-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
174 lines (147 loc) · 7.08 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
# python3
# -*- coding: utf-8 -*-
# @Time : 2021/11/10 15:23
# @Author : Liupeng Lin
# @Email : [email protected]
# Copyright (C) 2021 Liupeng Lin. All Rights Reserved.
import os
import gc
import re
import time
import glob
import h5py
import torch
import argparse
import warnings
import numpy as np
import torch.optim as optim
from torch import nn
from dataset import *
from fdfnet import FDFNet
from torch.autograd import Variable
from torch.utils.data import DataLoader
from torch.optim.lr_scheduler import StepLR
warnings.filterwarnings("ignore")
parser = argparse.ArgumentParser(description='Pytorch FDFNet')
parser.add_argument('--model', default='FDFNet', type=str, help='choose path of model')
parser.add_argument('--batchsize', default=64, type=int, help='batch size')
parser.add_argument('--epoch', default=200, type=int, help='number of train epoch')
parser.add_argument('--lr', default=1e-4, type=float, help='initial learning rate for Adam')
parser.add_argument('--train_data', default='data/train/train_hv_vv_y4r.h5', type=str, help='path of train data')
parser.add_argument('--nFeat', default=64, type=int, help='the number of feature maps')
parser.add_argument('--nDense', type=int, default=6, help='nDenselayer of RDB')
parser.add_argument('--growthRate', type=int, default=32, help='growthRate of dense net')
parser.add_argument('--ncha_dualsar', default=2, type=int, help='the number of SinSAR channels')
parser.add_argument('--ncha_polsar', default=9, type=int, help='the number of PolSAR channels')
parser.add_argument('--ncha_pd', default=4, type=int, help='the number of polarimetric decomposition images channels')
parser.add_argument('--ncha_diff', default=2, type=int, help='the number of differential images channels')
parser.add_argument('--ratio', default=8, type=int, help='the ratio of channel attention module')
parser.add_argument('--gpu', default='0,1', type=str, help='gpu id')
args = parser.parse_args()
cuda = torch.cuda.is_available()
nGPU = torch.cuda.device_count()
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
savedir = os.path.join('model', args.model)
log_header = [
'epoch',
'iteration',
'train/loss',
]
if not os.path.exists(savedir):
os.mkdir(savedir)
if not os.path.exists(os.path.join(savedir, 'log.csv')):
with open(os.path.join(savedir, 'log.csv'), 'w') as f:
f.write(','.join(log_header) + '\n')
def find_checkpoint(savedir):
file_list = glob.glob(os.path.join(savedir, 'model_*.pth'))
if file_list:
epochs_exist = []
for m in file_list:
result = re.findall(".*model_(.*).pth.*", m)
epochs_exist.append(int(result[0]))
initial_epoch = max(epochs_exist)
else:
initial_epoch = 0
return initial_epoch
def main():
# load train data
print('==> Data loading')
hf = h5py.File(args.train_data, 'r+')
lr_polsar = np.float32(hf['data1'])
hr_dualsar = np.float32(hf['data2'])
hr_diff = np.float32(hf['data3'])
lr_pd = np.float32(hf['data4'])
hr_polsar = np.float32(hf['label'])
lr_polsar = torch.from_numpy(lr_polsar).view(-1, args.ncha_polsar, 20, 20)
hr_dualsar = torch.from_numpy(hr_dualsar).view(-1, args.ncha_dualsar, 40, 40)
