anomalydiffusion/train-localization.py at master · sjtuplayer/anomalydiffusion · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197

import torch
from torch import optim
from unet_utils.tensorboard_visualizer import TensorboardVisualizer
from unet_utils.loss import FocalLoss, SSIM
import os
from unet_utils.data_loader import MVTec_Anomaly_Detection,MVTecDRAEMTestDataset_partial
from torch.utils.data import DataLoader
import numpy as np
from sklearn.metrics import roc_auc_score, average_precision_score
from unet_utils.model_unet import DiscriminativeSubNetwork
import os
from unet_utils.au_pro_util import calculate_au_pro
def get_lr(optimizer):
    for param_group in optimizer.param_groups:
        return param_group['lr']

def weights_init(m):
    classname = m.__class__.__name__
    if classname.find('Conv') != -1:
        m.weight.data.normal_(0.0, 0.02)
    elif classname.find('BatchNorm') != -1:
        m.weight.data.normal_(1.0, 0.02)
        m.bias.data.fill_(0)
def test(args,obj_name, model_seg):
    mvtec_path = args.mvtec_path
    obj_ap_pixel_list = []
    obj_auroc_pixel_list = []
    obj_ap_image_list = []
    obj_auroc_image_list = []
    img_dim = 256
    model_seg.eval()
    dataset = MVTecDRAEMTestDataset_partial(mvtec_path +'/'+ obj_name + "/test/", resize_shape=[img_dim, img_dim])
    dataloader = DataLoader(dataset, batch_size=1,
                            shuffle=False, num_workers=0)

    total_pixel_scores = np.zeros((img_dim * img_dim * len(dataset)))
    total_gt_pixel_scores = np.zeros((img_dim * img_dim * len(dataset)))
    mask_cnt = 0

    anomaly_score_gt = []
    anomaly_score_prediction = []

    gt_masks=[]
    predicted_masks=[]

    for i_batch, sample_batched in enumerate(dataloader):

        gray_batch = sample_batched["image"].cuda()
        gray_batch=gray_batch[:,[2,1,0],:,:]

        is_normal = sample_batched["has_anomaly"].detach().numpy()[0 ,0]
        anomaly_score_gt.append(is_normal)
        true_mask = sample_batched["mask"]
        true_mask_cv = true_mask.detach().numpy()[0, :, :, :].transpose((1, 2, 0))
        out_mask = model_seg(gray_batch)
        out_mask_sm = torch.softmax(out_mask, dim=1)

        out_mask_cv = out_mask_sm[0 ,1 ,: ,:].detach().cpu().numpy()
        out_mask_averaged = torch.nn.functional.avg_pool2d(out_mask_sm[: ,1: ,: ,:], 21, stride=1,
                                                           padding=21 // 2).cpu().detach().numpy()
        image_score = np.max(out_mask_averaged)
        anomaly_score_prediction.append(image_score)

        flat_true_mask = true_mask_cv.flatten()
        flat_out_mask = out_mask_cv.flatten()
        gt_masks.append(true_mask_cv.squeeze())
        predicted_masks.append(out_mask_cv.squeeze())

        total_pixel_scores[mask_cnt * img_dim * img_dim:(mask_cnt + 1) * img_dim * img_dim] = flat_out_mask
        total_gt_pixel_scores[mask_cnt * img_dim * img_dim:(mask_cnt + 1) * img_dim * img_dim] = flat_true_mask
        mask_cnt += 1

    anomaly_score_prediction = np.array(anomaly_score_prediction)
    anomaly_score_gt = np.array(anomaly_score_gt)
    auroc = roc_auc_score(anomaly_score_gt, anomaly_score_prediction)
    ap = average_precision_score(anomaly_score_gt, anomaly_score_prediction)

    total_gt_pixel_scores = total_gt_pixel_scores.astype(np.uint8)
    total_gt_pixel_scores = total_gt_pixel_scores[:img_dim * img_dim * mask_cnt]
    total_pixel_scores = total_pixel_scores[:img_dim * img_dim * mask_cnt]
    auroc_pixel = roc_auc_score(total_gt_pixel_scores, total_pixel_scores)
    ap_pixel = average_precision_score(total_gt_pixel_scores, total_pixel_scores)
    pro_pixel, _ = calculate_au_pro(gt_masks, predicted_masks)
    obj_ap_pixel_list.append(ap_pixel)
    obj_auroc_pixel_list.append(auroc_pixel)
    obj_auroc_image_list.append(auroc)
    obj_ap_image_list.append(ap)
    print(obj_name)
    print("AUC Image:  " +str(auroc))
    print("AP Image:  " +str(ap))
    print("AUC Pixel:  " +str(auroc_pixel))
    #print("AUC Pixel:  " +str(auroc_pixel))
    print("AP Pixel:  " +str(ap_pixel))
    print('PRO Pixel:' +str(pro_pixel))
    print("==============================")
    return float(auroc),float(auroc_pixel),float(ap_pixel),float(pro_pixel)


