forked from isalirezag/TReS
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodels.py
380 lines (256 loc) · 12.4 KB
/
models.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
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
import torch
import torchvision.models as models
import torchvision
import torch.nn.functional as F
from torch import nn, Tensor
import numpy as np
from scipy import stats
from tqdm import tqdm
import os
import math
import csv
import copy
import json
from typing import Optional, List
import data_loader
from transformers import Transformer
from posencode import PositionEmbeddingSine
class L2pooling(nn.Module):
def __init__(self, filter_size=5, stride=1, channels=None, pad_off=0):
super(L2pooling, self).__init__()
self.padding = (filter_size - 2 )//2
self.stride = stride
self.channels = channels
a = np.hanning(filter_size)[1:-1]
g = torch.Tensor(a[:,None]*a[None,:])
g = g/torch.sum(g)
self.register_buffer('filter', g[None,None,:,:].repeat((self.channels,1,1,1)))
def forward(self, input):
input = input**2
out = F.conv2d(input, self.filter, stride=self.stride, padding=self.padding, groups=input.shape[1])
return (out+1e-12).sqrt()
class Net(nn.Module):
def __init__(self,cfg,device):
super(Net, self).__init__()
self.device = device
self.cfg = cfg
self.L2pooling_l1 = L2pooling(channels=256)
self.L2pooling_l2 = L2pooling(channels=512)
self.L2pooling_l3 = L2pooling(channels=1024)
self.L2pooling_l4 = L2pooling(channels=2048)
if cfg.network =='resnet50':
from resnet_modify import resnet50 as resnet_modifyresnet
dim_modelt = 3840
modelpretrain = models.resnet50(pretrained=True)
elif cfg.network =='resnet34':
from resnet_modify import resnet34 as resnet_modifyresnet
modelpretrain = models.resnet34(pretrained=True)
dim_modelt = 960
self.L2pooling_l1 = L2pooling(channels=64)
self.L2pooling_l2 = L2pooling(channels=128)
self.L2pooling_l3 = L2pooling(channels=256)
self.L2pooling_l4 = L2pooling(channels=512)
elif cfg.network == 'resnet18':
from resnet_modify import resnet18 as resnet_modifyresnet
modelpretrain = models.resnet18(pretrained=True)
dim_modelt = 960
self.L2pooling_l1 = L2pooling(channels=64)
self.L2pooling_l2 = L2pooling(channels=128)
self.L2pooling_l3 = L2pooling(channels=256)
self.L2pooling_l4 = L2pooling(channels=512)
torch.save(modelpretrain.state_dict(), 'modelpretrain')
self.model = resnet_modifyresnet()
self.model.load_state_dict(torch.load('modelpretrain'), strict=True)
self.dim_modelt = dim_modelt
os.remove("modelpretrain")
nheadt=cfg.nheadt
num_encoder_layerst=cfg.num_encoder_layerst
dim_feedforwardt=cfg.dim_feedforwardt
ddropout=0.5
normalize =True
self.transformer = Transformer(d_model=dim_modelt,nhead=nheadt,
num_encoder_layers=num_encoder_layerst,
dim_feedforward=dim_feedforwardt,
normalize_before=normalize,
dropout = ddropout)
self.position_embedding = PositionEmbeddingSine(dim_modelt // 2, normalize=True)
self.fc2 = nn.Linear(dim_modelt, self.model.fc.in_features)
self.fc = nn.Linear(self.model.fc.in_features*2, 1)
self.ReLU = nn.ReLU()
self.avg7 = nn.AvgPool2d((7, 7))
self.avg8 = nn.AvgPool2d((8, 8))
self.avg4 = nn.AvgPool2d((4, 4))
self.avg2 = nn.AvgPool2d((2, 2))
self.drop2d = nn.Dropout(p=0.1)
self.consistency = nn.L1Loss()
def forward(self, x):
self.pos_enc_1 = self.position_embedding(torch.ones(1, self.dim_modelt, 7, 7).to(self.device))
self.pos_enc = self.pos_enc_1.repeat(x.shape[0],1,1,1).contiguous()
out,layer1,layer2,layer3,layer4 = self.model(x)
layer1_t = self.avg8(self.drop2d(self.L2pooling_l1(F.normalize(layer1,dim=1, p=2))))
layer2_t = self.avg4(self.drop2d(self.L2pooling_l2(F.normalize(layer2,dim=1, p=2))))
layer3_t = self.avg2(self.drop2d(self.L2pooling_l3(F.normalize(layer3,dim=1, p=2))))
layer4_t = self.drop2d(self.L2pooling_l4(F.normalize(layer4,dim=1, p=2)))
layers = torch.cat((layer1_t,layer2_t,layer3_t,layer4_t),dim=1)
out_t_c = self.transformer(layers,self.pos_enc)
out_t_o = torch.flatten(self.avg7(out_t_c),start_dim=1)
