-
Notifications
You must be signed in to change notification settings - Fork 1
/
main-wgan-gp.py
369 lines (322 loc) · 14.6 KB
/
main-wgan-gp.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
import argparse, os
import pdb
import torch
import math, random
import torch.backends.cudnn as cudnn
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
from torch.utils.data import DataLoader
from torch import autograd
from model_wgan_bicubic64 import _netGL, _netDL,_netGH, _netDH, L1_Charbonnier_loss
from dataset import FastLoader
from torchvision import models, transforms
import torch.utils.model_zoo as model_zoo
import torch.nn.functional as F
import utils
import time
# Training settings
parser = argparse.ArgumentParser(description="PyTorch DGGAN WGAN")
parser.add_argument("--batchSize", type=int, default=64, help="training batch size")
parser.add_argument("--nEpochs", type=int, default=100, help="number of epochs to train for")
parser.add_argument('--lrG', type=float, default=1e-4, help='Learning Rate. Default=1e-4')
parser.add_argument('--lrD', type=float, default=1e-4, help='Learning Rate. Default=1e-4')
parser.add_argument("--step", type=int, default=60, help="Sets the learning rate to the initial LR decayed by momentum every n epochs, Default: n=10")
parser.add_argument("--cuda", action="store_true", help="Use cuda?")
parser.add_argument("--resume", default="", type=str, help="Path to checkpoint (default: none)")
parser.add_argument("--start-epoch", default=1, type=int, help="Manual epoch number (useful on restarts)")
parser.add_argument("--threads", type=int, default=32, help="Number of threads for data loader to use, Default: 1")
parser.add_argument("--momentum", default=0.9, type=float, help="Momentum, Default: 0.9")
parser.add_argument("--weight-decay", "--wd", default=1e-4, type=float, help="weight decay, Default: 1e-4")
parser.add_argument("--pretrained", default="", type=str, help="path to pretrained model (default: none)")
parser.add_argument('--clamp_lower', type=float, default=-0.01)
parser.add_argument('--clamp_upper', type=float, default=0.01)
def main(dataname,dataid):
global opt, model
opt = parser.parse_args()
print(opt)
cuda = opt.cuda
if cuda and not torch.cuda.is_available():
raise Exception("No GPU found, please run without --cuda")
opt.seed = random.randint(1, 10000)
print("Random Seed: ", opt.seed)
torch.manual_seed(opt.seed)
if cuda:
torch.cuda.manual_seed(opt.seed)
cudnn.benchmark = True
print("===> Loading datasets")
train_set = FastLoader('../../datasets/NYU/pkls/nyu_%d.pkl'%dataid)
training_data_loader = DataLoader(dataset=train_set, num_workers=opt.threads, batch_size=opt.batchSize, shuffle=True)
print('===> Building generator model')
netG_HR = _netGH()
netG_LR = _netGL()
print('===> Building discriminator model')
netD_HR = _netDH()
netD_LR = _netDL()
print('===> Loading VGG model')
model_urls = {
"vgg19": "https://download.pytorch.org/models/vgg19-dcbb9e9d.pth"
}
netVGG = models.vgg19()
netVGG.load_state_dict(model_zoo.load_url(model_urls['vgg19']))
weight = torch.FloatTensor(64,1,3,3)
parameters = list(netVGG.parameters())
for i in range(64):
weight[i,:,:,:] = parameters[0].data[i].mean(0)
bias = parameters[1].data
class _content_model(nn.Module):
def __init__(self):
super(_content_model, self).__init__()
self.conv = conv2d = nn.Conv2d(1, 64, kernel_size=3, padding=1)
self.feature = nn.Sequential(*list(netVGG.features.children())[1:-1])
self._initialize_weights()
def forward(self, x):
out = self.conv(x)
out = self.feature(out)
return out
def _initialize_weights(self):
self.conv.weight.data.copy_(weight)
self.conv.bias.data.copy_(bias)
netContent = _content_model()
print('===> Building Loss')
