-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
174 lines (146 loc) · 7.3 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
import os
import argparse
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import torchvision.utils as utils
from torch.autograd import Variable
from torch.utils.data import DataLoader
from utils import utils
import torch.nn.functional as F
from DerainDataset import *
from architect import Architect
from utils import *
from torch.optim.lr_scheduler import MultiStepLR
from skimage.metrics import peak_signal_noise_ratio as compare_psnr
from SSIM import SSIM
from utils.image_io import save_graph
# from networks import *
from P_3UNet_Darts import UNet
from percep_loss import networks
from percep_loss import vgg
parser = argparse.ArgumentParser(description="PReNet_train")
parser.add_argument("--preprocess", type=bool, default=True, help='run prepare_data or not')
parser.add_argument("--batch_size", type=int, default=8, help="Training batch size")
parser.add_argument('--train_portion', type=float, default=0.5, help='portion of training data')
parser.add_argument("--epochs", type=int, default=150, help="Number of training epochs")
parser.add_argument("--milestone", type=int, default=[30,50,80], help="When to decay learning rate")
parser.add_argument("--lr", type=float, default=1e-3, help="initial learning rate")
parser.add_argument('--arch_learning_rate', type=float, default=3e-4, help='learning rate for arch encoding')
parser.add_argument('--arch_weight_decay', type=float, default=1e-3, help='weight decay for arch encoding')
parser.add_argument("--save_path", type=str, default="logs/", help='path to save models and log files')
parser.add_argument("--save_freq",type=int,default=3,help='save intermediate model')
parser.add_argument("--data_path",type=str, default="./train/Rain100H",help='path to training data')
parser.add_argument("--use_gpu", type=bool, default=True, help='use GPU or not')
parser.add_argument("--gpu_id", type=str, default="0", help='GPU id')
parser.add_argument("--recurrent_iter", type=int, default=6, help='number of recursive stages')
opt = parser.parse_args()
def batch_PSNR(img, imclean, data_range):
Img = img.data.cpu().numpy().astype(np.float32)
Iclean = imclean.data.cpu().numpy().astype(np.float32)
PSNR = 0
for i in range(Img.shape[0]):
PSNR += compare_psnr(Iclean[i,:,:,:], Img[i,:,:,:], data_range=data_range)
return (PSNR/Img.shape[0])
self_device = torch.device('cuda:{}'.format('0'))
self_vgg = vgg.Vgg19(requires_grad=False).to(self_device)
criterionVgg = networks.VGGLoss1(self_device, vgg=self_vgg, normalize=False)
def main():
losses = []
print('Loading dataset ...\n')
dataset_train = Dataset(data_path=opt.data_path)
num_train = dataset_train.__len__()
indices = list(range(num_train))
split = int(np.floor(opt.train_portion * num_train))
loader_train = DataLoader(dataset=dataset_train, num_workers=0, batch_size=opt.batch_size,
sampler=torch.utils.data.sampler.SequentialSampler(indices[:split]),pin_memory=True)
loader_valid = DataLoader(dataset=dataset_train, num_workers=0, batch_size=opt.batch_size,
sampler=torch.utils.data.sampler.SequentialSampler(indices[split:num_train]),pin_memory=True)
print("# of training samples: %d\n" % int(len(dataset_train)))
# Build model
model = UNet()
# loss function
# criterion = nn.MSELoss(size_average=False)
criterion = SSIM()
mse = torch.nn.MSELoss()
# Move to GPU
if opt.use_gpu:
model = model.cuda()
criterion.cuda()
mse.cuda()
# Optimizer
optimizer = optim.Adam(model.parameters(), lr=opt.lr)
scheduler = MultiStepLR(optimizer, milestones=opt.milestone, gamma=0.2) # learning rates
architect = Architect(model, opt)
# start training
for epoch in range(opt.epochs):
scheduler.step(epoch)
print('epoch:', epoch)
for param_group in optimizer.param_groups:
print('learning rate %f' % param_group["lr"])
# 选出权重值大的2个节点并保留操作
genotype = model.genotype()
print('genotype = %s', genotype)
# print(F.softmax(model.alphas_conv, dim=-1))
print(F.softmax(model.alphas_upsample, dim=-1))
select = model.select_skip()
print('select_skip = %s', select)
print(F.softmax(model.alphas_skipup, dim=-1))
print(F.softmax(model.alphas_skipdown, dim=-1))
## epoch training start
objs = utils.AvgrageMeter()
for i, (input_train, target_train) in enumerate(loader_train, 0):
model.train()
model.zero_grad()
optimizer.zero_grad()
input_train, target_train = Variable(input_train), Variable(target_train)
n = input_train.size(0)
if opt.use_gpu:
input_train, target_train = input_train.cuda(), target_train.cuda()
#处理α
(input_search, target_search) = next(iter(loader_valid))
blended_search = Variable(input_search, requires_grad=False).cuda()
target_search = Variable(target_search, requires_grad=False).cuda(non_blocking=True)
output_search = model(blended_search)
architect.step(output_search, target_search, blended_search)
aloss = 0.1 * mse(output_search, target_search) + (1-criterion(output_search, target_search))
vgg_loss = criterionVgg(target_search, output_search) / criterionVgg(blended_search, output_search)
# a = vgg_loss / (vgg_loss+aloss)
# v = aloss / (vgg_loss+aloss)
print('aloss: %.4f' % (aloss.item()) + ' || vgg_loss: %.4f' % ( vgg_loss.item()))
# aloss = a * aloss + v * vgg_loss
aloss += 0.1 * vgg_loss
out_train = model(input_train)
pixel_metric = criterion(target_train, out_train)
loss = 0.1 * mse(target_train, out_train) + (1-pixel_metric)
vgg_loss = criterionVgg(target_train, out_train) / criterionVgg(input_train, out_train)
# a = vgg_loss / (vgg_loss + loss)
# v = loss / (vgg_loss + loss)
# loss = a * loss + v * vgg_loss
loss += 0.1 * vgg_loss
loss.backward()
optimizer.step()
objs.update(aloss.data.item(), n)
print(objs.avg)
## epoch training end
# # save model
# torch.save(model.state_dict(), os.path.join(opt.save_path, 'net_latest.pth'))
if epoch % opt.save_freq == 0:
torch.save(model.state_dict(), os.path.join(opt.save_path, 'net_epoch%d.pth' % (epoch+1)))
# save model
torch.save(model.state_dict(), os.path.join(opt.save_path, 'net_latest.pth'))
losses.append(objs.avg)
save_graph(str(epoch) + "_aloss", losses,
output_path=opt.save_path + '/')
if __name__ == "__main__":
if opt.preprocess:
if opt.data_path.find('RainTrainH') != -1:
prepare_data_RainTrainH(data_path=opt.data_path, patch_size=128, stride=80)
elif opt.data_path.find('RainTrainL') != -1:
prepare_data_RainTrainL(data_path=opt.data_path, patch_size=128, stride=80)
elif opt.data_path.find('Rain12600') != -1:
prepare_data_Rain12600(data_path=opt.data_path, patch_size=128, stride=100)
else:
print('unkown datasets: please define prepare data function in DerainDataset.py')
main()