ## Dataset You can download [MVTec AD: MVTec Software](https://www.mvtec.com/company/research/datasets/mvtec-ad/) and [VisA](https://amazon-visual-anomaly.s3.us-west-2.amazonaws.com/VisA_20220922.tar) Benchmarks. For preprocessing of VisA dataset check out the [Data preparation](https://github.com/amazon-science/spot-diff/tree/main) section of this repository. The dataset should be placed in the 'datasets' folder. The training dataset should only contain one subcategory consisting of nominal samples, which should be named 'good'. The test dataset should include one category named 'good' for nominal samples, and any other subcategories of anomalous samples. It should be made as follows: ```shell Name_of_Dataset |-- Category |-----|----- ground_truth |-----|----- test |-----|--------|------ good |-----|--------|------ ... |-----|--------|------ ... |-----|----- train |-----|--------|------ good ``` ## Results Running the code as explained in this file should achieve the following results for MVTec AD: Anomaly Detection (Image AUROC) and Anomaly Localization (Pixel AUROC, PRO) Expected results for MVTec AD: | Category | Carpet | Grid | Leather | Tile | Wood | Bottle | Cable | Capsule | Hazel nut | Metalnut | Pill | Screw | Toothbrush | Transistor | Zipper |Average |---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---| | Detection | 99.3% | 100% | 100% | 100% | 100% | 100% | 99.4% | 99.4% | 100% | 100% | 100% | 99.0% | 100% | 100% | 100% | 99.8% | Localization | (98.7%,93.9%) | (99.4%,97.3%) | (99.4%,97.7%) | (98.2%,93.1%) | (95.0%,82.9%) | (98.7%,91.8%) | (98.1%,88.9%) | (95.7%,93.4%) | (98.4%,86.7%) | (99.0%,91.1%) | (99.1%,95.5%) | (99.3%,96.3%) | (98.7%,92.6%) | (95.3%,90.1%) | (98.2%,93.2%) | (98,1%,92.3%) The settings used for these results are detailed in the table. | **Categories** | Carpet | Grid | Leather | Tile | Wood | Bottle | Cable | Capsule | Hazelnut | Metal nut | Pill | Screw | Toothbrush | Transistor | Zipper | | -------------- | ------ | ---- | ------- | ---- | ---- | ------ | ----- | ------- | -------- | --------- | ---- | ----- | ----------- | ---------- | ------ | | **\(w\)** | 0 | 4 | 11 | 4 | 11 | 3 | 3 | 8 | 5 | 7 | 9 | 2 | 0 | 0 | 10 | | **Training epochs** | 2500 | 2000 | 2000 | 1000 | 2000 | 1000 | 3000 | 1500 | 2000 | 3000 | 1000 | 2000 | 2000 | 2000 | 1000 | | **FE epochs** | 0 | 6 | 8 | 0 | 16 | 5 | 0 | 8 | 3 | 1 | 4 | 4 | 2 | 0 | 6 | Following is the expected results on VisA Dataset. | Category | Candle | Capsules | Cashew | Chewing gum | Fryum | Macaroni1 | Macaroni2 | PCB1 | PCB2 | PCB3 | PCB4 | Pipe fryum | Average |---|---|---|---|---|---|---|---|---|---|---|---|---|---| | Detection | 99.9% | 100% | 94.5% | 98.1% | 99.0% | 99.2% | 99.2% | 100% | 99.7% | 97.2% | 100% | 100% | 98.9% | Localization | (98.7%,96.6%) | (99.5%,95.0%) | (97.4%,80.3%) | (96.5%,85.2%) | (96.9%,94.2%) | (98.7%,98.5%) | (98.2%,99.3%) | (93.4%,93.3%) | (97.4%,93.3%) | (96.3%,86.6%) | (98.5%,95.5%) | (99.5%,94.7%) |(97.6%,92.7%) The settings used for these results are detailed in the table. | **Categories** | Candle | Capsules | Cashew | Chewing gum | Fryum | Macaroni1 | Macaroni2 | PCB1 | PCB2 | PCB3 | PCB4 | Pipe fryum | | ---------------- | ------ | -------- | ------ | ------------ | ----- | --------- | --------- | ---- | ---- | ---- | ---- | ---------- | | **\(w\)** | 6 | 5 | 0 | 6 | 4 | 5 | 2 | 9 | 5 | 6 | 6 | 8 | | **Training epochs** | 1000 | 1000 | 1750 | 1250 | 1000 | 500 | 500 | 500 | 500 | 500 | 500 | 500 | | **FE epochs** | 1 | 3 | 0 | 0 | 3 | 7 | 11 | 8 | 5 | 1 | 1 | 6 | ![Framework](images/Qualitative.png) ## Citation ``` @article{mousakhan2023anomaly, title={Anomaly Detection with Conditioned Denoising Diffusion Models}, author={Mousakhan, Arian and Brox, Thomas and Tayyub, Jawad}, journal={arXiv preprint arXiv:2305.15956}, year={2023} } ``` ## Feedback For any feedback or inquiries, please contact arian.mousakhan@gmail.com ================================================ FILE: anomaly_map.py ================================================ import torch import torch.nn.functional as F from kornia.filters import gaussian_blur2d from torchvision.transforms import transforms import math from dataset import * from visualize import * from feature_extractor import * import numpy as np def heat_map(output, target, FE, config): ''' Compute the anomaly map :param output: the output of the reconstruction :param target: the target image :param FE: the feature extractor :param sigma: the sigma of the gaussian kernel :param i_d: the pixel distance :param f_d: the feature distance ''' sigma = 4 kernel_size = 2 * int(4 * sigma + 0.5) +1 anomaly_map = 0 output = output.to(config.model.device) target = target.to(config.model.device) i_d = pixel_distance(output, target) f_d = feature_distance((output), (target), FE, config) f_d = torch.Tensor(f_d).to(config.model.device) anomaly_map += f_d + config.model.v * (torch.max(f_d)/ torch.max(i_d)) * i_d anomaly_map = gaussian_blur2d( anomaly_map , kernel_size=(kernel_size,kernel_size), sigma=(sigma,sigma) ) anomaly_map = torch.sum(anomaly_map, dim=1).unsqueeze(1) return anomaly_map def pixel_distance(output, target): ''' Pixel distance between image1 and image2 ''' distance_map = torch.mean(torch.abs(output - target), dim=1).unsqueeze(1) return distance_map def feature_distance(output, target, FE, config): ''' Feature distance between output and target ''' FE.eval() transform = transforms.Compose([ transforms.Lambda(lambda t: (t + 1) / (2)), transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ]) target = transform(target) output = transform(output) inputs_features = FE(target) output_features = FE(output) out_size = config.data.image_size anomaly_map = torch.zeros([inputs_features[0].shape[0] ,1 ,out_size, out_size]).to(config.model.device) for i in range(len(inputs_features)): if i == 0: continue a_map = 1 - F.cosine_similarity(patchify(inputs_features[i]), patchify(output_features[i])) a_map = torch.unsqueeze(a_map, dim=1) a_map = F.interpolate(a_map, size=out_size, mode='bilinear', align_corners=True) anomaly_map += a_map return anomaly_map #https://github.com/amazon-science/patchcore-inspection def patchify(features, return_spatial_info=False): """Convert a tensor into a tensor of respective patches. Args: x: [torch.Tensor, bs x c x w x h] Returns: x: [torch.Tensor, bs * w//stride * h//stride, c, patchsize, patchsize] """ patchsize = 3 stride = 1 padding = int((patchsize - 1) / 2) unfolder = torch.nn.Unfold( kernel_size=patchsize, stride=stride, padding=padding, dilation=1 ) unfolded_features = unfolder(features) number_of_total_patches = [] for s in features.shape[-2:]: n_patches = ( s + 2 * padding - 1 * (patchsize - 1) - 1 ) / stride + 1 number_of_total_patches.append(int(n_patches)) unfolded_features = unfolded_features.reshape( *features.shape[:2], patchsize, patchsize, -1 ) unfolded_features = unfolded_features.permute(0, 4, 1, 2, 3) max_features = torch.mean(unfolded_features, dim=(3,4)) features = max_features.reshape(features.shape[0], int(math.sqrt(max_features.shape[1])) , int(math.sqrt(max_features.shape[1])), max_features.shape[-1]).permute(0,3,1,2) if return_spatial_info: return unfolded_features, number_of_total_patches return features ================================================ FILE: config.yaml ================================================ data : name: MVTec #MVTec #MTD #VisA data_dir: datasets/MVTec #MVTec #VisA #MTD category: screw #['carpet', 'bottle', 'hazelnut', 'leather', 'cable', 'capsule', 'grid', 'pill', 'transistor', 