/
Train.py
160 lines (128 loc) · 6.63 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
import os
import torch
from pyramid_structure.Omi_LP import MSLT
from torch.utils.data import DataLoader
from torchvision import transforms, utils
import dataset.ImageDataset as ImageDataset
import dataset.loaddata as loaddata
from batch_transformers import BatchRandomResolution, BatchToTensor, BatchRGBToYCbCr, YCbCrToRGB, BatchTestResolution
import argparse
from torch import nn, optim
from skimage.metrics import structural_similarity as ssim
from skimage.metrics.simple_metrics import peak_signal_noise_ratio as psnr
from pytorch_ssim.wavelet_ssim_loss import WSloss
import torchvision
import cv2 as cv
def train(config):
model = MSLT()
#model = nn.DataParallel(model, device_ids=[0, 1, 2])
model = model.cuda()
if config.load_pretrain == True:
model.load_state_dict(torch.load(config.pretrain_dir))
optimizer = optim.Adam(model.parameters(), lr=config.lr, betas=(0.9, 0.999))
train_transform = transforms.Compose([
transforms.ToTensor(),
])
valid_transform = transforms.Compose([
BatchToTensor(),
])
l1_loss=nn.L1Loss(reduction='mean').cuda()
MSE_loss=nn.MSELoss(reduction='mean').cuda()
wsloss = WSloss().cuda()
train_data = ImageDataset.ImageSeqDataset(csv_file=os.path.join(config.datapath, 'test.txt'),
img_dir=config.datapath,
transform=train_transform)
train_loader = DataLoader(train_data,
batch_size=config.train_batch_size,
num_workers=config.num_workers,
pin_memory=True,
shuffle=True)
valid_data = loaddata.ImageSeqDataset(csv_file=os.path.join(config.validpath, 'test.txt'),
Train_img_seq_dir=config.validpath,
Label_img_dir=config.validlabel,
Train_transform=valid_transform,
Label_transform=transforms.ToTensor(),
randomlist=False)
valid_loader = DataLoader(valid_data,
batch_size=1,
shuffle=False,
pin_memory=True,
num_workers=1)
iters = len(train_loader)
scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(optimizer, T_0=iters*2, T_mult=1)
for epoch in range(config.num_epochs):
for step, sample_batched in enumerate(train_loader):
train_image, label_image = sample_batched['train'], sample_batched['label']
train_image = train_image.cuda()
label_image = label_image.cuda()
optimizer.zero_grad()
#content = torch.exp(content)
output = model(train_image)
total_loss = MSE_loss(output , label_image)
total_loss.backward()
optimizer.step()
scheduler.step(epoch + step / iters)
if ((step + 1) % config.display_iter) == 0:
print("Loss at iteration", step + 1, ":", total_loss.item())
if ((step + 1) % config.snapshot_iter) == 0:
torch.save(model.state_dict(), config.snapshots_folder + "Epoch" + str(epoch) + '.pth')
f = "loss.txt"
with open(f, "a") as file:
file.write("epoch="+str(epoch)+"loss="+str(total_loss.item())+"lr="+str(optimizer.param_groups[0]['lr'])+ "\n")
if epoch > 0 and epoch % 5 == 0:
time1 = 0
count = 1
evl1 = 0
evl2 = 0
evl_lpipsvgg = 0
evl_lpipsalex = 0
for step, sample_batched in enumerate(valid_loader):
test_image, label_image = sample_batched['Train'], sample_batched['Lable']
test_image = test_image.squeeze(0).cuda()
print(label_image.shape)
print(test_image.shape)
label_image = label_image.cuda()
for index in range(5):
out4 = model(test_image[index].unsqueeze(0))
validate_path = "./validation/"
if not os.path.exists(validate_path):
os.makedirs(validate_path)
torchvision.utils.save_image(out4, validate_path + str(step + 1) + "_" + str(index + 1) + ".jpg")
image = cv.imread(validate_path + str(step + 1) + "_" + str(index + 1) + ".jpg")
label = cv.imread("/home/ubuntu/liangjin/SEC/validation/label/" + str(step + 1) + ".jpg")
print(image.shape)
print(label.shape)
evl1 = evl1 + ssim(image, label, multichannel=True)
evl2 = evl2 + psnr(image, label)
evl_ssim = evl1 / (count)
evl_psnr = evl2 / (count)
count = count + 1
print("psnr", evl_psnr)
print("ssim", evl_ssim)
f = "./valid_record/valid.txt"
if not os.path.exists("./valid_record/"):
os.makedirs("./valid_record/")
with open(f, "a") as file: # ”w"代表着每次运行都覆盖内容
file.write("epoch="+str(epoch)+"_"+"ssim="+str(evl_ssim)+"psnr="+str(evl_psnr)+"\n")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--validpath', type=str, default="./data/MEdata/validation/")
parser.add_argument('--validlabel', type=str, default="./data/MEdata/validation/label/")
parser.add_argument('--datapath', type=str, default="./data/MEdata/imagepatch_512/")
parser.add_argument('--lr', type=float, default=0.001)
parser.add_argument('--weight_decay', type=float, default=0.9)
parser.add_argument('--grad_clip_norm', type=float, default=1)
parser.add_argument('--num_epochs', type=int, default=200)
parser.add_argument('--train_batch_size', type=int, default=32)
parser.add_argument('--val_batch_size', type=int, default=4)
parser.add_argument('--num_workers', type=int, default=4)
parser.add_argument('--display_iter', type=int, default=10)
parser.add_argument('--snapshot_iter', type=int, default=10)
parser.add_argument('--snapshots_folder', type=str, default="snapshots/")
parser.add_argument('--load_pretrain', type=bool, default= False)
parser.add_argument('--pretrain_dir', type=str, default= 'snapshots131/Epoch131.pth')
config = parser.parse_args()
if not os.path.exists(config.snapshots_folder):
os.mkdir(config.snapshots_folder)
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
train(config)