forked from asprasan/unified_framework
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
265 lines (210 loc) · 9.88 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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
'''
-----------------------------------
TRAINING CODE - SHIFTVARCONV + UNET
-----------------------------------
'''
import os
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import logging
import glob
import argparse
import time
from torch.utils import data
from torch.utils.tensorboard import SummaryWriter
## set random seed
torch.manual_seed(12)
np.random.seed(12)
from dataloader import Dataset_load
from sensor import C2B
from unet import UNet
from inverse import ShiftVarConv2D, StandardConv2D
import utils
## parse arguments
parser = argparse.ArgumentParser()
parser.add_argument('--expt', type=str, required=True, help='expt name')
parser.add_argument('--save_root',type=str,required=False,default='models', help='root path to save trained models')
parser.add_argument('--epochs', type=int, default=500, help='num epochs to train')
parser.add_argument('--batch', type=int, required=True, help='batch size for training and validation')
parser.add_argument('--lr', type=float, default=1e-4, help='learning rate')
parser.add_argument('--blocksize', type=int, default=8, help='tile size for code default 3x3')
parser.add_argument('--subframes', type=int, default=16, help='num sub frames')
parser.add_argument('--ckpt', type=str, default=None, help='checkpoint to load')
parser.add_argument('--gpu', type=str, required=True, help='GPU ID')
parser.add_argument('--mask', type=str, default='random', help='"impulse" or "random" or "opt" or "flutter"')
parser.add_argument('--two_bucket', action='store_true', help='1 bucket or 2 buckets')
parser.add_argument('--intermediate', action='store_true', help="intermediate reconstruction")
args = parser.parse_args()
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
## params for DataLoader
train_params = {'batch_size': args.batch,
'shuffle': True,
'num_workers': 20,
'pin_memory': True}
val_params = {'batch_size': args.batch,
'shuffle': False,
'num_workers': 20,
'pin_memory': True}
num_epochs = args.epochs
save_path = os.path.join(args.save_root, args.expt)
utils.create_dirs(save_path)
## tensorboard summary logger
writer = SummaryWriter(
log_dir=os.path.join(save_path, 'logs'))
## configure runtime logging
logging.basicConfig(level=logging.INFO,
filename=os.path.join(save_path, 'logs', 'logfile.log'),
format='%(asctime)s - %(message)s',
filemode='w')
# logger=logging.getLogger()#.setLevel(logging.INFO)
console = logging.StreamHandler()
console.setLevel(logging.INFO)
console.setFormatter(logging.Formatter('%(asctime)s - %(message)s'))
logging.getLogger('').addHandler(console)
logging.info(args)
## dataloaders using hdf5 file
data_path = None
try:
assert data_path is not None
except AssertionError:
print('path to hdf5 data not specified')
print('hdf5 data should contain two datasets train and test')
print('++++exiting++++')
exit(0)
## initializing training and validation data generators
training_set = Dataset_load(data_path, dataset='train', num_samples='all')
training_generator = data.DataLoader(training_set, **train_params)
logging.info('Loaded training set: %d videos'%(len(training_set)))
validation_set = Dataset_load(data_path, dataset='test', num_samples=60000)
validation_generator = data.DataLoader(validation_set, **val_params)
logging.info('Loaded validation set: %d videos'%(len(validation_set)))
## initialize nets
if args.intermediate:
num_features = args.subframes
else:
num_features = 64
c2b = C2B(block_size=args.blocksize, sub_frames=args.subframes, mask=args.mask, two_bucket=args.two_bucket).cuda()
if args.mask == 'flutter':
assert not args.two_bucket
invNet = StandardConv2D(out_channels=num_features, window=7).cuda()
else:
invNet = ShiftVarConv2D(out_channels=num_features, block_size=args.blocksize, two_bucket=args.two_bucket).cuda()
uNet = UNet(in_channel=num_features, out_channel=args.subframes, instance_norm=False).cuda()
## optimizer
optimizer = torch.optim.Adam(list(invNet.parameters())+list(uNet.parameters()),
lr=args.lr, weight_decay=1e-5)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', factor=0.9,
patience=5, min_lr=1e-6, verbose=True)
start_epoch = 0
logging.info('Starting training')
for i in range(start_epoch, start_epoch+num_epochs):
## TRAINING
train_iter = 0
interm_loss_sum = 0.
final_loss_sum = 0.
tv_loss_sum = 0.
loss_sum = 0.
psnr_sum = 0.
