-
Notifications
You must be signed in to change notification settings - Fork 3
/
train_denoise_LRD.py
191 lines (161 loc) · 7.9 KB
/
train_denoise_LRD.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
import gc
import time
import torch
import os
import sys
import torch.backends.cudnn as cudnn
import argparse
from tqdm import tqdm, trange
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from datasets.Denoise_dataset import NPYDataset_LRD
from models.denoise_model import UNetSeeInDark
import torch.nn.functional as F
from utils import get_config
from raw_process.SID_post_process import tensor2im, quality_assess
import numpy as np
# parse options
parser = argparse.ArgumentParser(description='Low-Light Raw Noise Synthesis args setting')
parser.add_argument('--config', type=str, default='./configs/Noise_Synthesis_LRD.yaml', help='Path to the config file.')
parser.add_argument('--output_path', type=str, default='./Denoise_results_LRD', help="outputs path")
parser.add_argument("--resume", action="store_true")
parser.add_argument('--seed', type=int, default=3407, help="random seed")
parser.add_argument('--gpu_ids', type=str, default='0,1', help='gpu ids: e.g. 0 0,1,2, 0,2. use -1 for CPU')
parser.add_argument('--batch_size', type=int, default=1, help='input batch size')
parser.add_argument('--patch_size', type=int, default=512, help='crop patch size')
parser.add_argument('--log_ids', type=str, default='exp0', help='log ID')
parser.add_argument('--height', type=int, default=512, help='input height size')
parser.add_argument('--width', type=int, default=512, help='input width size')
parser.add_argument('--learning_rate', type=int, default=0.0002, help='input width size')
parser.add_argument('--synthesis_root', default='/home/dancer/train_datasets/Generated_train_samples_LRD/')
parser.add_argument('--max_epoch', type=int, default=200)
parser.add_argument('--image_save_epoch', type=int, default=10)
parser.add_argument('--snapshot_save_epoch', type=int, default=10)
opts = parser.parse_args()
os.environ['CUDA_VISIBLE_DEVICES'] = '1'
torch.manual_seed(opts.seed)
torch.cuda.manual_seed(opts.seed)
config = get_config(opts.config)
def stack2raw_RGGB(var):
_, h, w = var.shape
if var.is_cuda:
res = torch.cuda.FloatTensor(h * 2, w * 2)
else:
res = torch.FloatTensor(h * 2, w * 2)
res[0::2, 0::2] = var[0]
res[0::2, 1::2] = var[1]
res[1::2, 0::2] = var[2]
res[1::2, 1::2] = var[3]
return res
def raw2stack_RGGB(var):
h, w = var.shape
if var.is_cuda:
res = torch.cuda.FloatTensor(4, h // 2, w // 2).fill_(0)
else:
res = torch.FloatTensor(4, h // 2, w // 2).fill_(0)
res[0] = var[0::2, 0::2]
res[1] = var[0::2, 1::2]
res[2] = var[1::2, 0::2]
res[3] = var[1::2, 1::2]
return res
def forward_patches(model, input_img, patch_size=256, pad=32):
shift = patch_size - pad * 2
input_img = F.pad(input_img, (pad, pad, pad, pad), mode='reflect')
denoised = torch.zeros_like(input_img)
_, _, H, W = input_img.shape
for i in np.arange(0, H, shift):
for j in np.arange(0, W, shift):
h_end, w_end = min(i + patch_size, H), min(j + patch_size, W)
h_start, w_start = h_end - patch_size, w_end - patch_size
input_var = input_img[..., h_start: h_end, w_start: w_end]
with torch.no_grad():
out_var = model(input_var)
denoised[..., h_start + pad: h_end - pad, w_start + pad: w_end - pad] = \
out_var[..., pad:-pad, pad:-pad]
denoised = denoised[..., pad:-pad, pad:-pad]
denoised = denoised.clip(0, 1)
return denoised
def main():
cudnn.benchmark = True
# Loss functions
device = torch.device("cuda" if len(opts.gpu_ids) > 0 else "cpu")
