-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_filter.py
executable file
·224 lines (172 loc) · 8.39 KB
/
train_filter.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
import os
import time
import argparse
import numpy as np
import cv2
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
from tensorboardX import SummaryWriter
from dataset import MRIDataset
from env import Env
from model import MyFcn
from pixel_wise_a2c import PixelWiseA2C
from utils import adjust_learning_rate as adjust_learning_rate
from utils import Config as Config
from mri_utils import PSNR, SSIM
from filter_model import FilterModel
def parse():
parser = argparse.ArgumentParser()
parser.add_argument('--config', default='config.yml', type=str,
dest='config', help='to set the parameters')
parser.add_argument('--log_dir', default='log', type=str,
dest='log_dir', help='the root of log')
parser.add_argument('--gpu', default=[0, 1], nargs='+', type=int,
dest='gpu', help='the gpu used')
parser.add_argument('--root', default='/home/lwt/MRI_RL/DAGAN/data/MICCAI13_RL/', type=str,
dest='root', help='the root of images')
#parser.add_argument('--train_dir', nargs='+', type=str,
# dest='train_dir', help='the path of train file')
return parser.parse_args()
def computePSNR(o_, p_, i_):
return PSNR(o_, p_), PSNR(o_, i_)
def computeSSIM(o_, p_, i_):
return SSIM(o_, p_), SSIM(o_, i_)
def test(model, a2c, config, args, **kwargs):
env = Env(config.move_range)
env.set_param(**kwargs)
test_loader = torch.utils.data.DataLoader(
dataset = MRIDataset(root=args.root, image_set='test', transform=False),
batch_size=config.batch_size, shuffle=False,
num_workers=config.workers, pin_memory=False)
start = time.time()
reward_sum = 0
PSNR_list = []
SSIM_list = []
for i, (ori_image, image) in enumerate(test_loader):
ori_image = ori_image.numpy()
image = image.numpy()
previous_image = image.copy()
env.reset(ori_image=ori_image, image=image)
for j in range(config.episode_len):
image_input = Variable(torch.from_numpy(image).cuda(), volatile=True)
pout, vout = model(image_input)
actions = a2c.act(pout, deterministic=True)
image, reward = env.step(actions)
image = np.clip(image, 0, 1)
reward_sum += np.mean(reward)
for ii in range(image.shape[0]):
PSNR_list.append(computePSNR(ori_image[ii, 0], previous_image[ii, 0], image[ii, 0]))
SSIM_list.append(computeSSIM(ori_image[ii, 0], previous_image[ii, 0], image[ii, 0]))
if i == 100:
i += 1
actions = actions.astype(np.uint8)
total = actions.size
a0 = actions[0]
B = image[0, 0].copy()
for a in range(config.num_actions):
print(a, 'actions', np.sum(actions==a) / total)
A = np.zeros((*B.shape, 3))
#print(A, B)
A[..., 0] += B * 255
A[..., 1] += B * 255
A[..., 2] += B * 255
A[a0==a, 0] += 250
cv2.imwrite('actions/'+str(a)+'.jpg', A)
break
psnr_res = np.mean(np.array(PSNR_list), axis=0)
ssim_res = np.mean(np.array(SSIM_list), axis=0)
print('PSNR', psnr_res)
print('SSIM', ssim_res)
avg_reward = reward_sum / i
print('test finished: reward ', avg_reward)
return avg_reward, psnr_res, ssim_res
def train_filter(model, a2c):
args = parse()
config = Config('filter_config.yml')
torch.backends.cudnn.benchmark = True
#log_dir = os.path.expanduser(args.log_dir)
env = Env(config.move_range, reward_method=config.reward_method)
#model = MyFcn(num_actions=config.num_actions)
#model = torch.nn.DataParallel(model, device_ids=args.gpu).cuda()
#a2c = PixelWiseA2C(model=None, optimizer=None, t_max=100000, gamma=config.gamma, beta=1e-2)
filter_model = FilterModel()
filter_model = filter_model.cuda()
