-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathtrain.py
297 lines (257 loc) · 14.4 KB
/
train.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
import os
import argparse
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import torchvision
import torchvision.utils as utils
from torch.autograd import Variable
from torch.utils.data import DataLoader
from tensorboardX import SummaryWriter
from model_rdn import Generator_RDN, Discriminator, FeatureExtractor
from loss import GeneratorLoss
#from dataset import prepare_data, Dataset
from datasets import TrainDataset, EvalDataset
from skimage.io import imsave
from my_utils import *
import sys
import cv2
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
parser = argparse.ArgumentParser(description="SRGAN-RDN")
parser.add_argument("--preprocess", type=bool, default=False, help='run prepare_data or not')
parser.add_argument('--upscale_factor', type=str, default='3', help='super resolution scale')
parser.add_argument('--patch_size', type=int, default=32, help='output patch size')
parser.add_argument('--G0', type=int, default=64, help='default number of filters. (Use in RDN)')
parser.add_argument('--RDNkSize', type=int, default=3, help='default kernel size. (Use in RDN)')
parser.add_argument('--RDNconfig', type=str, default='B', help='parameters config of RDN. (Use in RDN)')
parser.add_argument("--batchSize", type=int, default=16, help="Training batch size")
parser.add_argument("--num_of_layers", type=int, default=17, help="Number of total layers")
parser.add_argument("--epochs", type=int, default=150, help="Number of training epochs")
parser.add_argument("--milestone", type=int, default=30, help="When to decay learning rate; should be less than epochs")
parser.add_argument('--generatorLR', type=float, default=0.0001, help='learning rate for generator')
parser.add_argument('--discriminatorLR', type=float, default=0.0001, help='learning rate for discriminator')
parser.add_argument("--outf", type=str, default="DIV/logs_sr_rdgan", help='path of log files')
parser.add_argument("--mode", type=str, default="S", help='with known noise level (S) or blind training (B)')
parser.add_argument("--noiseL", type=float, default=25, help='noise level; ignored when mode=B')
parser.add_argument("--val_noiseL", type=float, default=25, help='noise level used on validation set')
parser.add_argument('--n_colors', type=int, default=3, help='number of color channels to use')
parser.add_argument('--cuda', type=bool, default=True, help='enable cuda')
opt = parser.parse_args()
def save_result(result,path):
path = path if path.find('.') != -1 else path+'.png'
imsave(path, np.clip(result,0,1))
def make_grid(images,nrow,nline):
rows=[]
for i in range(nrow):
tmp=images[i]
for j in range(nline-1):
tmp=np.vstack((tmp,images[(j+1)*nrow+i]))
rows.append(tmp)
tmp=rows[0]
for i in range(nrow-1):
tmp=np.hstack((tmp,rows[i+1]))
return tmp
def main():
# Load dataset
print('Loading dataset ...\n')
#dataset_train = Dataset(train=True)
#dataset_val = Dataset(train=False)
dataset_train = TrainDataset('train_DIV_new.h5', opt.patch_size, int(opt.upscale_factor[0]))
dataset_val = EvalDataset('test_DIV.h5')
loader_train = DataLoader(dataset=dataset_train, num_workers=1, batch_size=opt.batchSize, shuffle=True)
loader_val = DataLoader(dataset=dataset_val, batch_size=1)
print("# of training samples: %d\n" % int(len(dataset_train)))
# Build model
netG = Generator_RDN(opt)
print('# generator parameters:', sum(param.numel() for param in netG.parameters()))
netD = Discriminator()
print('# discriminator parameters:', sum(param.numel() for param in netD.parameters()))
# net.apply(weights_init_kaiming)
# content_criterion = nn.MSELoss()
# feature_extractor = FeatureExtractor(torchvision.models.vgg19(pretrained=True))
content_criterion = nn.L1Loss()
adversarial_criterion = nn.BCELoss()
ones_const = Variable(torch.ones(opt.batchSize, 1))
# Move to GPU
if torch.cuda.is_available():
netG.cuda()
netD.cuda()
content_criterion.cuda()
adversarial_criterion.cuda()
ones_const = ones_const.cuda()
# feature_extractor.cuda()
optim_rdn = optim.Adam(netG.parameters(), lr=1e-4)
# Optimizer
# training
writer = SummaryWriter(opt.outf)
step = 0
# noiseL_B=[0,55] # ingnored when opt.mode=='S'
# Generator Pretraining(Using MSE Loss)
for epoch in range(10):
mean_generator_content_loss = 0.0
for i, (lrimg, hrimg) in enumerate(loader_train):
# adding noise
#print(lrimg[-1].shape)
#print(hrimg[-1].shape)
#cv2.imshow('win1',lrimg[-1].detach().numpy().transpose((1,2,0)))
#cv2.imshow('win2',hrimg[-1].detach().numpy().transpose((1,2,0)))
#cv2.waitKey(0)
for j in range(opt.batchSize):
noise = torch.FloatTensor(lrimg[j].size()).normal_(mean=0.0, std=opt.noiseL/255.)
