anomalydiffusion/train-classification.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

import torch
from torch import optim
from unet_utils.data_loader import MVTec_classification_train,MVTec_classification_test
from torch.utils.data import DataLoader
import os
from torchvision.models import resnet34
import torch.nn as nn
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,anomaly_names):
    model.eval()

    dataset = MVTec_classification_test(args,obj_name,anomaly_names)
    dataloader = DataLoader(dataset, batch_size=100,
                            shuffle=False, num_workers=0)

    for i_batch, sample_batched in enumerate(dataloader):
        image, label = sample_batched
        image = image.cuda()
        label = label.cuda()
        y_pred = model(image)
        prediction = torch.argmax(y_pred, 1)
        correct = (prediction == label).sum().float()
        print("Accuracy: %.4f"%(correct/len(label)))
    return correct/len(label)


def train_on_device(obj_names, args):

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


    for obj_name in obj_names:
        print(obj_name)
        run_name = obj_name
        dataset = MVTec_classification_train(args,obj_name)
        class_num=dataset.class_num()
        anomaly_names =dataset.return_anomaly_names()
        model = resnet34(pretrained=True, progress=True)
        model.fc = nn.Linear(model.fc.in_features, class_num)
        model=model.cuda()

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

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

        criterion = nn.CrossEntropyLoss()
        dataloader = DataLoader(dataset, batch_size=args.bs,
                                shuffle=True, num_workers=16)
        max_acc=0
        for epoch in range(args.epochs):
            model.train()
            print("Epoch: "+str(epoch),end=' ')
            for i_batch, sample_batched in enumerate(dataloader):
                image,label=sample_batched
                image=image.cuda()
                label=label.cuda()
                y_pred=model(image)
                loss=criterion(y_pred,label)
                optimizer.zero_grad()

                loss.backward()
                optimizer.step()

            scheduler.step()
            acc = test(args,obj_name, model, anomaly_names)
            if acc> max_acc:
                max_acc=acc
                torch.save(model.state_dict(), os.path.join(args.checkpoint_path, run_name+".pckl"))

if __name__=="__main__":
    import argparse

    parser = argparse.ArgumentParser()
    parser.add_argument('--anomaly_id',  type=int, default=None)
    parser.add_argument('--sample_name', type=str, default='all')
    parser.add_argument('--mvtec_path', type=str,required=True)
    parser.add_argument('--generated_data_path', type=str, required=True)
    parser.add_argument('--bs', action='store', type=int, default=8)
    parser.add_argument('--lr', action='store', type=float, default=0.0001)
    parser.add_argument('--epochs', action='store', type=int, default=30)
    parser.add_argument(
        "--reverse",
        action="store_true", default=False,
    )
    parser.add_argument('--checkpoint_path', default='checkpoints/classification', type=str)

    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

    train_on_device(picked_classes, args)
#python train-classification.py