'metal_nut', 'screw','toothbrush', 'zipper', 'tile', 'wood'] # ['candle', 'capsules', 'cashew', 'chewinggum', 'fryum', 'macaroni1', 'macaroni2', 'pcb1', 'pcb2' ,'pcb3', 'pcb4', 'pipe_fryum'] image_size: 256 batch_size: 32 # 32 for DDAD and 16 for DDADS DA_batch_size: 16 #16 for MVTec and [macaroni2, pcb1] in VisA, and 32 for other categories in VisA test_batch_size: 16 #16 for MVTec, 32 for VisA mask : True input_channel : 3 model: DDADS: False checkpoint_dir: checkpoints/MVTec #MTD #MVTec #VisA checkpoint_name: weights exp_name: default feature_extractor: wide_resnet101_2 #wide_resnet101_2 # wide_resnet50_2 #resnet50 learning_rate: 3e-4 weight_decay: 0.05 epochs: 3000 load_chp : 2000 # From this epoch checkpoint will be loaded. Every 250 epochs a checkpoint is saved. Try to load 750 or 1000 epochs for Visa and 1000-1500-2000 for MVTec. DA_epochs: 4 # Number of epochs for Domain adaptation. DA_chp: 4 v : 1 #7 # 1 for MVTec and cashew in VisA, and 7 for VisA (1.5 for cashew). Control parameter for pixel-wise and feature-wise comparison. v * D_p + D_f w : 2 # Conditionig parameter. The higher the value, the more the model is conditioned on the target image. "Fine tuninig this parameter results in better performance". w_DA : 3 #3 # Conditionig parameter for domain adaptation. The higher the value, the more the model is conditioned on the target image. DLlambda : 0.1 # 0.1 for MVTec and 0.01 for VisA trajectory_steps: 1000 test_trajectoy_steps: 250 # Starting point for denoining trajectory. test_trajectoy_steps_DA: 250 # Starting point for denoining trajectory for domain adaptation. skip : 25 # Number of steps to skip for denoising trajectory. skip_DA : 25 eta : 1 # Stochasticity parameter for denoising process. beta_start : 0.0001 beta_end : 0.02 device: 'cuda' #<"cpu", "gpu", "tpu", "ipu"> save_model: True num_workers : 2 seed : 42 metrics: auroc: True pro: True misclassifications: False visualisation: False ================================================ FILE: dataset.py ================================================ import os from glob import glob from pathlib import Path import shutil import numpy as np import csv import torch import torch.utils.data from PIL import Image from torchvision import transforms import torch.nn.functional as F import torchvision.datasets as datasets from torchvision.datasets import CIFAR10 class Dataset_maker(torch.utils.data.Dataset): def __init__(self, root, category, config, is_train=True): self.image_transform = transforms.Compose( [ transforms.Resize((config.data.image_size, config.data.image_size)), transforms.ToTensor(), # Scales data into [0,1] transforms.Lambda(lambda t: (t * 2) - 1) # Scale between [-1, 1] ] ) self.config = config self.mask_transform = transforms.Compose( [ transforms.Resize((config.data.image_size, config.data.image_size)), transforms.ToTensor(), # Scales data into [0,1] ] ) if is_train: if category: self.image_files = glob( os.path.join(root, category, "train", "good", "*.png") ) else: self.image_files = glob( os.path.join(root, "train", "good", "*.png") ) else: if category: self.image_files = glob(os.path.join(root, category, "test", "*", "*.png")) else: self.image_files = glob(os.path.join(root, "test", "*", "*.png")) self.is_train = is_train def __getitem__(self, index): image_file = self.image_files[index] image = Image.open(image_file) image = self.image_transform(image) if(image.shape[0] == 1): image = image.expand(3, self.config.data.image_size, self.config.data.image_size) if self.is_train: label = 'good' return image, label else: if self.config.data.mask: if os.path.dirname(image_file).endswith("good"): target = torch.zeros([1, image.shape[-2], image.shape[-1]]) label = 'good' else : if self.config.data.name == 'MVTec': target = Image.open( image_file.replace("/test/", "/ground_truth/").replace( ".png", "_mask.png" ) ) else: target = Image.open( image_file.replace("/test/", "/ground_truth/")) target = self.mask_transform(target) label = 'defective' else: if os.path.dirname(image_file).endswith("good"): target = torch.zeros([1, image.shape[-2], image.shape[-1]]) label = 'good' else : target = torch.zeros([1, image.shape[-2], image.shape[-1]]) label = 'defective' return image, target, label def __len__(self): return len(self.image_files) ================================================ FILE: ddad.py ================================================ from asyncio import constants from typing import Any import torch from unet import * from dataset import * from visualize import * from anomaly_map import * from metrics import * from feature_extractor import * from reconstruction import * os.environ['CUDA_VISIBLE_DEVICES'] = "0,1,2" class DDAD: def __init__(self, unet, config) -> None: self.test_dataset = Dataset_maker( root= config.data.data_dir, category=config.data.category, config = config, is_train=False, ) self.testloader = torch.utils.data.DataLoader( self.test_dataset, batch_size= config.data.test_batch_size, shuffle=False, num_workers= config.model.num_workers, drop_last=False, ) self.unet = unet self.config = config self.reconstruction = Reconstruction(self.unet, self.config) self.transform = transforms.Compose([ transforms.CenterCrop((224)), ]) def __call__(self) -> Any: feature_extractor = domain_adaptation(self.unet, self.config, fine_tune=False) feature_extractor.eval() labels_list = [] predictions= [] anomaly_map_list = [] gt_list = [] reconstructed_list = [] forward_list = [] with torch.no_grad(): for input, gt, labels in self.testloader: input = input.to(self.config.model.device) x0 = self.reconstruction(input, input, self.config.model.w)[-1] anomaly_map = heat_map(x0, input, feature_extractor, self.config) anomaly_map = self.transform(anomaly_map) gt = self.transform(gt) forward_list.append(input) anomaly_map_list.append(anomaly_map) gt_list.append(gt) reconstructed_list.append(x0) for pred, label in zip(anomaly_map, labels): labels_list.append(0 if label == 'good' else 1) predictions.append(torch.max(pred).item()) metric = Metric(labels_list, predictions, anomaly_map_list, gt_list, self.config) metric.optimal_threshold() if self.config.metrics.auroc: print('AUROC: ({:.1f},{:.1f})'.format(metric.image_auroc() * 100, metric.pixel_auroc() * 100)) if self.config.metrics.pro: print('PRO: {:.1f}'.format(metric.pixel_pro() * 100)) if self.config.metrics.misclassifications: metric.miscalssified() reconstructed_list = torch.cat(reconstructed_list, dim=0) forward_list = torch.cat(forward_list, dim=0) anomaly_map_list = torch.cat(anomaly_map_list, dim=0) pred_mask = (anomaly_map_list > metric.threshold).float() gt_list = torch.cat(gt_list, dim=0) if not os.path.exists('results'): os.mkdir('results') if self.config.metrics.visualisation: visualize(forward_list, reconstructed_list, gt_list, pred_mask, anomaly_map_list, self.config.data.category) ================================================ FILE: feature_extractor.py ================================================ import logging import torch from dataset import * from dataset import * from unet import * from visualize import * from resnet import * import torchvision.transforms as T from reconstruction import * os.environ['CUDA_VISIBLE_DEVICES'] = "0,1,2" def loss_fucntion(a, b, c, d, config): cos_loss = torch.nn.CosineSimilarity() loss1 = 0 loss2 = 0 loss3 = 0 for item in range(len(a)): loss1 += torch.mean(1-cos_loss(a[item].view(a[item].shape[0],-1),b[item].view(b[item].shape[0],-1))) loss2 += torch.mean(1-cos_loss(b[item].view(b[item].shape[0],-1),c[item].view(c[item].shape[0],-1))) * config.model.DLlambda