-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathtrainRM_ST.py
205 lines (157 loc) · 7.68 KB
/
trainRM_ST.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
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
198
199
200
201
202
203
204
import os
import time
import numpy as np
import torch
from tqdm import tqdm
from datasets.mvtec import MVTecDataset
from utils.util import AverageMeter,readYamlConfig,computeAUROC,loadWeights,load_dataset
from utils.functions import (
cal_loss,
cal_loss_cosine,
cal_loss_orth,
cal_anomaly_maps
)
from models.ST.teacherST import wide_resnet50_2,resnet18
from models.ST.resnetRM import resnet18Memory,resnet50Memory
from utils.visualization import visu
class NetTrainer:
def __init__(self, data,device):
self.device = device
self.validation_ratio = 0.2
self.data_path = data['data_path']
self.dataset=data['dataset']
self.obj = data['obj']
self.img_resize = data['TrainingData']['img_size']
self.img_cropsize = data['TrainingData']['crop_size']
self.num_epochs = data['TrainingData']['epochs']
self.lr = data['TrainingData']['lr']
self.batch_size = data['TrainingData']['batch_size']
self.embedDim = data['TrainingData']['embedDim']
self.n_embed = data['TrainingData']['n_embed']
self.lambda1 = data['TrainingData']['lambda1']
self.lambda2 = data['TrainingData']['lambda1']
self.model_dir = data['save_path'] + "/models" + "/" + self.obj
os.makedirs(self.model_dir, exist_ok=True)
self.visu=data['visu']
if (self.visu):
self.img_dir = data['save_path'] + "/images" + "/" + self.obj
os.makedirs(self.img_dir, exist_ok=True)
self.modelName = data['backbone']
self.load_model()
load_dataset(self)
self.optimizer = torch.optim.Adam(self.student.parameters(), lr=self.lr, betas=(0.5, 0.999))
self.scheduler = torch.optim.lr_scheduler.OneCycleLR(self.optimizer,max_lr=self.lr*10,epochs=self.num_epochs,steps_per_epoch=len(self.train_loader))
def load_model(self):
print("loading and training")
if self.modelName == "resnet18":
self.teacher=resnet18(pretrained=True)
self.student=resnet18Memory(embedDim=self.embedDim).to(self.device)
elif self.modelName == "wide_resnet50_2":
self.teacher=wide_resnet50_2(pretrained=True).to(self.device)
self.student=resnet50Memory(embedDim=self.embedDim).to(self.device) # ! cf Supplementary material
else :
print("Invalid/unconfigured model name")
exit()
self.teacher=self.teacher.to(self.device)
self.teacher.eval()
for param in self.teacher.parameters():
param.requires_grad = False
def train(self):
print("training " + self.obj)
self.student.train()
best_score = None
start_time = time.time()
epoch_time = AverageMeter()
epoch_bar = tqdm(total=len(self.train_loader) * self.num_epochs,desc="Training",unit="batch")
for _ in range(1, self.num_epochs + 1):
losses = AverageMeter()
for (image,_, _),(imageExamplar,_,_) in zip(self.train_loader, self.train_examplar_loader):
image= image.to(self.device)
imageExamplar= imageExamplar.to(self.device)
self.optimizer.zero_grad()
with torch.set_grad_enabled(True):
features_s,features_t,features_t_examplar,features_t_examplar_norm = self.infer(image,imageExamplar)
loss_KD=cal_loss_cosine(features_s, features_t)
loss_NM=cal_loss(features_t_examplar, features_t_examplar_norm)
loss_ORTH=cal_loss_orth(self.student)
loss=loss_KD+self.lambda1*loss_NM +self.lambda2*loss_ORTH
losses.update(loss.sum().item(), image.size(0))
loss.backward()
self.optimizer.step()
self.scheduler.step()
epoch_bar.set_postfix({"Loss": loss.item()})
epoch_bar.update()
val_loss = self.val(epoch_bar)
if best_score is None:
best_score = val_loss
self.save_checkpoint()
elif val_loss < best_score:
best_score = val_loss
self.save_checkpoint()
epoch_time.update(time.time() - start_time)
start_time = time.time()
epoch_bar.close()
print("Training end.")
def val(self, epoch_bar):
self.student.eval()
losses = AverageMeter()
for (image,_, _),(imageExamplar,_,_) in zip(self.val_loader,self.val_examplar_loader):
image= image.to(self.device)
imageExamplar= imageExamplar.to(self.device)
with torch.set_grad_enabled(False):
features_s,features_t,features_t_examplar,features_t_examplar_norm = self.infer(image,imageExamplar)
loss_KD=cal_loss_cosine(features_s, features_t)
loss_NM=cal_loss(features_t_examplar, features_t_examplar_norm)
loss_ORTH=cal_loss_orth(self.student)
loss=loss_KD+self.lambda1*loss_NM +self.lambda2*loss_ORTH
losses.update(loss.item(), image.size(0))
epoch_bar.set_postfix({"Loss": loss.item()})
return losses.avg
def save_checkpoint(self):
state = {"model": self.student.state_dict()}
torch.save(state, os.path.join(self.model_dir, "student.pth"))
@torch.no_grad()
def test(self):
self.student=loadWeights(self.student,self.model_dir,"student.pth")
scores = []
test_imgs = []
gt_list = []
gt_mask_list = []
progressBar = tqdm(self.test_loader)
for image, label, mask in self.test_loader:
test_imgs.extend(image.cpu().numpy())
gt_list.extend(label.cpu().numpy())
gt_mask_list.append(mask.squeeze().cpu().numpy())
image = image.to(self.device)
with torch.set_grad_enabled(False):
features_t = self.teacher(image)
features_s=self.student(image)
score =cal_anomaly_maps(features_s,features_t,self.img_cropsize)
progressBar.update()
scores.append(score)
progressBar.close()
scores = np.asarray(scores)
gt_list = np.asarray(gt_list)
img_roc_auc,img_scores,map_scores=computeAUROC(scores,gt_list,self.obj,"forward distillation")
if self.visu:
visu(self,gt_list, img_scores, gt_mask_list, map_scores, test_imgs)
return img_roc_auc
def infer(self, img,imgExamplar):
features_t_examplar = self.teacher.forward_normality_embedding(imgExamplar)
features_t_examplar = [features_t_examplar[1],features_t_examplar[2]]
features_t_examplar_norm=[self.student.memory1(features_t_examplar[0],normality=True),
self.student.memory2(features_t_examplar[1],normality=True)]
features_t = self.teacher(img)
features_s=self.student(img)
return features_s,features_t ,features_t_examplar,features_t_examplar_norm
if __name__ == "__main__":
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
data=readYamlConfig("config.yaml")
distill = NetTrainer(data,device)
if data['phase'] == "train":
distill.train()
distill.test()
elif data['phase'] == "test":
distill.test()
else:
print("Phase argument must be train or test.")