-
Notifications
You must be signed in to change notification settings - Fork 15
/
Copy pathTrain_clothing1M.py
447 lines (359 loc) · 16.9 KB
/
Train_clothing1M.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
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
from __future__ import print_function
import sys
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
import torchvision
import torchvision.models as models
import random
import os
import argparse
import numpy as np
import dataloader_clothing1M as dataloader
from sklearn.mixture import GaussianMixture
import copy
import torchnet
from Contrastive_loss import *
from PreResNet_source import *
parser = argparse.ArgumentParser(description='PyTorch Clothing1M Training')
parser.add_argument('--batch_size', default=32, type=int, help='train batchsize')
parser.add_argument('--lr', '--learning_rate', default=0.005, type=float, help='initial learning rate') ## Set the learning rate to 0.005 for faster training at the beginning
parser.add_argument('--alpha', default=0.5, type=float, help='parameter for Beta')
parser.add_argument('--lambda_c', default=0.025, type=float, help='weight for contrastive loss')
parser.add_argument('--T', default=0.5, type=float, help='sharpening temperature')
parser.add_argument('--d_u', default=0.7, type=float)
parser.add_argument('--num_epochs', default=200, type=int)
parser.add_argument('--id', default='clothing1m')
parser.add_argument('--tau', default=5, type=float, help='filtering coefficient')
parser.add_argument('--data_path', default='./data/Clothing1M_org', type=str, help='path to dataset')
parser.add_argument('--seed', default=123)
parser.add_argument('--gpuid', default=0, type=int)
parser.add_argument('--pretrained', default=True, type=bool)
parser.add_argument('--num_class', default=14, type=int)
parser.add_argument('--num_batches', default=1000, type=int)
parser.add_argument('--dataset', default="Clothing1M", type=str)
parser.add_argument('--resume', default=False, type=bool, help = 'Resume from the warmup checkpoint')
args = parser.parse_args()
torch.cuda.set_device(args.gpuid)
random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
contrastive_criterion = SupConLoss()
## For plotting the logs
# import wandb
# wandb.init(project="noisy-label-project-clothing1M", entity="...")
## Training
def train(epoch, net, net2, optimizer, labeled_trainloader, unlabeled_trainloader):
net2.eval() # Fix one network and train the other
net.train()
unlabeled_train_iter = iter(unlabeled_trainloader)
num_iter_un = (len(unlabeled_trainloader.dataset)//args.batch_size)+1
num_iter_lab = (len(labeled_trainloader.dataset)//args.batch_size)+1
loss_x = 0
loss_u = 0
loss_scl = 0
loss_ucl = 0
num_iter = num_iter_lab
for batch_idx, (inputs_x, inputs_x2, inputs_x3, inputs_x4, labels_x, w_x) in enumerate(labeled_trainloader):
try:
inputs_u, inputs_u2, inputs_u3, inputs_u4 = unlabeled_train_iter.next()
except:
unlabeled_train_iter = iter(unlabeled_trainloader)
inputs_u, inputs_u2, inputs_u3, inputs_u4 = unlabeled_train_iter.next()
batch_size = inputs_x.size(0)
# Transform label to one-hot
labels_x = torch.zeros(batch_size, args.num_class).scatter_(1, labels_x.view(-1,1), 1)
w_x = w_x.view(-1,1).type(torch.FloatTensor)
inputs_x, inputs_x2, inputs_x3, inputs_x4, labels_x, w_x = inputs_x.cuda(), inputs_x2.cuda(), inputs_x3.cuda(), inputs_x4.cuda(), labels_x.cuda(), w_x.cuda()
inputs_u, inputs_u2, inputs_u3, inputs_u4 = inputs_u.cuda(), inputs_u2.cuda(), inputs_u3.cuda(), inputs_u4.cuda()
with torch.no_grad():
# Label co-guessing of unlabeled samples
_, outputs_u11 = net(inputs_u3)
_, outputs_u12 = net(inputs_u4)
