-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
146 lines (128 loc) · 6.43 KB
/
main.py
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
import os
import torch
import numpy as np
import argparse
import torchvision.transforms as transforms
from torchvision.datasets import ImageFolder
import logging
import model.utils as utils
from data.data_loader import MVTecTrainDataset,MVTecTestDataset, VisATrainDataset, VisATestDataset
import json
import random
import script.backbones as backbones
LOGGER = logging.getLogger(__name__)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--gpu", type=int, default=0)
parser.add_argument("--seed", type=int, default=0)
parser.add_argument("--backbone_name", type=str, default="wideresnet50")
parser.add_argument("--layers_to_extract_from", "-le", type=str, default=["layer2","layer3"])
parser.add_argument("--pretrain_embed_dimension", type=int, default=1536)
parser.add_argument("--target_embed_dimension", type=int, default=1536)
parser.add_argument("--dsc_hidden", type=int, default=1024)
parser.add_argument("--patchsize", type=int, default=3)
parser.add_argument("--patchstride", type=int, default=1)
parser.add_argument("--patchoverlap", type=float, default=0.0)
parser.add_argument("--epochs", type=int, default=4, help="train epochs")
parser.add_argument("--meta_epochs", type=int, default=20, help="train")
parser.add_argument("--n_layers", type=int, default=2, help="layers of discriminator")
parser.add_argument("--lr_perlin", type=float, default=0.0001, help="dis_perlin lr")
parser.add_argument("--lr_gaussian", type=float, default=0.0002, help="dis_gaussian lr")
parser.add_argument("--lr_proj", type=float, default=0.0005, help="proj lr")
#save_path
parser.add_argument("--save_path", type=str, default="2_shot")
#dataset
parser.add_argument('--dataset_name', action='store', type=str, default='visa')
parser.add_argument('--anomaly_source_path', action='store', type=str, default='../datasets/dtd/images/')
parser.add_argument('--data_path',type=str, default="/opt/data/private/datasets/VisA/")
parser.add_argument("--batch_size", default=8, type=int)
parser.add_argument("--num_workers", default=16, type=int)
parser.add_argument("--resize", default=256, type=int)
parser.add_argument("--imagesize", default=256, type=int)
parser.add_argument('--k_shot',default=1, type=int)
parser.add_argument('--num',default=80, type=int, help='number of augmented images')
return parser.parse_args()
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
def train_and_infer(args):
if args.dataset_name == 'visa':
classes = ['pcb1', 'pcb2', 'pcb3', 'pcb4', 'candle', 'pipe_fryum', 'capsules',
'cashew', 'chewinggum', 'fryum','macaroni1', 'macaroni2',]
if args.dataset_name == 'mvtec':
classes = ["bottle", "cable", "capsule", "carpet", "grid","hazelnut", "leather", "metal_nut", "pill",
"screw", "tile", "toothbrush", "transistor", "wood", "zipper"]
with open("./data/anomaly_mask.json") as f:
anomaly_mask = json.load(f)
result_collect = []
run_save_path = utils.create_storage_folder('results',args.save_path, mode="iterate")
models_dir = os.path.join(run_save_path, "models")
i = 0
for _class_ in classes:
print("processing:{}/{}".format(i+1,len(classes)))
print("current class:",_class_)
i += 1
if args.dataset_name == 'mvtec':
bg_re = anomaly_mask[_class_]["bg_reverse"]
use_mask = anomaly_mask[_class_]["use_mask"]
train_path = args.data_path + _class_ +'/train/good/'
test_path = args.data_path + _class_ + '/test/'
train_data = MVTecTrainDataset(train_path, args.anomaly_source_path, resize_shape=args.imagesize, k_shot=args.k_shot, num=args.num, use_mask=use_mask, bg_reverse=bg_re)
test_data = MVTecTestDataset(test_path,resize_shape=args.imagesize)
if args.dataset_name == 'visa':
train_path = args.data_path + _class_ +'/train/good/'
test_path = args.data_path + _class_ + '/test/'
train_data = VisATrainDataset(train_path, args.anomaly_source_path, resize_shape=args.imagesize, k_shot=args.k_shot, num=args.num)
test_data = VisATestDataset(test_path,resize_shape=args.imagesize)
train_dataloader = torch.utils.data.DataLoader(train_data, batch_size=args.batch_size)
test_dataloader = torch.utils.data.DataLoader(test_data, batch_size=1)
device = utils.set_torch_device(args.gpu)
net = backbones.load(args.backbone_name)
DFD = utils.model(device)
DFD.load(
device=device,
backbone=net,
layers_to_extract_from=args.layers_to_extract_from,
input_shape=(3, args.imagesize, args.imagesize),
pretrain_embed_dimension=args.pretrain_embed_dimension,
target_embed_dimension=args.target_embed_dimension,
patchsize=args.patchsize,
patchstride=args.patchstride,
lr_perlin=args.lr_perlin,
lr_gaussian=args.lr_gaussian,
lr_proj=args.lr_proj,
dsc_hidden=args.dsc_hidden,
epochs=args.epochs,
meta_epochs=args.meta_epochs,
)
DFD.set_model_dir(models_dir, _class_)
auroc_px, auroc_sp, pro_auc = DFD.train(train_dataloader, test_dataloader)
result_collect.append(
{
"dataset_name": _class_,
"sample_auroc": round(auroc_sp*100,3),
"pixel_auroc" :round(auroc_px*100,3),
"aupro" :round(pro_auc*100,3),
}
)
for key, item in result_collect[-1].items():
if key != "dataset_name":
LOGGER.info("{0}: {1:3.3f}".format(key, item))
print("{0}: {1:3.3f}".format(key, item))
result_scores = [list(results.values())[1:] for results in result_collect]
result_metric_names = list(result_collect[-1].keys())[1:]
result_dataset_names = [results["dataset_name"] for results in result_collect]
utils.compute_and_store_final_results(
run_save_path,
result_scores,
column_names=result_metric_names,
row_names=result_dataset_names,
)
if __name__ == '__main__':
args = parse_args()
setup_seed(args.seed)
train_and_infer(args)