def train_on_device(obj_names, args):

    if not os.path.exists(args.save_path):
        os.makedirs(args.save_path)

    if not os.path.exists(args.log_path):
        os.makedirs(args.log_path)

    for obj_name in obj_names:

        run_name = obj_name

        model_seg = DiscriminativeSubNetwork(in_channels=3, out_channels=2)
        model_seg.cuda()
        model_seg.apply(weights_init)

        optimizer = torch.optim.Adam([
                                      {"params": model_seg.parameters(), "lr": args.lr}])

        scheduler = optim.lr_scheduler.MultiStepLR(optimizer,[args.epochs*0.8,args.epochs*0.9],gamma=0.2, last_epoch=-1)

        loss_focal = FocalLoss()

        dataset = MVTec_Anomaly_Detection(args,obj_name,length=500)
        dataloader = DataLoader(dataset, batch_size=args.bs,
                                shuffle=True, num_workers=16)

        n_iter = 0
        last_sum=0
        for epoch in range(args.epochs):
            model_seg.train()
            print("Epoch: "+str(epoch))
            for i_batch, sample_batched in enumerate(dataloader):
                aug_gray_batch = sample_batched["image"].cuda()
                anomaly_mask = sample_batched["mask"].cuda()
                out_mask = model_seg(aug_gray_batch)
                out_mask_sm = torch.softmax(out_mask, dim=1)
                segment_loss = loss_focal(out_mask_sm, anomaly_mask)
                loss = segment_loss
                optimizer.zero_grad()
                loss.backward()
                optimizer.step()
                n_iter +=1
            scheduler.step()

            auroc,auroc_px,ap_px,pro_px=test(args,obj_name, model_seg)
            sum_metric=auroc+auroc_px+ap_px+pro_px
            if sum_metric>last_sum:
                torch.save(model_seg.state_dict(), os.path.join(args.save_path, run_name + ".pckl"))
                last_sum=sum_metric

if __name__=="__main__":
    import argparse

    parser = argparse.ArgumentParser()
    parser.add_argument('--sample_name', type=str, default='all')
    parser.add_argument('--generated_data_path', action='store', type=str, required=True)
    parser.add_argument('--save_path', default='checkpoints/localization', type=str)
    parser.add_argument('--mvtec_path', action='store', type=str, required=True)
    parser.add_argument('--bs', action='store', type=int,default=8, required=False)
    parser.add_argument('--lr', action='store', type=float,default=0.0001, required=False)
    parser.add_argument('--epochs', action='store', type=int,default=200, required=False)
    parser.add_argument('--gpu_id', action='store', type=int, default=0, required=False)
    parser.add_argument('--log_path', action='store', type=str,default='./logs/ ', required=False)
    parser.add_argument('--visualize', action='store_true')
    parser.add_argument('--test_separately', action='store_true',default=False)
    parser.add_argument('--reverse', action='store_true',default=False)
    parser.add_argument('--data_name',type=str, default='text_inversion')
    args = parser.parse_args()

    obj_batch =  [
                    'bottle',
                    'capsule',
                     'carpet',
                     'leather',
                     'pill',
                     'transistor',
                     'tile',
                     'cable',
                     'zipper',
                     'toothbrush',
                     'metal_nut',
                     'hazelnut',
                     'screw',
                     'grid',
                     'wood'
                     ]
    if args.reverse:
        obj_batch=reversed(obj_batch)
    if args.sample_name!='all':
        obj_list=[args.sample_name]
        picked_classes = obj_list
    else:
        picked_classes = obj_batch

    with torch.cuda.device(args.gpu_id):
        train_on_device(picked_classes, args)
#python train-unet.py --data_path $path_to_the_generated_data  --save_path ./ --mvtec_path=$path_to_mvtec --sample_name=capsule