out_t_o = self.fc2(out_t_o)
layer4_o = self.avg7(layer4)
layer4_o = torch.flatten(layer4_o,start_dim=1)
predictionQA = self.fc(torch.flatten(torch.cat((out_t_o,layer4_o),dim=1),start_dim=1))
# =============================================================================
# =============================================================================
fout,flayer1,flayer2,flayer3,flayer4 = self.model(torch.flip(x, [3]))
flayer1_t = self.avg8( self.L2pooling_l1(F.normalize(flayer1,dim=1, p=2)))
flayer2_t = self.avg4( self.L2pooling_l2(F.normalize(flayer2,dim=1, p=2)))
flayer3_t = self.avg2( self.L2pooling_l3(F.normalize(flayer3,dim=1, p=2)))
flayer4_t = self.L2pooling_l4(F.normalize(flayer4,dim=1, p=2))
flayers = torch.cat((flayer1_t,flayer2_t,flayer3_t,flayer4_t),dim=1)
fout_t_c = self.transformer(flayers,self.pos_enc)
fout_t_o = torch.flatten(self.avg7(fout_t_c),start_dim=1)
fout_t_o = (self.fc2(fout_t_o))
flayer4_o = self.avg7(flayer4)
flayer4_o = torch.flatten(flayer4_o,start_dim=1)
fpredictionQA = (self.fc(torch.flatten(torch.cat((fout_t_o,flayer4_o),dim=1),start_dim=1)))
consistloss1 = self.consistency(out_t_c,fout_t_c.detach())
consistloss2 = self.consistency(layer4,flayer4.detach())
consistloss = 1*(consistloss1+consistloss2)
return predictionQA, consistloss
class TReS(object):
def __init__(self, config, device, svPath, datapath, train_idx, test_idx,Net):
super(TReS, self).__init__()
self.device = device
self.epochs = config.epochs
self.test_patch_num = config.test_patch_num
self.l1_loss = torch.nn.L1Loss()
self.lr = 2e-5
self.lrratio = 10
self.weight_decay = config.weight_decay
self.net = Net(config,device).to(device)
self.droplr = config.droplr
self.config = config
self.clsloss = nn.CrossEntropyLoss()
self.paras = [{'params': self.net.parameters(), 'lr': self.lr} ]
self.solver = torch.optim.Adam(self.paras, weight_decay=self.weight_decay)
train_loader = data_loader.DataLoader(config.dataset, datapath,
train_idx, config.patch_size,
config.train_patch_num,
batch_size=config.batch_size, istrain=True)
test_loader = data_loader.DataLoader(config.dataset, datapath,
test_idx, config.patch_size,
config.test_patch_num, istrain=False)
self.train_data = train_loader.get_data()
self.test_data = test_loader.get_data()
def train(self,seed,svPath):
best_srcc = 0.0
best_plcc = 0.0
print('Epoch\tTrain_Loss\tTrain_SRCC\tTest_SRCC\tTest_PLCC\tLearning_Rate\tdroplr')
steps = 0
results = {}
performPath = svPath +'/' + 'PLCC_SRCC_'+str(self.config.vesion)+'_'+str(seed)+'.json'
with open(performPath, 'w') as json_file2:
json.dump( {} , json_file2)
for epochnum in range(self.epochs):
self.net.train()
epoch_loss = []
pred_scores = []
gt_scores = []
pbar = tqdm(self.train_data, leave=False)
for img, label in pbar:
img = torch.as_tensor(img.to(self.device)).requires_grad_(False)
label = torch.as_tensor(label.to(self.device)).requires_grad_(False)
steps+=1
self.net.zero_grad()
pred,closs = self.net(img)
pred2,closs2 = self.net(torch.flip(img, [3]))
pred_scores = pred_scores + pred.flatten().cpu().tolist()
gt_scores = gt_scores + label.cpu().tolist()
loss_qa = self.l1_loss(pred.squeeze(), label.float().detach())
loss_qa2 = self.l1_loss(pred2.squeeze(), label.float().detach())
# =============================================================================
# =============================================================================
indexlabel = torch.argsort(label) # small--> large
anchor1 = torch.unsqueeze(pred[indexlabel[0],...].contiguous(),dim=0) # d_min
positive1 = torch.unsqueeze(pred[indexlabel[1],...].contiguous(),dim=0) # d'_min+
negative1_1 = torch.unsqueeze(pred[indexlabel[-1],...].contiguous(),dim=0) # d_max+
anchor2 = torch.unsqueeze(pred[indexlabel[-1],...].contiguous(),dim=0)# d_max
positive2 = torch.unsqueeze(pred[indexlabel[-2],...].contiguous(),dim=0)# d'_max+
negative2_1 = torch.unsqueeze(pred[indexlabel[0],...].contiguous(),dim=0)# d_min+