loss3 += torch.mean(1-cos_loss(a[item].view(a[item].shape[0],-1),d[item].view(d[item].shape[0],-1))) * config.model.DLlambda loss = loss1+loss2+loss3 return loss def domain_adaptation(unet, config, fine_tune): if config.model.feature_extractor == 'wide_resnet101_2': feature_extractor = wide_resnet101_2(pretrained=True) frozen_feature_extractor = wide_resnet101_2(pretrained=True) elif config.model.feature_extractor == 'wide_resnet50_2': feature_extractor = wide_resnet50_2(pretrained=True) frozen_feature_extractor = wide_resnet50_2(pretrained=True) elif config.model.feature_extractor == 'resnet50': feature_extractor = resnet50(pretrained=True) frozen_feature_extractor = resnet50(pretrained=True) else: logging.warning("Feature extractor is not correctly selected, Default: wide_resnet101_2") feature_extractor = wide_resnet101_2(pretrained=True) frozen_feature_extractor = wide_resnet101_2(pretrained=True) feature_extractor.to(config.model.device) frozen_feature_extractor.to(config.model.device) frozen_feature_extractor.eval() feature_extractor = torch.nn.DataParallel(feature_extractor) frozen_feature_extractor = torch.nn.DataParallel(frozen_feature_extractor) train_dataset = Dataset_maker( root= config.data.data_dir, category= config.data.category, config = config, is_train=True, ) trainloader = torch.utils.data.DataLoader( train_dataset, batch_size=config.data.DA_batch_size, shuffle=True, num_workers=config.model.num_workers, drop_last=True, ) if fine_tune: unet.eval() feature_extractor.train() transform = transforms.Compose([ transforms.Lambda(lambda t: (t + 1) / (2)), transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ]) optimizer = torch.optim.AdamW(feature_extractor.parameters(),lr= 1e-4) torch.save(frozen_feature_extractor.state_dict(), os.path.join(os.path.join(os.getcwd(), config.model.checkpoint_dir), config.data.category,f'feat0')) reconstruction = Reconstruction(unet, config) for epoch in range(config.model.DA_epochs): for step, batch in enumerate(trainloader): half_batch_size = batch[0].shape[0]//2 target = batch[0][:half_batch_size].to(config.model.device) input = batch[0][half_batch_size:].to(config.model.device) x0 = reconstruction(input, target, config.model.w_DA)[-1].to(config.model.device) x0 = transform(x0) target = transform(target) reconst_fe = feature_extractor(x0) target_fe = feature_extractor(target) target_frozen_fe = frozen_feature_extractor(target) reconst_frozen_fe = frozen_feature_extractor(x0) loss = loss_fucntion(reconst_fe, target_fe, target_frozen_fe,reconst_frozen_fe, config) optimizer.zero_grad() loss.backward() optimizer.step() print(f"Epoch {epoch+1} | Loss: {loss.item()}") # if (epoch+1) % 5 == 0: torch.save(feature_extractor.state_dict(), os.path.join(os.path.join(os.getcwd(), config.model.checkpoint_dir), config.data.category,f'feat{epoch+1}')) else: checkpoint = torch.load(os.path.join(os.path.join(os.getcwd(), config.model.checkpoint_dir), config.data.category,f'feat{config.model.DA_chp}'))#{config.model.DA_chp} feature_extractor.load_state_dict(checkpoint) return feature_extractor ================================================ FILE: loss.py ================================================ import torch import torch.nn as nn import numpy as np def get_loss(model, x_0, t, config): x_0 = x_0.to(config.model.device) betas = np.linspace(config.model.beta_start, config.model.beta_end, config.model.trajectory_steps, dtype=np.float64) b = torch.tensor(betas).type(torch.float).to(config.model.device) e = torch.randn_like(x_0, device = x_0.device) at = (1-b).cumprod(dim=0).index_select(0, t).view(-1, 1, 1, 1) x = at.sqrt() * x_0 + (1- at).sqrt() * e output = model(x, t.float()) return (e - output).square().sum(dim=(1, 2, 3)).mean(dim=0) ================================================ FILE: main.py ================================================ import torch import numpy as np import os import argparse from unet import * from omegaconf import OmegaConf from train import trainer from feature_extractor import * from ddad import * os.environ['CUDA_VISIBLE_DEVICES'] = "0,1,2" def build_model(config): if config.model.DDADS: unet = UNetModel(config.data.image_size, 32, dropout=0.3, n_heads=2 ,in_channels=config.data.input_channel) else: unet = UNetModel(config.data.image_size, 64, dropout=0.0, n_heads=4 ,in_channels=config.data.input_channel) return unet def train(config): torch.manual_seed(42) np.random.seed(42) unet = build_model(config) print(" Num params: ", sum(p.numel() for p in unet.parameters())) unet = unet.to(config.model.device) unet.train() unet = torch.nn.DataParallel(unet) # checkpoint = torch.load(os.path.join(os.path.join(os.getcwd(), config.model.checkpoint_dir), config.data.category,'1000')) # unet.load_state_dict(checkpoint) trainer(unet, config.data.category, config)#config.data.category, def detection(config): unet = build_model(config) checkpoint = torch.load(os.path.join(os.getcwd(), config.model.checkpoint_dir, config.data.category, str(config.model.load_chp))) unet = torch.nn.DataParallel(unet) unet.load_state_dict(checkpoint) unet.to(config.model.device) checkpoint = torch.load(os.path.join(os.getcwd(), config.model.checkpoint_dir, config.data.category, str(config.model.load_chp))) unet.eval() ddad = DDAD(unet, config) ddad() def finetuning(config): unet = build_model(config) checkpoint = torch.load(os.path.join(os.getcwd(), config.model.checkpoint_dir, config.data.category, str(config.model.load_chp))) unet = torch.nn.DataParallel(unet) unet.load_state_dict(checkpoint) unet.to(config.model.device) unet.eval() domain_adaptation(unet, config, fine_tune=True) def parse_args(): cmdline_parser = argparse.ArgumentParser('DDAD') cmdline_parser.add_argument('-cfg', '--config', default= os.path.join(os.path.dirname(os.path.abspath(__file__)),'config.yaml'), help='config file') cmdline_parser.add_argument('--train', default= False, help='Train the diffusion model') cmdline_parser.add_argument('--detection', default= False, help='Detection anomalies') cmdline_parser.add_argument('--domain_adaptation', default= False, help='Domain adaptation') args, unknowns = cmdline_parser.parse_known_args() return args if __name__ == "__main__": torch.cuda.empty_cache() args = parse_args() config = OmegaConf.load(args.config) print("Class: ",config.data.category, " w:", config.model.w, " v:", config.model.v, " load_chp:", config.model.load_chp, " feature extractor:", config.model.feature_extractor," w_DA: ",config.model.w_DA," DLlambda: ",config.model.DLlambda) print(f'{config.model.test_trajectoy_steps=} , {config.data.test_batch_size=}') torch.manual_seed(42) np.random.seed(42) if torch.cuda.is_available(): torch.cuda.manual_seed_all(42) if args.train: print('Training...') train(config) if args.domain_adaptation: print('Domain Adaptation...') finetuning(config) if args.detection: print('Detecting Anomalies...') detection(config) ================================================ FILE: metrics.py ================================================ import torch from torchmetrics import ROC, AUROC, F1Score import os from torchvision.transforms import transforms from skimage import measure import pandas as pd from statistics import mean import numpy as np from sklearn.metrics import auc from sklearn import metrics from sklearn.metrics import roc_auc_score, roc_curve class Metric: def __init__(self,labels_list, predictions, anomaly_map_list, gt_list, config) -> None: self.labels_list = labels_list self.predictions = predictions self.anomaly_map_list = anomaly_map_list