-
Notifications
You must be signed in to change notification settings - Fork 2
/
train.py
162 lines (141 loc) · 6.77 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
import torch as t
import torch.utils.data as data
import torch.optim as optim
from data import train_dataset_color, valid_dataset_color, train_dataset_gray, valid_dataset_gray
from cfg import par
from Loss_Function import myLoss_gray, myLoss_color
from skimage.measure import compare_psnr, compare_ssim
from Network import Denoising_Net_gray, Denoising_Net_color
from utils import gray_reconstruction, color_reconstruction
import warnings
warnings.filterwarnings("ignore")
Validation_color = ['CBSD68', 'Kodak24', 'McMaster'] # 'Kodak24', 'CBSD68', 'McMaster'
Validation_gray = ['Set12', 'BSD68'] # 'Set12', 'BSD68'
beginner = par.beginner
mode = par.mode
print('Training Mode: ' + mode)
if mode == 'gray':
print('Validation set: Set12, BSD68')
t.cuda.set_device(1)
log = open('./log/log.txt', 'w')
log.truncate()
log.close()
# network
net = Denoising_Net_gray()
# loss function
loss_fn = myLoss_gray()
else:
print('Validation set: CBSD68, Kodak24, McMaster')
t.cuda.set_device(0)
log = open('./log/log_c.txt', 'w')
log.truncate()
log.close()
# network
net = Denoising_Net_color()
# loss function
loss_fn = myLoss_color()
# define optimizer
optimizer = optim.Adam(net.parameters(), lr=1e-4, weight_decay=0.0005)
for i in range(beginner, beginner+50):
for group in optimizer.param_groups:
if (i >= beginner and i < beginner+15):
group['lr'] = 1e-4
if (i >= beginner+15 and i < beginner+30):
group['lr'] = 1e-5
if (i >= beginner+30 and i < beginner+40):
group['lr'] = 1e-6
if (i >= beginner+40 and i < beginner+50):
group['lr'] = 1e-7
if mode == 'gray':
if i != 0 and i == beginner:
net.load_state_dict(t.load('./model/model_gray_epoch_'+str(beginner))) # modify model name for resuming training
net.cuda()
log = open('./log/log.txt', 'a')
log.write('\n\nEpoch: %d, Learning_rate: %.2e, batch_size: %d' % (i+1, optimizer.param_groups[0]['lr'], par.bs))
log.close()
else:
if i != 0 and i == beginner:
net.load_state_dict(t.load('./model/model_gray_epoch_'+str(beginner))) # modify model name for resuming training
net.cuda()
log = open('./log/log_c.txt', 'a')
log.write('\n\nEpoch: %d, Learning_rate: %.2e, batch_size: %d' % (i + 1, optimizer.param_groups[0]['lr'], par.bs))
log.close()
# dataset
if mode == 'color':
dataset = train_dataset_color(pic=4744 + 1000)
data_loader = data.DataLoader(dataset, batch_size=par.bs, num_workers=0, shuffle=True, pin_memory=False)
elif mode == 'gray':
dataset = train_dataset_gray(pic=4744 + 400 + 600)
data_loader = data.DataLoader(dataset, batch_size=par.bs, num_workers=0, shuffle=True, pin_memory=False)
else:
print('Please select a training mode: gray or color.\n(Using Default Mode: color)')
dataset = train_dataset_color(pic=4744 + 1000)
data_loader = data.DataLoader(dataset, batch_size=par.bs, num_workers=0, shuffle=True, pin_memory=False)
Loss_epoch = 0
for batch_idx, (noisy, target) in enumerate(data_loader):
optimizer.zero_grad()
output = net(noisy.cuda())
Loss = loss_fn(output, target.cuda())
Loss.backward()
optimizer.step()
if batch_idx % 100 == 99 or batch_idx == 0:
print('#batch: %4d, Loss: %f' % (batch_idx + 1, Loss / par.bs))
Loss_epoch = Loss_epoch + Loss / par.bs
print('~~~~~~~~~~~~~~~~~~~~#epoch: %d, Loss: %.4f~~~~~~~~~~~~~~~~~~~~' % (i + 1, Loss_epoch / (batch_idx+1)))
if mode == 'gray':
log = open('./log/log.txt', 'a')
log.write('\nLoss %.4f' % (Loss_epoch / (batch_idx+1)))
log.close()
else:
log = open('./log/log_c.txt', 'a')
log.write('\nLoss %.4f' % (Loss_epoch / (batch_idx + 1)))
log.close()
# validation
if mode == 'gray':
Validation = Validation_gray
t.save(net.state_dict(), './model/model_gray_epoch_' + str(i + 1))
else:
Validation = Validation_color
t.save(net.state_dict(), './model/model_color_epoch_' + str(i + 1))
for valid_idx in range(len(Validation)):
dataset_name = Validation[valid_idx]
if mode == 'gray':
valid_set = valid_dataset_gray(root='/home/rick/database/Denoising/testset/', dataset_name=dataset_name)
valid_loader = data.DataLoader(valid_set, batch_size=1, num_workers=1, shuffle=False, pin_memory=False)
else:
valid_set = valid_dataset_color(root='/home/rick/database/Denoising/testset/', dataset_name=dataset_name)
valid_loader = data.DataLoader(valid_set, batch_size=1, num_workers=1, shuffle=False, pin_memory=False)
PSNR = 0
SSIM = 0
for b_idx, (img_n, img_clean) in enumerate(valid_loader):
img_denoised = net(img_n.cuda())
img_denoised = img_denoised.cpu().detach().numpy().squeeze()
if mode == 'gray':
img_clean = img_clean.numpy().squeeze() # W*H
img_denoised = gray_reconstruction(img_denoised)
img_denoised = img_denoised[0:img_clean.shape[0],0:img_clean.shape[1]]
img_denoised = img_denoised.clip(0, 1)
psnr = round(compare_psnr(img_denoised, img_clean, data_range=1), 3)
ssim = round(compare_ssim(img_denoised, img_clean, data_range=1, multichannel=False), 3)
else:
img_clean = img_clean.numpy().squeeze().transpose(1, 2, 0) # W*H*C
img_denoised = color_reconstruction(img_denoised) # C*W*H
img_denoised = img_denoised[:, 0:img_clean.shape[0], 0:img_clean.shape[1]]
img_denoised = img_denoised.transpose(1, 2, 0).clip(0, 1)
psnr = round(compare_psnr(img_denoised, img_clean, data_range=1), 3)
ssim = round(compare_ssim(img_denoised, img_clean, data_range=1, multichannel=True), 3)
PSNR = PSNR + psnr
SSIM = SSIM + ssim
if dataset_name == 'Set12':
print('#pic: %d, PSNR: %.4f, SSIM: %.4f' %(b_idx+1, psnr, ssim))
PSNR = PSNR/(b_idx+1)
SSIM = SSIM/(b_idx+1)
print(dataset_name + ': ave_PSNR: %f, ave_SSIM: %f' % (PSNR, SSIM))
if mode == 'gray':
log = open('./log/log.txt', 'a')
log.write('\n' + dataset_name + ': ave_PSNR: %.4f, ave_SSIM: %.4f' % (PSNR, SSIM))
log.close()
else:
log = open('./log/log_c.txt', 'a')
log.write('\n' + dataset_name + ': ave_PSNR: %.4f, ave_SSIM: %.4f' % (PSNR, SSIM))
log.close()