#lrimg[j] = lrimg[j] + noise
lrimg[j] = lrimg[j]
# Generate real and fake inputs
if opt.cuda:
high_res_real = Variable(hrimg.cuda())
high_res_fake = netG(Variable(lrimg).cuda())
else:
high_res_real = Variable(hrimg)
high_res_fake = netG(Variable(lrimg))
######### Train generator #########
netG.zero_grad()
generator_content_loss = content_criterion(high_res_fake, high_res_real)
mean_generator_content_loss += generator_content_loss.data
generator_content_loss.backward()
optim_rdn.step()
######### Status and display #########
# sys.stdout.write('\r[%d/%d][%d/%d] Generator_MSE_Loss: %.4f' % (epoch, 2, i, len(loader_train), generator_content_loss.data))
# visualizer.show(low_res, high_res_real.cpu().data, high_res_fake.cpu().data)
out_train = torch.clamp(high_res_fake, 0., 1.)
psnr_train, ssim_train = batch_PSNR(out_train, high_res_real, scale=3.0, data_range=1.)
if step % 10 == 0:
# Log the scalar values
writer.add_scalar('generator_content_loss', generator_content_loss.item(), step)
writer.add_scalar('PSNR on training data', psnr_train, step)
step += 1
# sys.stdout.write('\r[%d/%d][%d/%d] PSNR: %.4f, SSIM:%.4f' % (epoch, 2, i, len(loader_train), psnr_train, ssim_train))
sys.stdout.write('\r[%d/%d][%d/%d] Generator_MSE_Loss: %.4f\n' % (epoch, 2, i, len(loader_train), mean_generator_content_loss/len(loader_train)))
#log_value('generator_mse_loss', mean_generator_content_loss/len(dataloader), epoch)
# Do checkpointing
torch.save(netG.state_dict(), '%s/model/generator_pretrain.pth' % opt.outf)
#SRGAN-RDN training
optim_generator = optim.Adam(netG.parameters(), lr=opt.generatorLR)
optim_discriminator = optim.Adam(netD.parameters(), lr=opt.discriminatorLR)
scheduler_dis = torch.optim.lr_scheduler.StepLR(optim_discriminator,50,0.1)
scheduler_gen = torch.optim.lr_scheduler.StepLR(optim_generator,50,0.1)
print('SRGAN training')
step_new = 0
for epoch in range(opt.epochs):
mean_generator_content_loss = 0.0
mean_generator_adversarial_loss = 0.0
mean_generator_total_loss = 0.0
mean_discriminator_loss = 0.0
netG.train()
scheduler_gen.step()
scheduler_dis.step()
for i, (lrimg, hrimg) in enumerate(loader_train):
#print(lrimg[-1].shape)
#print(hrimg[-1].shape)
#cv2.imshow('win1',lrimg[-1].detach().numpy().transpose((1,2,0)))
#cv2.imshow('win2',hrimg[-1].detach().numpy().transpose((1,2,0)))
#cv2.waitKey(0)
for j in range(opt.batchSize):
noise = torch.FloatTensor(lrimg[j].size()).normal_(mean=0, std=opt.noiseL/255.)
#lrimg[j] = lrimg[j] + noise
lrimg[j] = lrimg[j]
# Generate real and fake inputs
if opt.cuda:
high_res_real = Variable(hrimg.cuda())
high_res_fake = netG(Variable(lrimg).cuda())
target_real = Variable(torch.rand(opt.batchSize,1)*0.5 + 0.7).cuda()
target_fake = Variable(torch.rand(opt.batchSize,1)*0.3).cuda()
else:
high_res_real = Variable(hrimg)
high_res_fake = netG(Variable(lrimg))
target_real = Variable(torch.rand(opt.batchSize,1)*0.5 + 0.7)
target_fake = Variable(torch.rand(opt.batchSize,1)*0.3)
######### Train discriminator #########
netD.zero_grad()
discriminator_loss = adversarial_criterion(netD(high_res_real), target_real) + \
adversarial_criterion(netD(Variable(high_res_fake.data)), target_fake)
mean_discriminator_loss += discriminator_loss.data
discriminator_loss.backward()
optim_discriminator.step()
######### Train generator #########
netG.zero_grad()
#real_features = Variable(feature_extractor(high_res_real).data)
#fake_features = feature_extractor(high_res_fake)
generator_content_loss = content_criterion(high_res_fake, high_res_real) #+ 0.006*content_criterion(fake_features, real_features)
mean_generator_content_loss += generator_content_loss.data
generator_adversarial_loss = adversarial_criterion(netD(high_res_fake), ones_const)
mean_generator_adversarial_loss += generator_adversarial_loss.data
generator_total_loss = generator_content_loss + 1e-3*generator_adversarial_loss
mean_generator_total_loss += generator_total_loss.data
generator_total_loss.backward()
optim_generator.step()
######### Status and display #########
sys.stdout.write('\r[%d/%d][%d/%d] Discriminator_Loss: %.4f Generator_Loss (Content/Advers/Total): %.4f/%.4f/%.4f' % (epoch, opt.epochs, i, len(loader_train),
discriminator_loss.data, generator_content_loss.data, generator_adversarial_loss.data, generator_total_loss.data))
# visualizer.show(low_res, high_res_real.cpu().data, high_res_fake.cpu().data)
out_train = torch.clamp(high_res_fake, 0., 1.)