criterion_pixel = torch.nn.L1Loss().to(device)
# model set and data loader
model = UNetSeeInDark(in_channels=4, out_channels=4).to(device)
# set logger and output folder
writer = SummaryWriter(os.path.join(opts.output_path + "/logs/" + opts.log_ids))
output_directory = os.path.join(opts.output_path + "/outputs")
# set checkpoint folder
checkpoint_directory = os.path.join(output_directory, 'checkpoints')
if not os.path.exists(checkpoint_directory):
print("Creating directory: {}".format(checkpoint_directory))
os.makedirs(checkpoint_directory)
# set image folder
image_directory = os.path.join(output_directory, 'images')
if not os.path.exists(image_directory):
print("Creating directory: {}".format(image_directory))
os.makedirs(image_directory)
train_set = NPYDataset_LRD(opts.synthesis_root, config['test_img_list'], mode='train', patch_size=opts.patch_size, crop=True)
train_loader = torch.utils.data.DataLoader(train_set, batch_size=opts.batch_size, shuffle=True, drop_last=True,
num_workers=0, pin_memory=True)
val_set = NPYDataset_LRD(config['LRD_root'], config['test_img_list'], mode='test', patch_size=opts.patch_size, crop=False)
val_loader = torch.utils.data.DataLoader(val_set, batch_size=1, shuffle=False, num_workers=0, pin_memory=True)
epoch_iter = 0
total_iters = 0 # the total number of training iterations
f = open("%s/Metrics_results.txt" % output_directory, 'a')
f.truncate(0)
print('start training')
print(opts.synthesis_root)
# init_weights(model, init_type='kaiming')
optim = torch.optim.Adam(model.parameters(), lr=opts.learning_rate, betas=(0.9, 0.999))
# scheduler = torch.optim.lr_scheduler.MultiStepLR(optim, milestones=[15000, 25000], gamma=0.5)
scheduler = torch.optim.lr_scheduler.MultiStepLR(optim, milestones=[100, 180], gamma=0.5)
model.train()
for epoch in trange(epoch_iter, opts.max_epoch, desc='Train', ncols=80):
epoch_start_time = time.time() # timer for entire epoch
for batch_idx, data in tqdm(enumerate(train_loader), total=len(train_loader),
desc='Train epoch={}, iter={}'.format(epoch_iter, total_iters), ncols=80,
leave=False):
total_iters += 1
gc.collect()
clean_raw = data['clean'].to(device)
noise_raw = data['noisy'].to(device)
optim.zero_grad()
denoise_raw = model(noise_raw)
loss = criterion_pixel(denoise_raw, clean_raw)
loss.backward()
optim.step()
writer.add_scalar('Training/loss', loss, epoch + 1)
scheduler.step()
if (epoch + 1) % opts.image_save_epoch == 0:
print("[*] Evaluating for phase train / epoch %d..." % (epoch + 1))
psnr = 0
ssim = 0
count = 0
for batch_idx, data in tqdm(enumerate(val_loader), total=len(val_loader), desc='Valid', ncols=80,
leave=False):
gc.collect()
clean_raw = data['clean'].to(device)
noise_raw = data['noisy'].to(device)
clean_path = data['clean_path'][0]
model.eval()
with torch.no_grad():
denoise_raw = forward_patches(model, noise_raw)
clean_numpy = tensor2im(clean_raw)
denoised_numpy = tensor2im(denoise_raw)
res = quality_assess(denoised_numpy, clean_numpy, data_range=255)
psnr += res['PSNR']
ssim += res['SSIM']
count += 1
metrics_data = ["Epoch: %s, PSNR: %s, SSIM: %s" % (epoch + 1, psnr / count, ssim / count), '\n']
f.writelines(metrics_data)
print("===> Iteration[{}]: psnr: {}, ssim:{}".format(epoch + 1, psnr / count, ssim / count))
if (epoch + 1) % opts.snapshot_save_epoch == 0:
torch.save(model.state_dict(), checkpoint_directory + '/check_' + str(epoch + 1) + '.pth')
if epoch + 1 >= opts.max_epoch:
writer.close()
end_time = time.time() - epoch_start_time
print("Finish training in [%4.4f]" % end_time)
sys.exit('Finish training')
if __name__ == '__main__':
main()