self.gt_list = gt_list self.config = config self.threshold = 0.5 def image_auroc(self): auroc_image = roc_auc_score(self.labels_list, self.predictions) return auroc_image def pixel_auroc(self): resutls_embeddings = self.anomaly_map_list[0] for feature in self.anomaly_map_list[1:]: resutls_embeddings = torch.cat((resutls_embeddings, feature), 0) resutls_embeddings = ((resutls_embeddings - resutls_embeddings.min())/ (resutls_embeddings.max() - resutls_embeddings.min())) gt_embeddings = self.gt_list[0] for feature in self.gt_list[1:]: gt_embeddings = torch.cat((gt_embeddings, feature), 0) resutls_embeddings = resutls_embeddings.clone().detach().requires_grad_(False) gt_embeddings = gt_embeddings.clone().detach().requires_grad_(False) auroc_p = AUROC(task="binary") gt_embeddings = torch.flatten(gt_embeddings).type(torch.bool).cpu().detach() resutls_embeddings = torch.flatten(resutls_embeddings).cpu().detach() auroc_pixel = auroc_p(resutls_embeddings, gt_embeddings) return auroc_pixel def optimal_threshold(self): fpr, tpr, thresholds = roc_curve(self.labels_list, self.predictions) # Calculate Youden's J statistic for each threshold youden_j = tpr - fpr # Find the optimal threshold that maximizes Youden's J statistic optimal_threshold_index = np.argmax(youden_j) optimal_threshold = thresholds[optimal_threshold_index] self.threshold = optimal_threshold return optimal_threshold def pixel_pro(self): #https://github.com/hq-deng/RD4AD/blob/main/test.py#L337 def _compute_pro(masks, amaps, num_th = 200): resutls_embeddings = amaps[0] for feature in amaps[1:]: resutls_embeddings = torch.cat((resutls_embeddings, feature), 0) amaps = ((resutls_embeddings - resutls_embeddings.min())/ (resutls_embeddings.max() - resutls_embeddings.min())) amaps = amaps.squeeze(1) amaps = amaps.cpu().detach().numpy() gt_embeddings = masks[0] for feature in masks[1:]: gt_embeddings = torch.cat((gt_embeddings, feature), 0) masks = gt_embeddings.squeeze(1).cpu().detach().numpy() min_th = amaps.min() max_th = amaps.max() delta = (max_th - min_th) / num_th binary_amaps = np.zeros_like(amaps) df = pd.DataFrame([], columns=["pro", "fpr", "threshold"]) for th in np.arange(min_th, max_th, delta): binary_amaps[amaps <= th] = 0 binary_amaps[amaps > th] = 1 pros = [] for binary_amap, mask in zip(binary_amaps, masks): for region in measure.regionprops(measure.label(mask)): axes0_ids = region.coords[:, 0] axes1_ids = region.coords[:, 1] tp_pixels = binary_amap[axes0_ids, axes1_ids].sum() pros.append(tp_pixels / region.area) inverse_masks = 1 - masks fp_pixels = np.logical_and(inverse_masks , binary_amaps).sum() fpr = fp_pixels / inverse_masks.sum() # print(f"Threshold: {th}, FPR: {fpr}, PRO: {mean(pros)}") df = pd.concat([df, pd.DataFrame({"pro": mean(pros), "fpr": fpr, "threshold": th}, index=[0])], ignore_index=True) # df = df.concat({"pro": mean(pros), "fpr": fpr, "threshold": th}, ignore_index=True) # Normalize FPR from 0 ~ 1 to 0 ~ 0.3 df = df[df["fpr"] < 0.3] df["fpr"] = df["fpr"] / df["fpr"].max() pro_auc = auc(df["fpr"], df["pro"]) return pro_auc pro = _compute_pro(self.gt_list, self.anomaly_map_list, num_th = 200) return pro def miscalssified(self): predictions = torch.tensor(self.predictions) labels_list = torch.tensor(self.labels_list) predictions0_1 = (predictions > self.threshold).int() for i,(l,p) in enumerate(zip(labels_list, predictions0_1)): print('Sample : ', i, ' predicted as: ',p.item() ,' label is: ',l.item(),'\n' ) if l != p else None ================================================ FILE: reconstruction.py ================================================ from typing import Any import torch # from forward_process import * import numpy as np import os os.environ['CUDA_VISIBLE_DEVICES'] = "0,1,2" class Reconstruction: ''' The reconstruction process :param y: the target image :param x: the input image :param seq: the sequence of denoising steps :param unet: the UNet model :param x0_t: the prediction of x0 at time step t ''' def __init__(self, unet, config) -> None: self.unet = unet self.config = config def __call__(self, x, y0, w) -> Any: def _compute_alpha(t): betas = np.linspace(self.config.model.beta_start, self.config.model.beta_end, self.config.model.trajectory_steps, dtype=np.float64) betas = torch.tensor(betas).type(torch.float).to(self.config.model.device) beta = torch.cat([torch.zeros(1).to(self.config.model.device), betas], dim=0) beta = beta.to(self.config.model.device) a = (1 - beta).cumprod(dim=0).index_select(0, t + 1).view(-1, 1, 1, 1) return a test_trajectoy_steps = torch.Tensor([self.config.model.test_trajectoy_steps]).type(torch.int64).to(self.config.model.device).long() at = _compute_alpha(test_trajectoy_steps) xt = at.sqrt() * x + (1- at).sqrt() * torch.randn_like(x).to(self.config.model.device) seq = range(0 , self.config.model.test_trajectoy_steps, self.config.model.skip) with torch.no_grad(): n = x.size(0) seq_next = [-1] + list(seq[:-1]) xs = [xt] for index, (i, j) in enumerate(zip(reversed(seq), reversed(seq_next))): t = (torch.ones(n) * i).to(self.config.model.device) next_t = (torch.ones(n) * j).to(self.config.model.device) at = _compute_alpha(t.long()) at_next = _compute_alpha(next_t.long()) xt = xs[-1].to(self.config.model.device) self.unet = self.unet.to(self.config.model.device) et = self.unet(xt, t) yt = at.sqrt() * y0 + (1- at).sqrt() * et et_hat = et - (1 - at).sqrt() * w * (yt-xt) x0_t = (xt - et_hat * (1 - at).sqrt()) / at.sqrt() c1 = ( self.config.model.eta * ((1 - at / at_next) * (1 - at_next) / (1 - at)).sqrt() ) c2 = ((1 - at_next) - c1 ** 2).sqrt() xt_next = at_next.sqrt() * x0_t + c1 * torch.randn_like(x) + c2 * et_hat xs.append(xt_next) return xs ================================================ FILE: requirements.txt ================================================ kornia==0.6.12 matplotlib==3.7.1 numpy==1.24.3 omegaconf==2.1.2 opencv-python-headless==4.5.5.64 pandas==2.0.1 Pillow==9.5.0 scikit-image==0.19.2 scikit-learn==1.2.2 scipy==1.10.1 sklearn==0.0.post5 torch==2.0.1 torchmetrics==0.11.4 torchvision==0.15.2 ================================================ FILE: resnet.py ================================================ import torch from torch import Tensor import torch.nn as nn try: from torch.hub import load_state_dict_from_url except ImportError: from torch.utils.model_zoo import load_url as load_state_dict_from_url from typing import Type, Any, Callable, Union, List, Optional __all__ = ['ResNet', 'resnet18', 'resnet34', 'resnet50', 'resnet101', 'resnet152', 'resnext50_32x4d', 'resnext101_32x8d', 'wide_resnet50_2', 'wide_resnet101_2'] model_urls = { 'resnet18': 'https://download.pytorch.org/models/resnet18-f37072fd.pth', 'resnet34': 'https://download.pytorch.org/models/resnet34-b627a593.pth', 'resnet50': 'https://download.pytorch.org/models/resnet50-0676ba61.pth', 'resnet101': 'https://download.pytorch.org/models/resnet101-63fe2227.pth', 'resnet152': 'https://download.pytorch.org/models/resnet152-394f9c45.pth', 'resnext50_32x4d': 'https://download.pytorch.org/models/resnext50_32x4d-7cdf4587.pth', 'resnext101_32x8d': 'https://download.pytorch.org/models/resnext101_32x8d-8ba56ff5.pth', 'wide_resnet50_2': 'https://download.pytorch.org/models/wide_resnet50_2-95faca4d.pth', 'wide_resnet101_2': 'https://download.pytorch.org/models/wide_resnet101_2-32ee1156.pth', } def conv3x3(in_planes: int, out_planes: int, stride: int = 1, groups: int = 1, dilation: int = 1) -> nn.Conv2d: """3x3 convolution with padding""" return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=dilation, groups=groups, bias=False, dilation=dilation) def conv1x1(in_planes: int, out_planes: int, stride: int = 1) -> nn.Conv2d: """1x1 convolution""" return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False) class BasicBlock(nn.Module): expansion: int = 1 def __init__( self, inplanes: int, planes: int, stride: int = 1, downsample: Optional[nn.Module] = None, groups: int = 1, base_width: int = 64, dilation: int = 1, norm_layer: Optional[Callable[..., nn.Module]] = None ) -> None: super(BasicBlock, self).