_, outputs_u21 = net2(inputs_u3)
_, outputs_u22 = net2(inputs_u4)
pu = (torch.softmax(outputs_u11, dim=1) + torch.softmax(outputs_u12, dim=1) + torch.softmax(outputs_u21, dim=1) + torch.softmax(outputs_u22, dim=1)) / 4
ptu = pu**(1/args.T) ## Temparature Sharpening
targets_u = ptu / ptu.sum(dim=1, keepdim=True) ## Normalize
targets_u = targets_u.detach()
## Label refinement of labeled samples
_, outputs_x = net(inputs_x3)
_, outputs_x2 = net(inputs_x4)
px = (torch.softmax(outputs_x, dim=1) + torch.softmax(outputs_x2, dim=1)) / 2
px = w_x*labels_x + (1-w_x)*px
ptx = px**(1/args.T) ## Temparature sharpening
targets_x = ptx / ptx.sum(dim=1, keepdim=True) ## normalize
targets_x = targets_x.detach()
## Mixmatch
l = np.random.beta(args.alpha, args.alpha)
l = max(l,1-l)
## Unsupervised Contrastive Loss
f1, _ = net(inputs_u)
f2, _ = net(inputs_u2)
f1 = F.normalize(f1, dim=1)
f2 = F.normalize(f2, dim=1)
features = torch.cat([f1.unsqueeze(1), f2.unsqueeze(1)], dim=1)
loss_simCLR = contrastive_criterion(features)
all_inputs = torch.cat([inputs_x, inputs_x2, inputs_u, inputs_u2], dim=0)
all_targets = torch.cat([targets_x, targets_x, targets_u, targets_u], dim=0)
idx = torch.randperm(all_inputs.size(0))
input_a , input_b = all_inputs , all_inputs[idx]
target_a, target_b = all_targets, all_targets[idx]
## Mixing inputs
mixed_input = (l * input_a[: batch_size * 2] + (1 - l) * input_b[: batch_size * 2])
mixed_target = (l * target_a[: batch_size * 2] + (1 - l) * target_b[: batch_size * 2])
_, logits = net(mixed_input)
Lx = -torch.mean(
torch.sum(F.log_softmax(logits, dim=1) * mixed_target, dim=1)
)
## Regularization
prior = torch.ones(args.num_class) / args.num_class
prior = prior.cuda()
pred_mean = torch.softmax(logits, dim=1).mean(0)
penalty = torch.sum(prior * torch.log(prior / pred_mean))
loss = Lx + args.lambda_c*loss_simCLR + penalty
loss_x += Lx.item()
loss_ucl += loss_simCLR.item()
## Compute gradient and do SGD step
optimizer.zero_grad()
loss.backward()
optimizer.step()
sys.stdout.write('\r')
sys.stdout.write('%s: | Epoch [%3d/%3d] Iter[%3d/%3d]\t Labeled loss: %.2f Contrative Loss:%.4f'
%(args.dataset, epoch, args.num_epochs, batch_idx+1, num_iter, loss_x/(batch_idx+1), loss_ucl/(batch_idx+1)))
sys.stdout.flush()
def warmup(net,optimizer,dataloader):
net.train()
for batch_idx, (inputs, labels, path) in enumerate(dataloader):
inputs, labels = inputs.cuda(), labels.cuda()
optimizer.zero_grad()
_ , outputs = net(inputs)
loss = CEloss(outputs, labels)
penalty = conf_penalty(outputs)
L = loss + penalty
L.backward()
optimizer.step()
sys.stdout.write('\r')
sys.stdout.write('|Warm-up: Iter[%3d/%3d]\t CE-loss: %.4f Conf-Penalty: %.4f'
%(batch_idx+1, args.num_batches, loss.item(), penalty.item()))
sys.stdout.flush()
def val(net,val_loader,k):
net.eval()
correct = 0
total = 0
with torch.no_grad():
for batch_idx, (inputs, targets) in enumerate(val_loader):
inputs, targets = inputs.cuda(), targets.cuda()
_ , outputs = net(inputs)
_ , predicted = torch.max(outputs, 1)
total += targets.size(0)
correct += predicted.eq(targets).cpu().sum().item()
acc = 100.*correct/total
print("\n| Validation\t Net%d Acc: %.2f%%" %(k,acc))
if acc > best_acc[k-1]:
best_acc[k-1] = acc
print('| Saving Best Net%d ...'%k)
save_point = os.path.join(model_save_loc, '%s_net%d.pth.tar'%(args.id,k))
torch.save(net.state_dict(), save_point)
return acc
def test(net1,net2,test_loader):
acc_meter.reset()
net1.eval()
net2.eval()
correct = 0
total = 0
with torch.no_grad():
for batch_idx, (inputs, targets) in enumerate(test_loader):
inputs, targets = inputs.cuda(), targets.cuda()
_, outputs1 = net1(inputs)
_, outputs2 = net2(inputs)
outputs = outputs1+outputs2
