-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_mpevnet.py
182 lines (162 loc) · 8.97 KB
/
train_mpevnet.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
175
176
177
178
179
180
181
182
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 tensorboardX import SummaryWriter
# from DerainDataset import *
from utils import *
from torch.optim.lr_scheduler import MultiStepLR, ReduceLROnPlateau
from SSIM import seq_SSIM
from mpevnet import *
torch.cuda.empty_cache()
import random
random.seed(0)
torch.manual_seed(0)
np.random.seed(0)
parser = argparse.ArgumentParser(description="MPEVNet_train")
parser.add_argument("--preprocess", action="store_true", default=False, help='run prepare_data or not')
parser.add_argument("--batch_size", type=int, default=8, help="Training batch size")
parser.add_argument("--epochs", type=int, default=100, 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("--save_path", type=str, default="logs/mpevNet/s7_i6_100_ev_sa2", help='path to save models and log files')
parser.add_argument("--save_freq",type=int,default=1,help='save intermediate model')
parser.add_argument("--data_path",type=str, default="dataset/train/RainVIDSS",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=str, default='0,0,6', help='number of recursive stages')
parser.add_argument("--sa_loss_epoch", type=int, default=2, help='when (in which epoch) to use self-alignment loss')
parser.add_argument("--sb", type=int, default=0, help='small batch for quick test or not')
parser.add_argument("--data_name", type=str, default="NTURain", help='data name')
parser.add_argument("--dim_feature", type=int, default=16, help='dimension of hidden features')
opt = parser.parse_args()
print(opt)
if opt.use_gpu:
os.environ["CUDA_VISIBLE_DEVICES"] = opt.gpu_id
def main():
print('Loading dataset ...\n')
dataset_train = dataLoader_evnet.NTURainData(data_path=opt.data_path, if_event=True, data_name=opt.data_name)
loader_train = DataLoader(dataset=dataset_train, num_workers=4, batch_size=opt.batch_size, shuffle=True)
print("# of training samples: %d\n" % int(len(dataset_train)))
# Build model
model = MPEVNet(recurrent_iter=[int(i) for i in opt.recurrent_iter.split(',')], use_GPU=opt.use_gpu, opt_dim=opt.dim_feature)
print_network(model)
# loss function
# criterion = nn.MSELoss(size_average=False)
criterion = seq_SSIM()
# Move to GPU
if opt.use_gpu:
model = model.cuda()
criterion.cuda()
# Optimizer
optimizer = optim.Adam(model.parameters(), lr=opt.lr)
# scheduler = MultiStepLR(optimizer, milestones=opt.milestone, gamma=0.2) # learning rates
scheduler = ReduceLROnPlateau(optimizer, mode="max", factor=0.1, patience=1, threshold=1e-3) # learning rates
# record training
writer = SummaryWriter(opt.save_path)
# load the lastest model
initial_epoch = findLastCheckpoint(save_dir=opt.save_path)
if initial_epoch > 0:
print('resuming by loading epoch %d' % initial_epoch)
model.load_state_dict(torch.load(os.path.join(opt.save_path, 'net_epoch%d.pth' % initial_epoch)))
optimizer.load_state_dict(torch.load(os.path.join(opt.save_path, 'param_epoch%d.pth' % initial_epoch)))
# start training
step = initial_epoch * dataset_train.__len__() / opt.batch_size
prev_psnr = 0
if_sa_loss = True
for epoch in range(initial_epoch, opt.epochs):
for param_group in optimizer.param_groups:
print('learning rate %f' % param_group["lr"])
epoch_loss = 0
epoch_psnr = 0
## epoch training start
for i, (input_train, target_train, event_train) in enumerate(loader_train, 0):
if opt.sb == 1 and i > 10:
break
model.train()
model.zero_grad()
optimizer.zero_grad()
# input_train, target_train = Variable(input_train), Variable(target_train)
if opt.use_gpu:
input_train, target_train, event_train = input_train.cuda(), target_train.cuda(), event_train.cuda()
input_train.requires_grad_()
event_train.requires_grad_()
# print(input_train.requires_grad, event_train.requires_grad, target_train.requires_grad)
out_train = model(input_train, event_train)
# print(out_train.requires_grad)
if (epoch >= opt.sa_loss_epoch and prev_psnr >= 34):
if_sa_loss = True
prev_psnr = 0
pixel_metric = criterion(target_train, out_train, if_sa_loss)
loss = -pixel_metric
loss.backward()
# nn.utils.clip_grad_norm_(model.parameters(), max_norm=2., norm_type=2)
optimizer.step()
torch.cuda.empty_cache()
# training curve
model.eval()
with torch.no_grad():
out_train = out_train.detach()#model(input_train, event_train)
out_train = torch.clamp(out_train, 0., 1.)