__init__() if norm_layer is None: norm_layer = nn.BatchNorm2d if groups != 1 or base_width != 64: raise ValueError('BasicBlock only supports groups=1 and base_width=64') if dilation > 1: raise NotImplementedError("Dilation > 1 not supported in BasicBlock") # Both self.conv1 and self.downsample layers downsample the input when stride != 1 self.conv1 = conv3x3(inplanes, planes, stride) self.bn1 = norm_layer(planes) self.relu = nn.ReLU(inplace=True) self.conv2 = conv3x3(planes, planes) self.bn2 = norm_layer(planes) self.downsample = downsample self.stride = stride def forward(self, x: Tensor) -> Tensor: identity = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) if self.downsample is not None: identity = self.downsample(x) out += identity out = self.relu(out) return out class Bottleneck(nn.Module): # Bottleneck in torchvision places the stride for downsampling at 3x3 convolution(self.conv2) # while original implementation places the stride at the first 1x1 convolution(self.conv1) # according to "Deep residual learning for image recognition"https://arxiv.org/abs/1512.03385. # This variant is also known as ResNet V1.5 and improves accuracy according to # https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch. expansion: int = 4 def __init__( self, inplanes: int, planes: int, stride: int = 1, downsample: Optional[nn.Module] = None, groups: int = 1, base_width: int = 64, dilation: int = 1, norm_layer: Optional[Callable[..., nn.Module]] = None ) -> None: super(Bottleneck, self).__init__() if norm_layer is None: norm_layer = nn.BatchNorm2d width = int(planes * (base_width / 64.)) * groups # Both self.conv2 and self.downsample layers downsample the input when stride != 1 self.conv1 = conv1x1(inplanes, width) self.bn1 = norm_layer(width) self.conv2 = conv3x3(width, width, stride, groups, dilation) self.bn2 = norm_layer(width) self.conv3 = conv1x1(width, planes * self.expansion) self.bn3 = norm_layer(planes * self.expansion) self.relu = nn.ReLU(inplace=True) self.downsample = downsample self.stride = stride def forward(self, x: Tensor) -> Tensor: identity = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.relu(out) out = self.conv3(out) out = self.bn3(out) if self.downsample is not None: identity = self.downsample(x) out += identity out = self.relu(out) return out class ResNet(nn.Module): def __init__( self, block: Type[Union[BasicBlock, Bottleneck]], layers: List[int], num_classes: int = 1000, zero_init_residual: bool = False, groups: int = 1, width_per_group: int = 64, replace_stride_with_dilation: Optional[List[bool]] = None, norm_layer: Optional[Callable[..., nn.Module]] = None ) -> None: super(ResNet, self).__init__() if norm_layer is None: norm_layer = nn.BatchNorm2d self._norm_layer = norm_layer self.inplanes = 64 self.dilation = 1 if replace_stride_with_dilation is None: # each element in the tuple indicates if we should replace # the 2x2 stride with a dilated convolution instead replace_stride_with_dilation = [False, False, False] if len(replace_stride_with_dilation) != 3: raise ValueError("replace_stride_with_dilation should be None " "or a 3-element tuple, got {}".format(replace_stride_with_dilation)) self.groups = groups self.base_width = width_per_group self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = norm_layer(self.inplanes) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, layers[0]) self.layer2 = self._make_layer(block, 128, layers[1], stride=2, dilate=replace_stride_with_dilation[0]) self.layer3 = self._make_layer(block, 256, layers[2], stride=2, dilate=replace_stride_with_dilation[1]) self.layer4 = self._make_layer(block, 512, layers[3], stride=2, dilate=replace_stride_with_dilation[2]) self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) self.fc = nn.Linear(512 * block.expansion, num_classes) for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu') elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)): nn.init.constant_(m.weight, 1) nn.init.constant_(m.bias, 0) # Zero-initialize the last BN in each residual branch, # so that the residual branch starts with zeros, and each residual block behaves like an identity. # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677 if zero_init_residual: for m in self.modules(): if isinstance(m, Bottleneck): nn.init.constant_(m.bn3.weight, 0) # type: ignore[arg-type] elif isinstance(m, BasicBlock): nn.init.constant_(m.bn2.weight, 0) # type: ignore[arg-type] def _make_layer(self, block: Type[Union[BasicBlock, Bottleneck]], planes: int, blocks: int, stride: int = 1, dilate: bool = False) -> nn.Sequential: norm_layer = self._norm_layer downsample = None previous_dilation = self.dilation if dilate: self.dilation *= stride stride = 1 if stride != 1 or self.inplanes != planes * block.expansion: downsample = nn.Sequential( conv1x1(self.inplanes, planes * block.expansion, stride), norm_layer(planes * block.expansion), ) layers = [] layers.append(block(self.inplanes, planes, stride, downsample, self.groups, self.base_width, previous_dilation, norm_layer)) self.inplanes = planes * block.expansion for _ in range(1, blocks): layers.append(block(self.inplanes, planes, groups=self.groups, base_width=self.base_width, dilation=self.dilation, norm_layer=norm_layer)) return nn.Sequential(*layers) def _forward_impl(self, x: Tensor) -> Tensor: # See note [TorchScript super()] x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x = self.maxpool(x) feature_a = self.layer1(x) feature_b = self.layer2(feature_a) feature_c = self.layer3(feature_b) feature_d = self.layer4(feature_c) return [feature_a, feature_b, feature_c] def forward(self, x: Tensor) -> Tensor: return self._forward_impl(x) def _resnet( arch: str, block: Type[Union[BasicBlock, Bottleneck]], layers: List[int], pretrained: bool, progress: bool, **kwargs: Any ) -> ResNet: model = ResNet(block, layers, **kwargs) if pretrained: state_dict = load_state_dict_from_url(model_urls[arch], progress=progress) #for k,v in list(state_dict.items()): # if 'layer4' in k or 'fc' in k: # state_dict.pop(k) model.load_state_dict(state_dict) return model class AttnBasicBlock(nn.Module): expansion: int = 1 def __init__( self, inplanes: int, planes: int, stride: int = 1, downsample: Optional[nn.Module] = None, groups: int = 1, base_width: int = 64, dilation: int = 1, norm_layer: Optional[Callable[..., nn.Module]] = None, attention: bool = True, ) -> None: super(AttnBasicBlock, self).__init__() self.attention = attention #print("Attention:", self.attention) if norm_layer is None: norm_layer = nn.BatchNorm2d if groups != 1 or base_width != 64: raise ValueError('BasicBlock only supports groups=1 and base_width=64') if dilation > 1: raise NotImplementedError("Dilation > 1 not supported in BasicBlock") # Both self.conv1 and self.downsample layers downsample the input when stride != 1 self.conv1 = conv3x3(inplanes, planes, stride) self.bn1 = norm_layer(planes) self.relu = nn.ReLU(inplace=True) self.conv2 = conv3x3(planes, planes) self.bn2 = norm_layer(planes) #self.cbam = GLEAM(planes, 16) self.downsample = downsample self.stride = stride def forward(self, x: Tensor) -> Tensor: #if self.attention: # x = self.cbam(x) identity = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) if self.downsample is not None: identity = self.downsample(x) out += identity out = self.relu(out) return out class AttnBottleneck(nn.Module): expansion: int = 4 def __init__( self, inplanes: int, planes: int, stride: int = 1, downsample: Optional[nn.Module] = None, groups: int = 1, base_width: int = 64, dilation: int = 1, norm_layer: Optional[Callable[..., nn.Module]] = None, attention: bool = True, ) -> None: super(AttnBottleneck, self).