hr_diff = torch.from_numpy(hr_diff).view(-1, args.ncha_diff, 40, 40)
lr_pd = torch.from_numpy(lr_pd).view(-1, args.ncha_pd, 20, 20)
hr_polsar = torch.from_numpy(hr_polsar).view(-1, args.ncha_polsar, 40, 40)
train_set = FDFNetDataset(lr_polsar, hr_dualsar, hr_diff, lr_pd, hr_polsar)
train_loader = DataLoader(dataset=train_set, num_workers=8, drop_last=True, batch_size=64, shuffle=True, pin_memory=True)
print('==> Model building')
model = FDFNet(args)
criterion1 = nn.L1Loss(size_average=True)
criterion2 = nn.L1Loss(size_average=True)
criterion3 = nn.L1Loss(size_average=True)
if cuda:
print('==> GPU setting')
model = model.cuda()
criterion1 = criterion1.cuda()
criterion2 = criterion2.cuda()
criterion3 = criterion3.cuda()
print('==> Optimizer setting')
optimizer = optim.Adam(model.parameters(), lr=args.lr, weight_decay=0)
scheduler = StepLR(optimizer, step_size=50, gamma=0.5)
initial_epoch = find_checkpoint(savedir=savedir)
if initial_epoch > 0:
print('==> Resuming by loading epoch %03d' % initial_epoch)
model = torch.load(os.path.join(savedir, 'model_%03d.pth' % initial_epoch))
if len(args.gpu) > 1:
model = torch.nn.DataParallel(model)
for epoch in range(initial_epoch, args.epoch):
scheduler.step(epoch)
epoch_loss = 0
start_time = time.time()
# train
model.train()
num_train = len(train_loader.dataset)
for iteration, batch in enumerate(train_loader):
lr_polsar_batch, hr_dualsar_batch, hr_diff_batch, lr_pd_batch, hr_polsar_batch = Variable(batch[0]), Variable(batch[1]), Variable(batch[2]), Variable(batch[3]), Variable(batch[4])
if cuda:
lr_polsar_batch = lr_polsar_batch.cuda()
hr_dualsar_batch = hr_dualsar_batch.cuda()
hr_diff_batch = hr_diff_batch.cuda()
lr_pd_batch = lr_pd_batch.cuda()
hr_polsar_batch = hr_polsar_batch.cuda()
optimizer.zero_grad()
out = model(lr_polsar_batch, hr_dualsar_batch, hr_diff_batch, lr_pd_batch)
out_hh = out[:, 0, :, :].view(-1, 1, 40, 40)
out_hv = (0.5 * out[:, 5, :, :]).view(-1, 1, 40, 40)
out_vv = out[:, 8, :, :].view(-1, 1, 40, 40)
hr_sar_hv_batch = hr_dualsar_batch[:, 0, :, :].view(-1, 1, 40, 40)
hr_sar_vv_batch = hr_dualsar_batch[:, 1, :, :].view(-1, 1, 40, 40)
loss1 = criterion1(out, hr_polsar_batch)
loss2 = criterion2(out_hv, hr_sar_hv_batch)
loss3 = criterion3(out_vv, hr_sar_vv_batch)
lambda1 = loss1.data / (loss1.data + loss2.data + loss3.data)
lambda2 = loss2.data / (loss1.data + loss2.data + loss3.data)
lambda3 = loss3.data / (loss1.data + loss2.data + loss3.data)
loss = lambda1 * loss1 + lambda2 * loss2 + lambda3 * loss3
epoch_loss += loss.data/num_train
print('%4d %4d / %4d loss = %2.6f' % (epoch + 1, iteration, train_set.hr_polsar.size(0)/args.batchsize, loss.data))
loss.backward()
optimizer.step()
with open(os.path.join(savedir, 'log.csv'), 'a') as file:
log = [epoch, iteration] + [loss.data.item()]
log = map(str, log)
file.write(','.join(log) + '\n')
if len(args.gpu) > 1:
torch.save(model.module, os.path.join(savedir, 'model_%03d.pth' % (epoch + 1)))
else:
torch.save(model, os.path.join(savedir, 'model_%03d.pth' % (epoch + 1)))
gc.collect()
elapsed_time = time.time() - start_time
print('epcoh = %4d , time is %4.4f s' % (epoch + 1, elapsed_time))
if __name__ == '__main__':
main()