# =============================================================================
# =============================================================================
fanchor1 = torch.unsqueeze(pred2[indexlabel[0],...].contiguous(),dim=0)
fpositive1 = torch.unsqueeze(pred2[indexlabel[1],...].contiguous(),dim=0)
fnegative1_1 = torch.unsqueeze(pred2[indexlabel[-1],...].contiguous(),dim=0)
fanchor2 = torch.unsqueeze(pred2[indexlabel[-1],...].contiguous(),dim=0)
fpositive2 = torch.unsqueeze(pred2[indexlabel[-2],...].contiguous(),dim=0)
fnegative2_1 = torch.unsqueeze(pred2[indexlabel[0],...].contiguous(),dim=0)
consistency = nn.L1Loss()
assert (label[indexlabel[-1]]-label[indexlabel[1]])>=0
assert (label[indexlabel[-2]]-label[indexlabel[0]])>=0
triplet_loss1 = nn.TripletMarginLoss(margin=(label[indexlabel[-1]]-label[indexlabel[1]]), p=1) # d_min,d'_min,d_max
# triplet_loss2 = nn.TripletMarginLoss(margin=label[indexlabel[0]], p=1)
triplet_loss2 = nn.TripletMarginLoss(margin=(label[indexlabel[-2]]-label[indexlabel[0]]), p=1)
# triplet_loss1 = nn.TripletMarginLoss(margin=label[indexlabel[-1]], p=1)
# triplet_loss2 = nn.TripletMarginLoss(margin=label[indexlabel[0]], p=1)
tripletlosses = triplet_loss1(anchor1, positive1, negative1_1) + \
triplet_loss2(anchor2, positive2, negative2_1)
ftripletlosses = triplet_loss1(fanchor1, fpositive1, fnegative1_1) + \
triplet_loss2(fanchor2, fpositive2, fnegative2_1)
loss = loss_qa + closs + loss_qa2 + closs2 + 0.5*( self.l1_loss(tripletlosses,ftripletlosses.detach())+ self.l1_loss(ftripletlosses,tripletlosses.detach()))+0.05*(tripletlosses+ftripletlosses)
epoch_loss.append(loss.item())
loss.backward()
self.solver.step()
modelPath = svPath + '/model_{}_{}_{}'.format(str(self.config.vesion),str(seed),epochnum)
torch.save(self.net.state_dict(), modelPath)
train_srcc, _ = stats.spearmanr(pred_scores, gt_scores)
test_srcc, test_plcc = self.test(self.test_data,epochnum,svPath,seed)
results[epochnum]=(test_srcc, test_plcc)
with open(performPath, "r+") as file:
data = json.load(file)
data.update(results)
file.seek(0)
json.dump(data, file)
if test_srcc > best_srcc:
modelPathbest = svPath + '/bestmodel_{}_{}'.format(str(self.config.vesion),str(seed))
torch.save(self.net.state_dict(), modelPathbest)
best_srcc = test_srcc
best_plcc = test_plcc
print('{}\t{:4.3f}\t\t{:4.4f}\t\t{:4.4f}\t\t{:4.3f}\t\t{}\t\t{:4.3f}'.format(epochnum + 1, sum(epoch_loss) / len(epoch_loss), train_srcc, test_srcc, test_plcc,self.paras[0]['lr'],self.droplr ))
if (epochnum+1)==self.droplr or (epochnum+1)==(2*self.droplr) or (epochnum+1)==(3*self.droplr):
self.lr = self.lr /self.lrratio
self.paras = [{'params': self.net.parameters(), 'lr': self.lr} ]
self.solver = torch.optim.Adam(self.paras, weight_decay=self.weight_decay)
print('Best test SRCC %f, PLCC %f' % (best_srcc, best_plcc))
return best_srcc, best_plcc
def test(self, data,epochnum,svPath,seed,pretrained=0):
if pretrained:
self.net.load_state_dict(torch.load(svPath+'/bestmodel_{}_{}'.format(str(self.config.vesion),str(seed))))
self.net.eval()
pred_scores = []
gt_scores = []
pbartest = tqdm(data, leave=False)
with torch.no_grad():
steps2 = 0
for img, label in pbartest:
img = torch.as_tensor(img.to(self.device))
label = torch.as_tensor(label.to(self.device))
pred,_ = self.net(img)
pred_scores = pred_scores + pred.cpu().tolist()
gt_scores = gt_scores + label.cpu().tolist()
steps2 += 1
pred_scores = np.mean(np.reshape(np.array(pred_scores), (-1, self.test_patch_num)), axis=1)
gt_scores = np.mean(np.reshape(np.array(gt_scores), (-1, self.test_patch_num)), axis=1)
# if not pretrained:
dataPath = svPath + '/test_prediction_gt_{}_{}_{}.csv'.format(str(self.config.vesion),str(seed),epochnum)
with open(dataPath, 'w') as f:
writer = csv.writer(f)
writer.writerows(zip(pred_scores, gt_scores))
test_srcc, _ = stats.spearmanr(pred_scores, gt_scores)
test_plcc, _ = stats.pearsonr(pred_scores, gt_scores)
return test_srcc, test_plcc
if __name__=='__main__':
import os
import argparse
import random
import numpy as np
from args import *