__init__() self.attention = attention #print("Attention:",self.attention) if norm_layer is None: norm_layer = nn.BatchNorm2d width = int(planes * (base_width / 64.)) * groups # Both self.conv2 and self.downsample layers downsample the input when stride != 1 self.conv1 = conv1x1(inplanes, width) self.bn1 = norm_layer(width) self.conv2 = conv3x3(width, width, stride, groups, dilation) self.bn2 = norm_layer(width) self.conv3 = conv1x1(width, planes * self.expansion) self.bn3 = norm_layer(planes * self.expansion) self.relu = nn.ReLU(inplace=True) #self.cbam = GLEAM([int(planes * self.expansion/4), # int(planes * self.expansion//2), # planes * self.expansion], 16) self.downsample = downsample self.stride = stride def forward(self, x: Tensor) -> Tensor: #if self.attention: # x = self.cbam(x) identity = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.relu(out) out = self.conv3(out) out = self.bn3(out) if self.downsample is not None: identity = self.downsample(x) out += identity out = self.relu(out) return out def __init__(self, block: Type[Union[BasicBlock, Bottleneck]], layers: int, groups: int = 1, width_per_group: int = 64, norm_layer: Optional[Callable[..., nn.Module]] = None, ): super(BN_layer, self).__init__() if norm_layer is None: norm_layer = nn.BatchNorm2d self._norm_layer = norm_layer self.groups = groups self.base_width = width_per_group self.inplanes = 256 * block.expansion self.dilation = 1 self.bn_layer = self._make_layer(block, 512, layers, stride=2) self.conv1 = conv3x3(64 * block.expansion, 128 * block.expansion, 2) self.bn1 = norm_layer(128 * block.expansion) self.relu = nn.ReLU(inplace=True) self.conv2 = conv3x3(128 * block.expansion, 256 * block.expansion, 2) self.bn2 = norm_layer(256 * block.expansion) self.conv3 = conv3x3(128 * block.expansion, 256 * block.expansion, 2) self.bn3 = norm_layer(256 * block.expansion) self.conv4 = conv1x1(1024 * block.expansion, 512 * block.expansion, 1) self.bn4 = norm_layer(512 * block.expansion) for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu') elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)): nn.init.constant_(m.weight, 1) nn.init.constant_(m.bias, 0) def _make_layer(self, block: Type[Union[BasicBlock, Bottleneck]], planes: int, blocks: int, stride: int = 1, dilate: bool = False) -> nn.Sequential: norm_layer = self._norm_layer downsample = None previous_dilation = self.dilation if dilate: self.dilation *= stride stride = 1 if stride != 1 or self.inplanes != planes * block.expansion: downsample = nn.Sequential( conv1x1(self.inplanes*3, planes * block.expansion, stride), norm_layer(planes * block.expansion), ) layers = [] layers.append(block(self.inplanes*3, planes, stride, downsample, self.groups, self.base_width, previous_dilation, norm_layer)) self.inplanes = planes * block.expansion for _ in range(1, blocks): layers.append(block(self.inplanes, planes, groups=self.groups, base_width=self.base_width, dilation=self.dilation, norm_layer=norm_layer)) return nn.Sequential(*layers) def _forward_impl(self, x: Tensor) -> Tensor: # See note [TorchScript super()] #x = self.cbam(x) l1 = self.relu(self.bn2(self.conv2(self.relu(self.bn1(self.conv1(x[0])))))) l2 = self.relu(self.bn3(self.conv3(x[1]))) feature = torch.cat([l1,l2,x[2]],1) output = self.bn_layer(feature) #x = self.avgpool(feature_d) #x = torch.flatten(x, 1) #x = self.fc(x) return output.contiguous() def forward(self, x: Tensor) -> Tensor: return self._forward_impl(x) def resnet18(pretrained: bool = False, progress: bool = True,**kwargs: Any) -> ResNet: r"""ResNet-18 model from `"Deep Residual Learning for Image Recognition" `_. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ return _resnet('resnet18', BasicBlock, [2, 2, 2, 2], pretrained, progress, **kwargs) def resnet34(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""ResNet-34 model from `"Deep Residual Learning for Image Recognition" `_. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ return _resnet('resnet34', BasicBlock, [3, 4, 6, 3], pretrained, progress, **kwargs) def resnet50(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""ResNet-50 model from `"Deep Residual Learning for Image Recognition" `_. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ return _resnet('resnet50', Bottleneck, [3, 4, 6, 3], pretrained, progress, **kwargs) def resnet101(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""ResNet-101 model from `"Deep Residual Learning for Image Recognition" `_. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ return _resnet('resnet101', Bottleneck, [3, 4, 23, 3], pretrained, progress, **kwargs) def resnet152(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""ResNet-152 model from `"Deep Residual Learning for Image Recognition" `_. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ return _resnet('resnet152', Bottleneck, [3, 8, 36, 3], pretrained, progress, **kwargs) def resnext50_32x4d(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""ResNeXt-50 32x4d model from `"Aggregated Residual Transformation for Deep Neural Networks" `_. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ kwargs['groups'] = 32 kwargs['width_per_group'] = 4 return _resnet('resnext50_32x4d', Bottleneck, [3, 4, 6, 3], pretrained, progress, **kwargs) def resnext101_32x8d(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""ResNeXt-101 32x8d model from `"Aggregated Residual Transformation for Deep Neural Networks" `_. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ kwargs['groups'] = 32 kwargs['width_per_group'] = 8 return _resnet('resnext101_32x8d', Bottleneck, [3, 4, 23, 3], pretrained, progress, **kwargs) def wide_resnet50_2(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""Wide ResNet-50-2 model from `"Wide Residual Networks" `_. The model is the same as ResNet except for the bottleneck number of channels which is twice larger in every block. The number of channels in outer 1x1 convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048 channels, and in Wide ResNet-50-2 has 2048-1024-2048. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ kwargs['width_per_group'] = 64 * 2 return _resnet('wide_resnet50_2', Bottleneck, [3, 4, 6, 3], pretrained, progress, **kwargs) def wide_resnet101_2(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> ResNet: r"""Wide ResNet-101-2 model from `"Wide Residual Networks" `_. The model is the same as ResNet except for the bottleneck number of channels which is twice larger in every block. The number of channels in outer 1x1 convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048 channels, and in Wide ResNet-50-2 has 2048-1024-2048. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ kwargs['width_per_group'] = 64 * 2 return _resnet('wide_resnet101_2', Bottleneck, [3, 4, 23, 3], pretrained, progress, **kwargs) ================================================ FILE: train.py ================================================ import torch import os import torch.nn as nn from dataset import * from dataset import * from loss import * def trainer(model, category, config): ''' Training the UNet model :param model: the UNet model :param category: the category of the dataset ''' # optimizer = torch.optim.AdamW( # model.parameters(), lr=config.model.learning_rate) optimizer = torch.optim.Adam( model.parameters(), lr=config.model.learning_rate, weight_decay=config.model.weight_decay ) train_dataset = Dataset_maker( root= config.data.data_dir, category=category, config = config, is_train=True, ) trainloader = torch.utils.data.DataLoader( train_dataset, batch_size=config.data.batch_size, shuffle=True, num_workers=config.model.num_workers, drop_last=True, ) if not os.path.exists('checkpoints'): os.mkdir('checkpoints') if not os.path.exists(config.model.checkpoint_dir): os.mkdir(config.model.checkpoint_dir) for epoch in range(config.model.epochs): for step, batch in enumerate(trainloader): optimizer.zero_grad() # loss = 0 # for _ in range(2): t = torch.randint(0, config.model.trajectory_steps, (batch[0].shape[0],), device=config.model.device).long() loss = get_loss(model, batch[0], t, config) loss.backward() optimizer.step() if (epoch+1) % 25 == 0 and step == 0: print(f"Epoch {epoch+1} | Loss: {loss.item()}") if (epoch+1) %250 == 0 and epoch>0 and step ==0: if config.model.save_model: model_save_dir = os.path.join(os.getcwd(), config.model.checkpoint_dir, category) if not os.path.exists(model_save_dir): os.mkdir(model_save_dir) torch.save(model.state_dict(), os.path.join(model_save_dir, str(epoch+1))) ================================================ FILE: unet.py ================================================ # https://github.com/openai/guided-diffusion/tree/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924 import math from abc import abstractmethod import numpy as np import torch import torch.nn.functional as F from torch import nn class TimestepBlock(nn.Module): """ Any module where forward() takes timestep embeddings as a second argument. """ @abstractmethod def forward(self, x, emb): """ Apply the module to `x` given `emb` timestep embeddings. """ class TimestepEmbedSequential(nn.Sequential, TimestepBlock): """ A sequential module that passes timestep embeddings to the children that support it as an extra input. """ def forward(self, x, emb): for layer in self: if isinstance(layer, TimestepBlock): x = layer(x, emb) else: x = layer(x) return x class PositionalEmbedding(nn.Module): # PositionalEmbedding """ Computes Positional Embedding of the timestep """ def __init__(self, dim, scale=1): super().__init__() assert dim % 2 == 0 self.dim = dim self.scale = scale def forward(self, x): device = x.device half_dim = self.dim // 2 emb = np.log(10000) / half_dim emb = torch.exp(torch.arange(half_dim, device=device) * -emb) emb = torch.outer(x * self.scale, emb) emb = torch.cat((emb.sin(), emb.cos()), dim=-1) return emb class Downsample(nn.Module): def __init__(self, in_channels, use_conv, out_channels=None): super().__init__() self.channels = in_channels out_channels = out_channels or in_channels if use_conv: # downsamples by 1/2 self.downsample = nn.Conv2d(in_channels, out_channels, 3, stride=2, padding=1) else: assert in_channels == out_channels self.downsample = nn.AvgPool2d(kernel_size=2, stride=2) def forward(self, x, time_embed=None): assert x.shape[1] == self.channels return self.downsample(x) class Upsample(nn.Module): def __init__(self, in_channels, use_conv, out_channels=None): super().__init__() self.channels = in_channels self.use_conv = use_conv # uses upsample then conv to avoid checkerboard artifacts # self.upsample = nn.Upsample(scale_factor=2, mode="nearest") if use_conv: self.conv = nn.Conv2d(in_channels, out_channels, 3, padding=1) def forward(self, x, time_embed=None): assert x.shape[1] == self.channels x = F.interpolate(x, scale_factor=2, mode="nearest") if self.use_conv: x = self.conv(x) return x class AttentionBlock(nn.Module): """ An attention block that allows spatial positions to attend to each other. Originally ported from here, but adapted to the N-d case. https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66. """ def __init__(self, in_channels, n_heads=1, n_head_channels=-1): super().__init__() self.in_channels = in_channels self.norm = GroupNorm32(32, self.in_channels) if n_head_channels == -1: self.num_heads = n_heads else: assert ( in_channels % n_head_channels == 0 ), f"q,k,v channels {in_channels} is not divisible by num_head_channels {n_head_channels}" self.num_heads = in_channels // n_head_channels # query, key, value for attention self.to_qkv = nn.Conv1d(in_channels, in_channels * 3, 1) self.attention = QKVAttention(self.num_heads) self.proj_out = zero_module(nn.Conv1d(in_channels, in_channels, 1)) def forward(self, x, time=None): b, c, *spatial = x.shape x = x.reshape(b, c, -1) qkv = self.to_qkv(self.norm(x)) h = self.attention(qkv) h = self.proj_out(h) return (x + h).reshape(b, c, *spatial) class QKVAttention(nn.Module): """ A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping """ def __init__(self, n_heads): super().__init__() self.n_heads = n_heads def forward(self, qkv, time=None): """ Apply QKV attention. :param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs. :return: an [N x (H * C) x T] tensor after attention. """ bs, width, length = qkv.shape assert width % (3 * self.n_heads) == 0 ch = width // (3 * self.n_heads) q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1) scale = 1 / math.sqrt(math.sqrt(ch)) weight = torch.einsum( "bct,bcs->bts", q * scale, k * scale ) # More stable with f16 than dividing afterwards weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype) a = torch.einsum("bts,bcs->bct", weight, v) return a.reshape(bs, -1, length) class ResBlock(TimestepBlock): def __init__( self, in_channels, time_embed_dim, dropout, out_channels=None, use_conv=False, up=False, down=False ): super().__init__() out_channels = out_channels or in_channels self.in_layers = nn.Sequential( GroupNorm32(32, in_channels), nn.SiLU(), nn.Conv2d(in_channels, out_channels, 3, padding=1) ) self.updown = up or down if up: self.h_upd = Upsample(in_channels, False) self.x_upd = Upsample(in_channels, False) elif down: self.h_upd = Downsample(in_channels, False) self.x_upd = Downsample(in_channels, False) else: self.h_upd = self.x_upd = nn.Identity() self.embed_layers = nn.Sequential( nn.SiLU(), nn.Linear(time_embed_dim, out_channels) ) self.out_layers = nn.Sequential( GroupNorm32(32, out_channels), nn.SiLU(), nn.Dropout(p=dropout), zero_module(nn.Conv2d(out_channels, out_channels, 3, padding=1)) ) if out_channels == in_channels: self.skip_connection = nn.Identity() elif use_conv: self.skip_connection = nn.Conv2d(in_channels, out_channels, 3, padding=1) else: self.skip_connection = nn.Conv2d(in_channels, out_channels, 1) def forward(self, x, time_embed): if self.updown: in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1] h = in_rest(x) h = self.h_upd(h) x = self.x_upd(x) h = in_conv(h) else: h = self.in_layers(x) emb_out = self.embed_layers(time_embed).type(h.dtype) while len(emb_out.shape) < len(h.shape): emb_out = emb_out[..., None] h = h + emb_out h = self.out_layers(h) return self.skip_connection(x) + h class UNetModel(nn.Module): # UNet model def __init__( self, img_size, base_channels, conv_resample=True, n_heads=1, n_head_channels=-1, channel_mults="", num_res_blocks=2, dropout=0, attention_resolutions="32,16,8", biggan_updown=True, in_channels=1 ): self.dtype = torch.float32 super().