criterion = L1_Charbonnier_loss()
print("===> Setting GPU")
if cuda:
netG_HR = netG_HR.cuda()
netD_HR = netD_HR.cuda()
netG_LR = netG_LR.cuda()
netD_LR = netD_LR.cuda()
netContent = netContent.cuda()
criterion = criterion.cuda()
# optionally resume from a checkpoint
if opt.resume:
if os.path.isfile(opt.resume):
print("=> loading checkpoint '{}'".format(opt.resume))
checkpoint = torch.load(opt.resume)
opt.start_epoch = checkpoint["epoch"] + 1
netG_HR.load_state_dict(checkpoint["model"].state_dict())
else:
print("=> no checkpoint found at '{}'".format(opt.resume))
# optionally copy weights from a checkpoint
if opt.pretrained:
if os.path.isfile(opt.pretrained):
print("=> loading model '{}'".format(opt.pretrained))
weights = torch.load(opt.pretrained)
netG_HR.load_state_dict(weights['model'].state_dict())
else:
print("=> no model found at '{}'".format(opt.pretrained))
print("===> Setting Optimizer")
optimizerDL = optim.RMSprop(netD_LR.parameters(), lr = opt.lrD)
optimizerDH = optim.RMSprop(netD_HR.parameters(), lr = opt.lrD)
optimizerGL = optim.RMSprop(netG_LR.parameters(), lr = opt.lrG)
optimizerGH = optim.RMSprop(netG_HR.parameters(), lr = opt.lrG)
print("===> Training")
for epoch in range(opt.start_epoch, opt.nEpochs + 1):
e_start_time = time.time()
train(training_data_loader, optimizerGH, optimizerDH,optimizerGL, optimizerDL, netG_HR, netD_HR,netG_LR, netD_LR, netContent, criterion, epoch)
save_checkpoint(netG_HR, epoch, dataid)
e_end_time = time.time()
elapse_time = e_end_time-e_start_time
print('Time = %.5fs/Epoch'%elapse_time)
def adjust_learning_rate(optimizer, epoch, type):
"""Sets the learning rate to the initial LR decayed by 10 every 10 epochs"""
if type == 'G':
lr = opt.lrG * (0.1 ** (epoch // opt.step))
elif type == 'D':
lr = opt.lrD * (0.1 ** (epoch // opt.step))
return lr
def calc_gradient_penalty(netD, real_data, fake_data):
# print "real_data: ", real_data.size(), fake_data.size()
BATCH_SIZE = real_data.shape[0]
LAMBDA = 10
alpha = torch.rand(BATCH_SIZE, 1)
ag2 = real_data.nelement()//BATCH_SIZE
# ag2 must be int
alpha = alpha.expand(BATCH_SIZE, ag2)
alpha = alpha.contiguous()
# alpha = alpha.view(BATCH_SIZE, 3, 32, 32)
# single channel
# print(real_data.shape)
# print(fake_data.shape)
h,w = real_data.shape[2],real_data.shape[3]
alpha = alpha.view(BATCH_SIZE, 1, h, w)
if opt.cuda:
alpha = alpha.cuda()
interpolates = alpha * real_data + ((1 - alpha) * fake_data)
if opt.cuda:
interpolates = interpolates.cuda()
interpolates = autograd.Variable(interpolates, requires_grad=True)
disc_interpolates = netD(interpolates)
gradients = autograd.grad(outputs=disc_interpolates, inputs=interpolates,
grad_outputs=torch.ones(disc_interpolates.size()).cuda() if opt.cuda else torch.ones(
disc_interpolates.size()),
create_graph=True, retain_graph=True, only_inputs=True)[0]
gradients = gradients.view(gradients.size(0), -1)
gradient_penalty = ((gradients.norm(2, dim=1) - 1) ** 2).mean() * LAMBDA
return gradient_penalty
def train(training_data_loader, optimizerGH, optimizerDH,optimizerGL, optimizerDL, netG_HR, netD_HR,netG_LR, netD_LR, netContent, criterion, epoch):
lrGH = adjust_learning_rate(optimizerGH, epoch-1,'G')
lrGL = adjust_learning_rate(optimizerGL, epoch-1,'G')
lrDH = adjust_learning_rate(optimizerDH, epoch-1,'D')
lrDL = adjust_learning_rate(optimizerDL, epoch-1,'D')
for param_group in optimizerGH.param_groups:
param_group["lr"] = lrGH
for param_group in optimizerGL.param_groups:
param_group["lr"] = lrGL
for param_group in optimizerDH.param_groups:
param_group["lr"] = lrDH
for param_group in optimizerDL.param_groups:
param_group["lr"] = lrDL
netG_HR.train()
netD_HR.train()
netG_LR.train()
netD_LR.train()
one = torch.FloatTensor([1.])