psnr_train = batch_PSNR(out_train, target_train, 1.)
epoch_psnr += psnr_train
epoch_loss += pixel_metric.item()
print("[epoch %d][%d/%d] loss: %.4f, pixel_metric: %.4f, PSNR: %.4f" %
(epoch+1, i+1, len(loader_train), loss.item(), pixel_metric.item(), psnr_train))
if step % 10 == 0:
# Log the scalar values
writer.add_scalar('loss', loss.item(), step)
writer.add_scalar('PSNR on training data', psnr_train, step)
writer.add_scalar('Learning rate', optimizer.param_groups[0]["lr"], epoch)
step += 1
#if i == 10:
# break
## epoch training end
print("The average loss for epoch {} is {}".format(epoch, epoch_loss / len(loader_train)))
print("The average psnr for epoch {} is {}".format(epoch, epoch_psnr / len(loader_train)))
scheduler.step(epoch_loss / len(loader_train))
prev_psnr = epoch_psnr / len(loader_train)
# log the images
model.eval()
with torch.no_grad():
out_train = out_train.detach()#model(input_train, event_train)
out_train = torch.clamp(out_train, 0., 1.)
print(target_train.data.shape)
_, dim_c, dim_s, dim_w, dim_h = target_train.data.shape
im_target = utils.make_grid(target_train.data.transpose(2,1).reshape(-1, dim_c, dim_w, dim_h), nrow=dim_s, normalize=True, scale_each=True)
im_input = utils.make_grid(input_train.data.transpose(2,1).reshape(-1, dim_c, dim_w, dim_h), nrow=dim_s, normalize=True, scale_each=True)
im_event = utils.make_grid(event_train.data.transpose(2,1).reshape(-1, 2, dim_w, dim_h), nrow=dim_s+1, normalize=True, scale_each=True)
im_derain = utils.make_grid(out_train.data.transpose(2,1).reshape(-1, dim_c, dim_w, dim_h), nrow=dim_s, normalize=True, scale_each=True)
writer.add_image('clean image', im_target, epoch+1)
writer.add_image('rainy image', im_input, epoch+1)
writer.add_image('event image', im_event, epoch+1)
writer.add_image('deraining image', im_derain, epoch+1)
# save model
torch.save(model.state_dict(), os.path.join(opt.save_path, 'net_latest.pth'))
torch.save(optimizer.state_dict(), os.path.join(opt.save_path, 'param_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)))
torch.save(optimizer.state_dict(), os.path.join(opt.save_path, 'param_epoch%d.pth' % (epoch+1)))
writer.add_scalar('Epoch loss', epoch_loss / len(loader_train), epoch)
writer.add_scalar('Epoch psnr', epoch_psnr / len(loader_train), epoch)
if __name__ == "__main__":
if opt.preprocess:
if opt.data_path.find('RainTrainH') != -1:
print(opt.data_path.find('RainTrainH'))
prepare_data_RainTrainH(data_path=opt.data_path, patch_size=100, stride=80)
elif opt.data_path.find('RainTrainL') != -1:
prepare_data_RainTrainL(data_path=opt.data_path, patch_size=100, stride=80)
elif opt.data_path.find('Rain12600') != -1:
prepare_data_Rain12600(data_path=opt.data_path, patch_size=100, stride=100)
elif opt.data_path.find('NTURain') != -1:
dataLoader_evnet.event_preprocess(patch_size=128, stride=128, seq_crop_len=7, seq_crop_stride=7)
elif opt.data_path.find('RainVIDSS') != -1:
dataLoader_evnet.event_preprocess_RainVIDSS(patch_size=128, stride=128, seq_crop_len=7, seq_crop_stride=7)
else:
print('unkown datasets: please define prepare data function in DerainDataset.py')
main()