psnr_train, ssim_train = batch_PSNR(out_train, high_res_real, scale=3.0, data_range=1.)
if step_new % 10 == 0:
# Log the scalar values
writer.add_scalar('generator_content_loss', generator_content_loss.item(), step)
writer.add_scalar('PSNR on training data', psnr_train, step)
writer.add_scalar('discriminator_loss', discriminator_loss.data, step_new)
writer.add_scalar('generator_adversarial_loss', generator_adversarial_loss.item(), step_new)
writer.add_scalar('generator_total_loss', generator_total_loss, step_new)
step += 1
step_new += 1
sys.stdout.write('\r[%d/%d][%d/%d] Discriminator_Loss: %.4f Generator_Loss (Content/Advers/Total): %.4f/%.4f/%.4f\n' % (epoch, opt.epochs, i, len(loader_train),
mean_discriminator_loss/len(loader_train), mean_generator_content_loss/len(loader_train),
mean_generator_adversarial_loss/len(loader_train), mean_generator_total_loss/len(loader_train)))
# Do checkpointing
torch.save(netG.state_dict(), '%s/model/generator_final_%d.pth' % (opt.outf,epoch))
torch.save(netD.state_dict(), '%s/model/discriminator_final%d.pth' % (opt.outf,epoch))
## the end of each epoch
netG.eval()
# validate
psnr_val = 0
ssim_val = 0.0
val_images = []
num = 0
numofex=opt.noiseL
for index, (lrimg_val, hrimg_val) in enumerate(loader_val):
#lrimg_val, hrimg_val = dataset_val[k]
noise = torch.FloatTensor(lrimg_val.size()).normal_(mean=0, std=opt.val_noiseL/255.)
#lrimgn_val = lrimg_val + noise
lrimgn_val = lrimg_val
#lrimgn_val = torch.Tensor(np.expand_dims(lrimgn_val, axis=0))
#hrimg_val = torch.Tensor(np.expand_dims(hrimg_val, axis=0))
#lrimg_val = lrimg_val + noise
hrimg_val, lrimgn_val = Variable(hrimg_val.cuda(), volatile=True), Variable(lrimgn_val.cuda(), volatile=True)
#print(lrimgn_val[-1].shape)
#print(hrimg_val[-1].shape)
#cv2.imshow('win1', lrimgn_val[-1].detach().cpu().numpy().transpose((1,2,0)))
#cv2.imshow('win2', hrimg_val[-1].detach().cpu().numpy().transpose((1,2,0)))
#cv2.waitKey(0)
out_val = netG(lrimgn_val)
psnr_val_e, ssim_val_e = batch_PSNR(out_val, hrimg_val, scale=3.0, data_range=1.)
psnr_val += psnr_val_e
ssim_val += ssim_val_e
hrimg_val = np.transpose(hrimg_val[0].detach().cpu().numpy(), (1,2,0))
out_val = np.transpose(out_val[0].detach().cpu().numpy(),(1,2,0))
if num<5:
num+=1
# hrimg_val = hrimg_val[int(hrimg_val.shape[0] / 2) - 160:int(hrimg_val.shape[0] / 2) + 160,
# int(hrimg_val.shape[1] / 2) - 160:int(hrimg_val.shape[1] / 2) + 160]
# out_val = out_val[int(out_val.shape[0] / 2) - 160:int(out_val.shape[0] / 2) + 160,
# int(out_val.shape[1] / 2) - 160:int(out_val.shape[1] / 2) + 160]
val_images.extend([hrimg_val,out_val])
output_image=make_grid(val_images,nrow=2,nline=1)
if not os.path.exists('%s/training_results/%d/' % (opt.outf, numofex)):
os.makedirs('%s/training_results/%d/' % (opt.outf, numofex))
save_result(output_image,path='%s/training_results/%d/epoch%d.png' % (opt.outf,numofex,epoch))
psnr_val /= len(dataset_val)
ssim_val /= len(dataset_val)
print("\n[epoch %d] PSNR_val: %.4f" % (epoch+1, psnr_val))
print("\n[epoch %d] SSIM_val: %.4f" % (epoch+1, ssim_val))
writer.add_scalar('PSNR on validation data', psnr_val, epoch)
writer.add_scalar('SSIM on validation data', ssim_val, epoch)
if __name__ == "__main__":
if opt.preprocess:
prepare_data(data_path='data', patch_size=opt.patch_size, stride=30, scale_new=opt.upscale_factor, aug_times=2)
main()