_, predicted = torch.max(outputs, 1)
total += targets.size(0)
correct += predicted.eq(targets).cpu().sum().item()
acc_meter.add(outputs,targets)
acc = 100.*correct/total
print("\n| Test Acc: %.2f%%\n" %(acc))
accs = acc_meter.value()
return acc , accs
## Calculate the KL Divergence
def kl_divergence(p, q):
return (p * ((p+1e-10) / (q+1e-10)).log()).sum(dim=1)
## Jensen_Shannon divergence (Symmetric and Smoother than the KL divergence)
class Jensen_Shannon(nn.Module):
def __init__(self):
super(Jensen_Shannon,self).__init__()
pass
def forward(self, p,q):
m = (p+q)/2
return 0.5*kl_divergence(p, m) + 0.5*kl_divergence(q, m)
## Calculate JSD
def Calculate_JSD(epoch,model1, model2):
model1.eval()
model2.eval()
num_samples = args.num_batches*args.batch_size
prob = torch.zeros(num_samples)
JS_dist = Jensen_Shannon()
paths = []
n=0
for batch_idx, (inputs, targets, path) in enumerate(eval_loader):
inputs, targets = inputs.cuda(), targets.cuda()
batch_size = inputs.size()[0]
## Get the output of the Model
with torch.no_grad():
out1 = torch.nn.Softmax(dim=1).cuda()(model1(inputs)[1])
out2 = torch.nn.Softmax(dim=1).cuda()(model2(inputs)[1])
## Get the Prediction
out = 0.5*out1 + 0.5*out2
## Divergence clculator to record the diff. between ground truth and output prob. dist.
dist = JS_dist(out, F.one_hot(targets, num_classes = args.num_class))
prob[int(batch_idx*batch_size):int((batch_idx+1)*batch_size)] = dist
for b in range(inputs.size(0)):
paths.append(path[b])
n+=1
sys.stdout.write('\r')
sys.stdout.write('| Evaluating loss Iter %3d\t' %(batch_idx))
sys.stdout.flush()
return prob,paths
## Penalty for Asymmetric Noise
class NegEntropy(object):
def __call__(self,outputs):
probs = torch.softmax(outputs, dim=1)
return torch.mean(torch.sum(probs.log()*probs, dim=1))
## Get the pre-trained model
def get_model():
model = models.resnet50(pretrained=True)
model.fc = nn.Linear(2048, args.num_class)
return model
def create_model():
model = resnet50(num_classes=args.num_class)
model = model.cuda()
return model
## Threshold Adjustment
def linear_rampup(current, warm_up, rampup_length=5):
current = np.clip((current-warm_up) / rampup_length, 0.0, 1.0)
return args.lambda_u*float(current)
## Semi-Supervised Loss
class SemiLoss(object):
def __call__(self, outputs_x, targets_x, outputs_u, targets_u, epoch, warm_up):
probs_u = torch.softmax(outputs_u, dim=1)
Lx = -torch.mean(torch.sum(F.log_softmax(outputs_x, dim=1) * targets_x, dim=1))
Lu = torch.mean((probs_u - targets_u)**2)
return Lx, Lu, linear_rampup(epoch,warm_up)
log = open('./checkpoint/%s.txt'%args.id,'w')
log.flush()
loader = dataloader.clothing_dataloader(root=args.data_path, batch_size=args.batch_size, warmup_batch_size = args.batch_size*2, num_workers=8, num_batches=args.num_batches)
print('| Building Net')
model = get_model()
net1 = create_model()
net2 = create_model()
cudnn.benchmark = True
## Optimizer and Learning Rate Scheduler
optimizer1 = optim.SGD(net1.parameters(), lr=args.lr, momentum=0.9, weight_decay=1e-3)
optimizer2 = optim.SGD(net2.parameters(), lr=args.lr, momentum=0.9, weight_decay=1e-3)
scheduler1 = optim.lr_scheduler.CosineAnnealingLR(optimizer1, 100, 1e-5)
scheduler2 = optim.lr_scheduler.CosineAnnealingLR(optimizer2, 100, 1e-5)
## Cross-Entropy and Other Losses
CE = nn.CrossEntropyLoss(reduction='none')
CEloss = nn.CrossEntropyLoss()
conf_penalty = NegEntropy()
criterion = SemiLoss()
## Warm-up Epochs (maximum value is 2, we recommend 0 or 1)
warm_up = 0
## Copy Saved Data
if args.pretrained:
params = model.named_parameters()
params1 = net1.named_parameters()
params2 = net2.named_parameters()
dict_params2 = dict(params2)