for gt_vid in training_generator:
gt_vid = gt_vid.cuda()
if not args.two_bucket:
b1 = c2b(gt_vid) # (N,1,H,W)
# b1 = torch.mean(gt_vid, dim=1, keepdim=True)
interm_vid = invNet(b1)
else:
b1, b0 = c2b(gt_vid)
b_stack = torch.cat([b1,b0], dim=1)
interm_vid = invNet(b_stack)
highres_vid = uNet(interm_vid) # (N,16,H,W)
psnr_sum += utils.compute_psnr(highres_vid, gt_vid).item()
## LOSSES
if args.intermediate:
interm_loss = utils.weighted_L1loss(interm_vid, gt_vid)
interm_loss_sum += interm_loss.item()
final_loss = utils.weighted_L1loss(highres_vid, gt_vid)
final_loss_sum += final_loss.item()
tv_loss = utils.gradx(highres_vid).abs().mean() + utils.grady(highres_vid).abs().mean()
tv_loss_sum += tv_loss.item()
if args.intermediate:
loss = final_loss + 0.1*tv_loss + 0.5*interm_loss
else:
loss = final_loss + 0.1*tv_loss
loss_sum += loss.item()
## BACKPROP
optimizer.zero_grad()
loss.backward()
optimizer.step()
if train_iter % 1000 == 0:
logging.info('epoch: %3d \t iter: %5d \t loss: %.4f'%(i, train_iter, loss.item()))
train_iter += 1
logging.info('Total train iterations: %d'%(train_iter))
logging.info('Finished epoch %3d with loss: %.4f psnr: %.4f'
%(i, loss_sum/train_iter, psnr_sum/len(training_set)))
## dump tensorboard summaries
writer.add_scalar('training/loss',loss_sum/train_iter,i)
writer.add_scalar('training/interm_loss',interm_loss/train_iter,i)
writer.add_scalar('training/final_loss',final_loss/train_iter,i)
writer.add_scalar('training/tv_loss',tv_loss_sum/train_iter,i)
writer.add_scalar('training/psnr',psnr_sum/len(training_set) ,i)
logging.info('Dumped tensorboard summaries for epoch %4d'%(i))
## VALIDATION
if ((i+1) % 2 == 0) or ((i+1) == (start_epoch+num_epochs)):
logging.info('Starting validation')
val_iter = 0
val_loss_sum = 0.
val_psnr_sum = 0.
val_ssim_sum = 0.
invNet.eval()
uNet.eval()
with torch.no_grad():
for gt_vid in validation_generator:
gt_vid = gt_vid.cuda()
if not args.two_bucket:
b1 = c2b(gt_vid) # (N,1,H,W)
# b1 = torch.mean(gt_vid, dim=1, keepdim=True)
interm_vid = invNet(b1)
else:
b1, b0 = c2b(gt_vid)
b_stack = torch.cat([b1,b0], dim=1)
interm_vid = invNet(b_stack)
highres_vid = uNet(interm_vid) # (N,9,H,W)
val_psnr_sum += utils.compute_psnr(highres_vid, gt_vid).item()
val_ssim_sum += utils.compute_ssim(highres_vid, gt_vid).item()
psnr = utils.compute_psnr(highres_vid, gt_vid).item() / gt_vid.shape[0]
ssim = utils.compute_ssim(highres_vid, gt_vid).item() / gt_vid.shape[0]
## loss
if args.intermediate:
interm_loss = utils.weighted_L1loss(interm_vid, gt_vid).item()
final_loss = utils.weighted_L1loss(highres_vid, gt_vid).item()
tv_loss = utils.gradx(highres_vid).abs().mean().item() + utils.grady(highres_vid).abs().mean().item()
if args.intermediate:
val_loss_sum += final_loss + 0.1*tv_loss + 0.5*interm_loss
else:
val_loss_sum += final_loss + 0.1*tv_loss
if val_iter % 1000 == 0:
print('In val iter %d'%(val_iter))
val_iter += 1
logging.info('Total val iterations: %d'%(val_iter))
logging.info('Finished validation with loss: %.4f psnr: %.4f ssim: %.4f'
%(val_loss_sum/val_iter, val_psnr_sum/len(validation_set), val_ssim_sum/len(validation_set)))
scheduler.step(val_loss_sum)
invNet.train()
uNet.train()
## dump tensorboard summaries
writer.add_scalar('validation/loss',val_loss_sum/val_iter,i)
writer.add_scalar('validation/psnr',val_psnr_sum/len(validation_set),i)
writer.add_scalar('validation/ssim',val_ssim_sum/len(validation_set),i)
## CHECKPOINT
if ((i+1) % 50 == 0) or ((i+1) == (start_epoch+num_epochs)):
utils.save_checkpoint(state={'epoch': i,
'invnet_state_dict': invNet.state_dict(),
'unet_state_dict': uNet.state_dict(),
'c2b_state_dict': c2b.state_dict(),
'opt_state_dict': optimizer.state_dict()},
save_path=os.path.join(save_path, 'model'),
filename='model_%.6d.pth'%(i))
logging.info('Saved checkpoint for epoch {}'.format(i))
logger.writer.flush()
logging.info('Finished training')