__init__() if channel_mults == "": if img_size == 512: channel_mults = (0.5, 1, 1, 2, 2, 4, 4) elif img_size == 256: channel_mults = (1, 1, 2, 2, 4, 4)# elif img_size == 128: channel_mults = (1, 1, 2, 3, 4) elif img_size == 64: channel_mults = (1, 2, 3, 4) elif img_size == 32: channel_mults = (1, 2, 3, 4) else: raise ValueError(f"unsupported image size: {img_size}") attention_ds = [] for res in attention_resolutions.split(","): attention_ds.append(img_size // int(res)) self.image_size = img_size self.in_channels = in_channels self.model_channels = base_channels self.out_channels = in_channels self.num_res_blocks = num_res_blocks self.attention_resolutions = attention_resolutions self.dropout = dropout self.channel_mult = channel_mults self.conv_resample = conv_resample self.dtype = torch.float32 self.num_heads = n_heads self.num_head_channels = n_head_channels time_embed_dim = base_channels * 4 self.time_embedding = nn.Sequential( PositionalEmbedding(base_channels, 1), nn.Linear(base_channels, time_embed_dim), nn.SiLU(), nn.Linear(time_embed_dim, time_embed_dim), ) ch = int(channel_mults[0] * base_channels) self.down = nn.ModuleList( [TimestepEmbedSequential(nn.Conv2d(self.in_channels, base_channels, 3, padding=1))] ) channels = [ch] ds = 1 for i, mult in enumerate(channel_mults): # out_channels = base_channels * mult for _ in range(num_res_blocks): layers = [ResBlock( ch, time_embed_dim=time_embed_dim, out_channels=base_channels * mult, dropout=dropout, )] ch = base_channels * mult # channels.append(ch) if ds in attention_ds: layers.append( AttentionBlock( ch, n_heads=n_heads, n_head_channels=n_head_channels, ) ) self.down.append(TimestepEmbedSequential(*layers)) channels.append(ch) if i != len(channel_mults) - 1: out_channels = ch self.down.append( TimestepEmbedSequential( ResBlock( ch, time_embed_dim=time_embed_dim, out_channels=out_channels, dropout=dropout, down=True ) if biggan_updown else Downsample(ch, conv_resample, out_channels=out_channels) ) ) ds *= 2 ch = out_channels channels.append(ch) self.middle = TimestepEmbedSequential( ResBlock( ch, time_embed_dim=time_embed_dim, dropout=dropout ), AttentionBlock( ch, n_heads=n_heads, n_head_channels=n_head_channels ), ResBlock( ch, time_embed_dim=time_embed_dim, dropout=dropout ) ) self.up = nn.ModuleList([]) for i, mult in reversed(list(enumerate(channel_mults))): for j in range(num_res_blocks + 1): inp_chs = channels.pop() layers = [ ResBlock( ch + inp_chs, time_embed_dim=time_embed_dim, out_channels=base_channels * mult, dropout=dropout ) ] ch = base_channels * mult if ds in attention_ds: layers.append( AttentionBlock( ch, n_heads=n_heads, n_head_channels=n_head_channels ), ) if i and j == num_res_blocks: out_channels = ch layers.append( ResBlock( ch, time_embed_dim=time_embed_dim, out_channels=out_channels, dropout=dropout, up=True ) if biggan_updown else Upsample(ch, conv_resample, out_channels=out_channels) ) ds //= 2 self.up.append(TimestepEmbedSequential(*layers)) self.out = nn.Sequential( GroupNorm32(32, ch), nn.SiLU(), zero_module(nn.Conv2d(base_channels * channel_mults[0], self.out_channels, 3, padding=1)) ) def forward(self, x, time): time_embed = self.time_embedding(time) skips = [] h = x.type(self.dtype) for i, module in enumerate(self.down): h = module(h, time_embed) skips.append(h) h = self.middle(h, time_embed) for i, module in enumerate(self.up): h = torch.cat([h, skips.pop()], dim=1) h = module(h, time_embed) h = h.type(x.dtype) h = self.out(h) return h class GroupNorm32(nn.GroupNorm): def forward(self, x): return super().forward(x.float()).type(x.dtype) def zero_module(module): """ Zero out the parameters of a module and return it. """ for p in module.parameters(): p.detach().zero_() return module def update_ema_params(target, source, decay_rate=0.9999): targParams = dict(target.named_parameters()) srcParams = dict(source.named_parameters()) for k in targParams: targParams[k].data.mul_(decay_rate).add_(srcParams[k].data, alpha=1 - decay_rate) ================================================ FILE: visualize.py ================================================ import matplotlib.pyplot as plt from torchvision.transforms import transforms import numpy as np import torch import os from dataset import * def visualalize_reconstruction(input, recon, target): plt.figure(figsize=(11,11)) plt.subplot(1, 3, 1).axis('off') plt.subplot(1, 3, 2).axis('off') plt.subplot(1, 3, 3).axis('off') plt.subplot(1, 3, 1) plt.imshow(show_tensor_image(input)) plt.title('input image') plt.subplot(1, 3, 2) plt.imshow(show_tensor_mask(recon)) plt.title('recon image') plt.subplot(1, 3, 3) plt.imshow(show_tensor_mask(target)) plt.title('target image') k = 0 while os.path.exists('results/heatmap{}.png'.format(k)): k += 1 plt.savefig('results/heatmap{}.png'.format(k)) plt.close() # def visualize_reconstructed(input, data,s): # fig, axs = plt.subplots(int(len(data)/5),6) # row = 0 # col = 1 # axs[0,0].imshow(show_tensor_image(input)) # axs[0, 0].get_xaxis().set_visible(False) # axs[0, 0].get_yaxis().set_visible(False) # axs[0,0].set_title('input') # for i, img in enumerate(data): # axs[row, col].imshow(show_tensor_image(img)) # axs[row, col].get_xaxis().set_visible(False) # axs[row, col].get_yaxis().set_visible(False) # axs[row, col].set_title(str(i)) # col += 1 # if col == 6: # row += 1 # col = 0 # col = 6 # row = int(len(data)/5) # remain = col * row - len(data) -1 # for j in range(remain): # col -= 1 # axs[row-1, col].remove() # axs[row-1, col].get_xaxis().set_visible(False) # axs[row-1, col].get_yaxis().set_visible(False) # plt.subplots_adjust(left=0.1, # bottom=0.1, # right=0.9, # top=0.9, # wspace=0.4, # hspace=0.4) # k = 0 # while os.path.exists(f'results/reconstructed{k}{s}.png'): # k += 1 # plt.savefig(f'results/reconstructed{k}{s}.png') # plt.close() def visualize(image, noisy_image, GT, pred_mask, anomaly_map, category) : for idx, img in enumerate(image): plt.figure(figsize=(11,11)) plt.subplot(1, 2, 1).axis('off') plt.subplot(1, 2, 2).axis('off') plt.subplot(1, 2, 1) plt.imshow(show_tensor_image(image[idx])) plt.title('clear image') plt.subplot(1, 2, 2) plt.imshow(show_tensor_image(noisy_image[idx])) plt.title('reconstructed image') plt.savefig('results/{}sample{}.png'.format(category,idx)) plt.close() plt.figure(figsize=(11,11)) plt.subplot(1, 3, 1).axis('off') plt.subplot(1, 3, 2).axis('off') plt.subplot(1, 3, 3).axis('off') plt.subplot(1, 3, 1) plt.imshow(show_tensor_mask(GT[idx])) plt.title('ground truth') plt.subplot(1, 3, 2) plt.imshow(show_tensor_mask(pred_mask[idx])) plt.title('normal' if torch.max(pred_mask[idx]) == 0 else 'abnormal', color="g" if torch.max(pred_mask[idx]) == 0 else "r") plt.subplot(1, 3, 3) plt.imshow(show_tensor_image(anomaly_map[idx])) plt.title('heat map') plt.savefig('results/{}sample{}heatmap.png'.format(category,idx)) plt.close() def show_tensor_image(image): reverse_transforms = transforms.Compose([ transforms.Lambda(lambda t: (t + 1) / (2)), transforms.Lambda(lambda t: t.permute(1, 2, 0)), # CHW to HWC transforms.Lambda(lambda t: t * 255.), transforms.Lambda(lambda t: t.cpu().numpy().astype(np.uint8)), ]) # Takes the first image of batch if len(image.shape) == 4: image = image[0, :, :, :] return reverse_transforms(image) def show_tensor_mask(image): reverse_transforms = transforms.Compose([ # transforms.Lambda(lambda t: (t + 1) / (2)), transforms.Lambda(lambda t: t.permute(1, 2, 0)), # CHW to HWC transforms.Lambda(lambda t: t.cpu().numpy().astype(np.int8)), ]) # Takes the first image of batch if len(image.shape) == 4: image = image[0, :, :, :] return reverse_transforms(image)