dict_params1 = dict(params1)
for name1, param in params:
if name1 in dict_params2:
dict_params2[name1].data.copy_(param.data)
dict_params1[name1].data.copy_(param.data)
## Location for saving the models
folder = 'Clothing1M'
model_save_loc = './checkpoint/' + folder
if not os.path.exists(model_save_loc):
os.mkdir(model_save_loc)
net1 = nn.DataParallel(net1)
net2 = nn.DataParallel(net2)
## Loading Saved Weights
model_name_1 = 'clothing1m_net1.pth.tar'
model_name_2 = 'clothing1m_net2.pth.tar'
if args.resume:
net1.load_state_dict(torch.load(os.path.join(model_save_loc, model_name_1)))
net2.load_state_dict(torch.load(os.path.join(model_save_loc, model_name_2)))
best_acc = [0,0]
SR = 0
torch.backends.cudnn.benchmark = True
acc_meter = torchnet.meter.ClassErrorMeter(topk=[1,5], accuracy=True)
nb_repeat = 2
for epoch in range(0, args.num_epochs+1):
val_loader = loader.run(0, 'val')
if epoch>100:
nb_repeat =3 ## Change how many times we want to repeat on the same selection
if epoch<warm_up:
train_loader = loader.run(0,'warmup')
print('Warmup Net1')
warmup(net1,optimizer1,train_loader)
print('\nWarmup Net2')
train_loader = loader.run(0,'warmup')
warmup(net2, optimizer2, train_loader)
else:
num_samples = args.num_batches*args.batch_size
eval_loader = loader.run(0.5,'eval_train')
prob2, paths2 = Calculate_JSD(epoch, net1, net2) ## Calculate the JSD distances
threshold = torch.mean(prob2) ## Simply Take the average as the threshold
if threshold.item()>args.d_u:
threshold = threshold - (threshold-torch.min(JSD))/args.tau
SR = torch.sum(prob2<threshold).item()/prob2.size()[0] ## Calculate the Ratio of clean samples
for i in range(nb_repeat):
print('\n\nTrain Net1')
labeled_trainloader, unlabeled_trainloader = loader.run(SR, 'train', prob=prob2, paths=paths2) ## Uniform Selection
train(epoch, net1, net2, optimizer1, labeled_trainloader, unlabeled_trainloader) ## Train Net1
acc1 = val(net1,val_loader,1)
print('\n==== Net 1 evaluate next epoch training data loss ====')
eval_loader = loader.run(SR,'eval_train')
net1.load_state_dict(torch.load(os.path.join(model_save_loc, '%s_net1.pth.tar'%args.id)))
prob1, paths1 = Calculate_JSD(epoch,net2, net1)
threshold = torch.mean(prob1)
if threshold.item()>args.d_u:
threshold = threshold - (threshold-torch.min(JSD))/args.tau
SR = torch.sum(prob1<threshold).item()/prob1.size()[0]
for i in range(nb_repeat):
print('\nTrain Net2')
labeled_trainloader, unlabeled_trainloader = loader.run(SR, 'train', prob=prob1, paths=paths1) ## Uniform Selection
train(epoch,net2,net1,optimizer2,labeled_trainloader, unlabeled_trainloader) ## Train net2
acc2 = val(net2,val_loader,2)
scheduler1.step()
scheduler2.step()
acc1 = val(net1,val_loader,1)
acc2 = val(net2,val_loader,2)
log.write('Validation Epoch:%d Acc1:%.2f Acc2:%.2f\n'%(epoch,acc1,acc2))
net1.load_state_dict(torch.load(os.path.join(model_save_loc, '%s_net1.pth.tar'%args.id)))
net2.load_state_dict(torch.load(os.path.join(model_save_loc, '%s_net2.pth.tar'%args.id)))
log.flush()
test_loader = loader.run(0,'test')
acc, accs = test(net1,net2,test_loader)
print('\n| Epoch:%d \t Acc: %.2f%% (%.2f%%) \n'%(epoch,accs[0],accs[1]))
log.write('Epoch:%d \t Acc: %.2f%% (%.2f%%) \n'%(epoch,accs[0],accs[1]))
log.flush()
if epoch<warm_up:
model_name_1 = 'Net1_warmup_pretrained.pth'
model_name_2 = 'Net2_warmup_pretrained.pth'
print("Save the Warmup Model --- --")
checkpoint1 = {
'net': net1.state_dict(),
'Model_number': 1,
'epoch': epoch,
}
checkpoint2 = {
'net': net2.state_dict(),
'Model_number': 2,
'epoch': epoch,
}
torch.save(checkpoint1, os.path.join(model_save_loc, model_name_1))
torch.save(checkpoint2, os.path.join(model_save_loc, model_name_2))