-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathWINNet_denoise_train.py
163 lines (149 loc) · 7.58 KB
/
WINNet_denoise_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
import os
import argparse
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
from torch.utils.data import DataLoader
from utils.networks import WINNetklvl
from utils.dataset import prepare_data, Dataset
import time
from tqdm import tqdm
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
parser = argparse.ArgumentParser(description="WINNet")
parser.add_argument("--preprocess", type=bool, default=True, help='run prepare_data or not')
parser.add_argument("--batchSize", type=int, default=32, help="Training batch size")
parser.add_argument("--num_of_steps", type=int, default=4, help="Number of steps")
parser.add_argument("--num_of_layers", type=int, default=4, help="Number of layers")
parser.add_argument("--num_of_channels", type=int, default=32, help="Number of channels")
parser.add_argument("--lvl", type=int, default=1, help="number of levels")
parser.add_argument("--split", type=str, default="dct", help='splitting operator')
parser.add_argument("--dnlayers", type=int, default=4, help="Number of denoising layers")
parser.add_argument("--epochs", type=int, default=50, 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("--start_epoch", type=int, default=0, help="start epochs")
parser.add_argument("--lr", type=float, default=1e-3, help="Initial learning rate")
parser.add_argument("--decay_rate", type=float, default=0.1, help="decay rate")
parser.add_argument("--outf", type=str, default="logs", 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')
opt = parser.parse_args()
print(opt)
def main():
if not os.path.exists(opt.outf):
os.makedirs(opt.outf)
# Load dataset
print('Loading dataset ...\n')
dataset_train = Dataset(train=True)
loader_train = DataLoader(dataset=dataset_train, num_workers=4, batch_size=opt.batchSize, shuffle=True, pin_memory=True)
print("# of training samples: %d\n" % int(len(dataset_train)))
# Build model
net = WINNetklvl(steps=opt.num_of_steps, layers=opt.num_of_layers, channels=opt.num_of_channels, klvl=opt.lvl,
mode=opt.split, dnlayers=opt.dnlayers)
criterion = nn.MSELoss(size_average=False).cuda()
device_ids = [0]
model = nn.DataParallel(net, device_ids=device_ids).cuda()
torch.backends.cudnn.benchmark = True
pytorch_total_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
print('Total Number of Parameters: ', pytorch_total_params)
# optimizer
optimizer = optim.Adam(model.parameters(), lr=opt.lr)
# training
noiseL_B = [0, 55] # ingnored when opt.mode=='S'
start_epoch = opt.start_epoch
if start_epoch > 0:
print('Start Epoch: ', start_epoch)
model.load_state_dict(torch.load(os.path.join(opt.outf, 'net_WINNet_epoch_{}.pth'.format(start_epoch))))
for param_group in optimizer.param_groups:
current_lr = param_group["lr"]
current_lr = current_lr * (opt.decay_rate ** (start_epoch // opt.milestone))
param_group["lr"] = current_lr
print('Learning rate: ', current_lr)
else:
torch.save(model.state_dict(), os.path.join(opt.outf, 'net_WINNet_epoch_{}.pth'.format(start_epoch)))
epoch = start_epoch + 1
torch.cuda.synchronize()
t1 = time.time()
print('Epoch: ', epoch)
while epoch <= opt.epochs:
with tqdm(loader_train, unit='batch') as tepoch:
i = 0
for data in tepoch:
tepoch.set_description(f"Epoch {epoch}")
# training step
model.train()
model.zero_grad()
optimizer.zero_grad()
img_train = data.cuda()
########################################################################################################
if opt.mode == 'S':
stdN = opt.noiseL
if opt.mode == 'B':
stdRD = np.random.uniform(noiseL_B[0], noiseL_B[1], size=1)
stdN = stdRD[0]
stdNv = Variable(stdN*torch.ones(1, device=torch.device('cuda:0')))
noise = torch.cuda.FloatTensor(img_train.size()).normal_(mean=0, std=stdN / 255.)
########################################################################################################
imgn_train = img_train + noise
img_train, imgn_train = Variable(img_train), Variable(imgn_train)
outn_train, oloss = model(imgn_train, stdNv)
loss = criterion(outn_train, img_train) / (imgn_train.size()[0] * 2)
loss = loss + oloss
Loss = loss.detach()
if i % 10 == 0:
norm_PU = net.linear()
loss = loss + 1 * 1e-1 * norm_PU
########################################################################################################
loss.backward()
nn.utils.clip_grad_norm_(model.parameters(), max_norm=1, norm_type=2)# Gradient Clipping
optimizer.step()
tepoch.set_postfix(loss=Loss)
time.sleep(0.0001)
i += 1
torch.cuda.synchronize()
t2 = time.time()
print('Time:' f'{(t2 - t1) / 60: .3e} mins')
print('Orthogonal Loss: ', oloss)
print('Norm PU', norm_PU)
for param_group in optimizer.param_groups:
current_lr = param_group["lr"]
print('Learning rate: ', current_lr)
if torch.isnan(Loss) or torch.isinf(Loss):
if epoch > 0:
print('Load Model Epoch: ', epoch-1)
model.load_state_dict(torch.load(os.path.join(opt.outf, 'net_WINNet_epoch_{}.pth'.format(epoch-1))))
else:
net = WINNetklvl(steps=opt.num_of_steps, layers=opt.num_of_layers, channels=opt.num_of_channels, klvl=opt.lvl,
mode=opt.modeSM, dnlayers=opt.dnlayers)
model = nn.DataParallel(net, device_ids=device_ids).cuda()
for param_group in optimizer.param_groups:
current_lr = param_group["lr"]
current_lr = current_lr * 0.8
param_group["lr"] = current_lr
print('Learning rate: ', current_lr)
continue
torch.cuda.synchronize()
t1 = time.time()
print("Save Model Epoch: %d" % (epoch))
print('\n')
torch.save(model.state_dict(), os.path.join(opt.outf, 'net_WINNet_epoch_{}.pth'.format(epoch)))
epoch += 1
print('Epoch: ', epoch)
if (epoch > 0) and (epoch % opt.milestone == 0):
for param_group in optimizer.param_groups:
current_lr = param_group["lr"]
current_lr = current_lr * opt.decay_rate
param_group["lr"] = current_lr
print('Learning rate: ', current_lr)
if opt.mode == 'S':
prepare_data(data_path='data', patch_size=40, stride=10, aug_times=1)
if opt.mode == 'B':
prepare_data(data_path='data', patch_size=50, stride=10, aug_times=2)
if __name__ == "__main__":
if opt.preprocess:
if opt.mode == 'S':
prepare_data(data_path='data', patch_size=40, stride=10, aug_times=1)
if opt.mode == 'B':
prepare_data(data_path='data', patch_size=50, stride=10, aug_times=2)
main()