mone = one * -1
content_weight = torch.FloatTensor([0.1])
# content_weight = torch.FloatTensor([1.])
adversarial_weight = torch.FloatTensor([1.])
# adversarial_weight = torch.FloatTensor([0.01])
for iteration, batch in enumerate(training_data_loader, 1):
LR, HR, Ld, Hd = batch[0], batch[1], batch[2], batch[3]
# print(HR.shape,LR.shape)
d_ratio = (Hd/Ld).mean()
d_scale = torch.sigmoid(d_ratio)
# d_scale = F.tanh(d_ratio/4)
# d_scale = F.softmax(d_ratio)
# d_scale = d_ratio/4
# print(d_scale)
# 用dscale约束每次迭代的参数更新幅度
# for param_group in optimizerGH.param_groups:
# param_group["lr"] = lrGH*d_scale
# for param_group in optimizerGL.param_groups:
# param_group["lr"] = lrGL*d_scale
# for param_group in optimizerDH.param_groups:
# param_group["lr"] = lrDH*d_scale
# for param_group in optimizerDL.param_groups:
# param_group["lr"] = lrDL*d_scale
if opt.cuda:
LR = LR.cuda()
HR = HR.cuda()
one, mone, content_weight, adversarial_weight = one.cuda(), mone.cuda(), content_weight.cuda(), adversarial_weight.cuda()
############################
# (1) Update D network: loss = D(x)) - D(G(z))
###########################
# netD_HR --->fake HR and real HR
# netD_LR --->fake LR and real LR
# netG_HR ---> fake LR--> fake HR
# netG_LR ---> HR---> fake LR
errD_real_l = netD_LR(LR)
errD_real_l.backward(one, retain_graph=True)
input_G_hr = HR.data
fake_LR = netG_LR(input_G_hr).data
errD_fake_l = netD_LR(fake_LR)
errD_fake_l.backward(mone)
# train with gradient penalty
gradient_penalty_l = calc_gradient_penalty(netD_LR, LR.data, fake_LR)
gradient_penalty_l.backward()
errD_l = errD_real_l - errD_fake_l + gradient_penalty_l
Wasserstein_D_l = errD_real_l - errD_fake_l
optimizerDL.step()
# for p in netD_LR.parameters(): # reset requires_grad
# p.data.clamp_(opt.clamp_lower, opt.clamp_upper)
netD_LR.zero_grad()
netG_LR.zero_grad()
netContent.zero_grad()
# train with real
errD_real_h = netD_HR(HR)
errD_real_h.backward(one, retain_graph=True)
# train with fake
input_G_lr = fake_LR.data
fake_HR = netG_HR(input_G_lr).data
errD_fake_h = netD_HR(fake_HR)
errD_fake_h.backward(mone)
gradient_penalty_h = calc_gradient_penalty(netD_HR, HR.data, fake_HR)
gradient_penalty_h.backward()
errD_h = errD_real_h - errD_fake_h + gradient_penalty_h
Wasserstein_D_h = errD_real_h - errD_fake_h
optimizerDH.step()
# for p in netD_HR.parameters(): # reset requires_grad
# p.data.clamp_(opt.clamp_lower, opt.clamp_upper)
netD_HR.zero_grad()
netG_HR.zero_grad()
netContent.zero_grad()
############################
# (2) Update G network: loss = D(G(z))
###########################
#----------Part1 Downsample-------------
fake_D_lr = netG_LR(HR)