optimizer = torch.optim.SGD(filter_model.parameters(), config.base_lr, momentum=0)
train_loader = torch.utils.data.DataLoader(
dataset = MRIDataset(root=args.root, image_set='train', transform=True),
batch_size=config.batch_size, shuffle=True,
num_workers=config.workers, pin_memory=False)
writer = SummaryWriter('./filter_logs')
#for lp in [0, 0.01, 0.02, 0.08, 0.09, 0.095, 0.1, 0.105, 0.11]:
# print('lp', lp)
# avg_reward, psnr_res, ssim_res = test(model, a2c, config, args, laplace_param=lp)
for sobel_v1 in [0, 0.01, 0.02, 0.08, 0.09, 0.095, 0.1, 0.105, 0.11]:
print('sobel_v1', sobel_v1)
avg_reward, psnr_res, ssim_res = test(model, a2c, config, args, sobel_v1_param=sobel_v1)
episodes = 0
while episodes < config.num_episodes:
for i, (ori_image, image) in enumerate(train_loader):
learning_rate = adjust_learning_rate(optimizer, episodes, config.base_lr, policy=config.lr_policy, policy_parameter=config.policy_parameter)
ori_image_input = Variable(ori_image).cuda()
ori_image = ori_image.numpy()
image = image.numpy()
env.reset(ori_image=ori_image, image=image)
reward = np.zeros((1))
loss = Variable(torch.zeros(1)).cuda()
for j in range(config.episode_len):
image_input = Variable(torch.from_numpy(image).cuda(), volatile=True)
#reward_input = Variable(torch.from_numpy(reward).cuda())
pout, vout = model(image_input)
actions = a2c.act(pout, deterministic=True)
#print(actions)
mask_laplace = (actions==6)[:, np.newaxis]
action_mask = Variable(torch.from_numpy(mask_laplace.astype(np.float32))).cuda()
print(action_mask.mean())
xxx
image_input = Variable(torch.from_numpy(image).cuda())
output_laplace = filter_model(image_input)
ll = torch.abs(ori_image_input - output_laplace) * action_mask
#print(ll.shape)
loss += ll.mean()
previous_image = image
image, reward = env.step(actions)
#print(ori_image_input.shape, action_mask.shape, actions.shape, output_laplace.shape)
if i % 40 == 0:
print('reward', j, np.mean(reward))
print(computeSSIM(ori_image[0, 0], previous_image[0, 0], image[0, 0]))
print('diff', (torch.abs(ori_image_input.data - torch.from_numpy(image).cuda()) - torch.abs(ori_image_input.data - output_laplace.data) * action_mask.data).mean())
image = np.where(mask_laplace, output_laplace.cpu().data.numpy(), image)
image = np.clip(image, 0, 1)
#loss = a2c.stop_episode_and_compute_loss(reward=Variable(torch.from_numpy(reward).cuda()), done=True) / config.iter_size
loss.backward()
if not(episodes % config.iter_size):
optimizer.step()
optimizer.zero_grad()
lw = float(filter_model.state_dict()['conv_laplace.weight'].cpu().numpy())
print('loss:', ll.mean(), 'weight:', lw)
writer.add_scalar('weight', lw, episodes)
episodes += 1
if episodes % config.display == 0:
print('episode: ', episodes, 'loss: ', loss.data)
if not(episodes % config.save_episodes):
#torch.save(model.module.state_dict(), 'model/' + str(episodes) + '.pth')
print('model saved')
if not(episodes % config.test_episodes):
avg_reward, psnr_res, ssim_res = test(model, a2c, config, args)
#writer.add_scalar('psnr_ref', psnr_res[0], episodes)
#writer.add_scalar('psnr', psnr_res[1], episodes)
#writer.add_scalar('ssim_ref', ssim_res[0], episodes)
#writer.add_scalar('ssim', ssim_res[1], episodes)
if episodes == config.num_episodes:
writer.close()
break
if __name__ == "__main__":
os.environ['CUDA_VISIBLE_DEVICES'] = '1'
model = MyFcn(num_actions=13)
model.load_state_dict(torch.load('model/16000.pth_0718'))
model = model.cuda()
a2c = PixelWiseA2C(model=None, optimizer=None, t_max=100000, gamma=0.5, beta=1e-2)
train_filter(model, a2c)