# content_fake_lr = netContent(fake_D_lr)
# content_real_lr = netContent(LR)
# content_real_lr = Variable(content_real_lr.data)
# content loss for this task???
# content_loss_lr = criterion(content_fake_lr, content_real_lr)
# content_loss_lr.backward(content_weight, retain_graph=True)
# content_loss_l = content_loss_lr
adversarial_loss_l = netD_LR(fake_D_lr)
adversarial_loss_l.backward(adversarial_weight)
optimizerGL.step()
netD_LR.zero_grad()
netG_LR.zero_grad()
# netContent.zero_grad()
#---- cycle 1 distribution up_real_lr-->real HR
fake_D_lr_up = netG_HR(LR)
adversarial_loss_l_up = netD_HR(fake_D_lr_up)
adversarial_loss_l_up.backward(adversarial_weight)
optimizerGH.step()
netD_HR.zero_grad()
netG_HR.zero_grad()
#--------Part2 Upsample-----------
fake_D_lr = netG_LR(HR)
fake_D_x2 = netG_HR(fake_D_lr)
content_fake_x2 = netContent(fake_D_x2)
content_real_x2 = netContent(HR)
content_real_x2 = Variable(content_real_x2.data)
content_loss_x2 = criterion(content_fake_x2, content_real_x2)
content_loss_x2.backward(content_weight, retain_graph=True)
content_loss = content_loss_x2
adversarial_loss = netD_HR(fake_D_x2)
lossG = adversarial_loss+content_loss
adversarial_loss.backward(adversarial_weight)
optimizerGH.step()
netD_HR.zero_grad()
netG_HR.zero_grad()
netContent.zero_grad()
#---- cycle 2 fake HR->fake LR
fake_D_lr = netG_LR(HR)
fake_D_x2 = netG_HR(fake_D_lr)
fake_D_hr_down = netG_LR(fake_D_x2)
content_fake_hr_down = netContent(fake_D_hr_down)
content_real_lr = netContent(fake_D_lr)
content_real_lr = Variable(content_real_lr.data)
content_loss_down_lr = criterion(content_fake_hr_down, content_real_lr)
content_loss_down_lr.backward(content_weight, retain_graph=True)
content_loss_dl = content_loss_down_lr
adversarial_loss_dl = netD_LR(fake_D_hr_down)
adversarial_loss_dl.backward(adversarial_weight)
optimizerGL.step()
netD_LR.zero_grad()
netG_LR.zero_grad()
netContent.zero_grad()
# print network and loss
if iteration%5 == 0:
print("===> Epoch[{}]({}/{}): LossD: {:.5f} [{:.5f} - {:.5f}] LossG: {:.5f} [{:.5f} + {:.5f}]".format(
epoch, iteration, len(training_data_loader),
errD_h.item(), errD_real_h.item(), errD_fake_h.item(), lossG.item(), adversarial_loss.item(), content_loss.item()))
def save_checkpoint(model, epoch,dataid):
model_folder = './86_zssr/model_%d/'%dataid
model_out_path = model_folder + "biwgan_model_epoch_{}.pth".format(epoch)
state = {"epoch": epoch ,"model": model}
if not os.path.exists(model_folder):
os.makedirs(model_folder)
torch.save(state, model_out_path)
print("Checkpoint saved to {}".format(model_out_path))
if __name__ == "__main__":
utils.save_experiment()
# ims = [98,80,3,7,5,1,15,2,4,40,28,49,8,26,27,65,11,81,91,55,67,22,25,85,57,52,48,61,83,62,56,45,
# 50,44,18,35,43,97,86,0,88,58,30,87,78,74,99,29,75,64]
ims = [50,44,18,35,43,97,86,0,88,58,30,87,78,74,99,29,75,64]
# for i in range(86,87):
# main('NYU',i)
for i in range(len(ims)):
main('NYU',ims[i])