-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
187 lines (155 loc) · 8.27 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
from __future__ import division
import tensorflow as tf
import tensorflow.contrib.slim as slim
import numpy as np
import matplotlib.pyplot as plt
from networks import *
from utils import *
import argparse, sys
parser = argparse.ArgumentParser()
parser.add_argument("--task", default="pre-trained", help="path to folder containing the model")
parser.add_argument("--data_dir", default="./Dataset/inpainting/", help="path to real dataset")
parser.add_argument("--save_model_freq", default=1, type=int, help="frequency to save model")
parser.add_argument("--use_gpu", default=0, type=int, help="which gpu to use")
parser.add_argument("--use_da", default=0.0, type=float, help="[0~1], the precentage of synthesized dataset")
parser.add_argument("--is_hyper", default=1, type=int, help="use hypercolumn or not")
parser.add_argument("--is_training", default=1, help="training or testing")
parser.add_argument("--continue_training", action="store_true",
help="search for checkpoint in the subfolder specified by `task` argument")
parser.add_argument("--vgg_19_path", default='Models/imagenet-vgg-verydeep-19.mat', type=str,
help="path to vgg 19 path model")
ARGS = parser.parse_args()
task = 'logs/' + ARGS.task
is_training = ARGS.is_training == 1
continue_training = False
hyper = ARGS.is_hyper == 1
current_best = 65535
maxepoch = 151
EPS = 1e-12
channel = 64 # number of feature channels to build the model, set to 64
# vgg_19_path = scipy.io.loadmat('./Models/imagenet-vgg-verydeep-19.mat')
vgg_19_path = ARGS.vgg_19_path
test_w, test_h = 640, 480
if ARGS.use_gpu < 0:
os.environ['CUDA_VISIBLE_DEVICES'] = ''
else:
os.environ['CUDA_VISIBLE_DEVICES'] = str(ARGS.use_gpu)
train_real_root = [ARGS.data_dir]
# set up the model and define the graph
with tf.variable_scope(tf.get_variable_scope()):
input = tf.placeholder(tf.float32, shape=[None, None, None, 3])
target = tf.placeholder(tf.float32, shape=[None, None, None, 3])
gtmask = tf.placeholder(tf.float32, shape=[None, None, None, 1])
# build the model
shadow_free_image, predicted_mask = build_aggasatt_joint(input, channel, vgg_19_path=vgg_19_path)
loss_mask = tf.reduce_mean(tf.keras.losses.binary_crossentropy(gtmask, tf.nn.sigmoid(predicted_mask)))
# Perceptual Loss
loss_percep = compute_percep_loss(shadow_free_image, target, vgg_19_path=vgg_19_path)
# loss_percep = tf.reduce_mean(tf.abs(shadow_free_image - target))
# Adversarial Loss
with tf.variable_scope("discriminator"):
predict_real, pred_real_dict = build_discriminator(input, target)
with tf.variable_scope("discriminator", reuse=True):
predict_fake, pred_fake_dict = build_discriminator(input, shadow_free_image)
d_loss = (tf.reduce_mean(-(tf.log(predict_real + EPS) + tf.log(1 - predict_fake + EPS)))) * 0.5
g_loss = tf.reduce_mean(-tf.log(predict_fake + EPS))
loss = loss_percep * 0.2 + loss_mask
train_vars = tf.trainable_variables()
d_vars = [var for var in train_vars if 'discriminator' in var.name]
g_vars = [var for var in train_vars if 'g_' in var.name]
g_opt = tf.train.AdamOptimizer(learning_rate=0.0002).minimize(loss * 100 + g_loss,
var_list=g_vars) # optimizer for the generator
d_opt = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(d_loss,
var_list=d_vars) # optimizer for the discriminator
for var in tf.trainable_variables():
print("Listing trainable variables ... ")
print(var)
saver = tf.train.Saver(max_to_keep=None)
if not os.path.isdir(task):
os.makedirs(task)
######### Session #########
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
ckpt = tf.train.get_checkpoint_state(task)
print("[i] contain checkpoint: ", ckpt)
if ckpt and continue_training:
saver_restore = tf.train.Saver([var for var in tf.trainable_variables()])
print('loaded ' + ckpt.model_checkpoint_path)
saver_restore.restore(sess, ckpt.model_checkpoint_path)
# test doesn't need to load discriminator
elif not is_training:
saver_restore = tf.train.Saver([var for var in tf.trainable_variables() if 'discriminator' not in var.name])
print('loaded ' + ckpt.model_checkpoint_path)
saver_restore.restore(sess, ckpt.model_checkpoint_path)
sys.stdout.flush()
if is_training:
# please follow the dataset directory setup in README
input_images_path = prepare_data(train_real_root, stage=['train_A']) # no reflection ground truth for real images
syn_images = prepare_data(train_real_root, stage=['synC'])
print("[i] Total %d training images, first path of real image is %s." % (
len(input_images_path), input_images_path[0]))
num_train = len(input_images_path) + len(syn_images)
all_l = np.zeros(num_train, dtype=float)
all_percep = np.zeros(num_train, dtype=float)
all_grad = np.zeros(num_train, dtype=float)
all_g = np.zeros(num_train, dtype=float)
for epoch in range(1, maxepoch):
input_images_ids, target_images_ids = [None] * num_train, [None] * num_train
epoch_st = time.time()
if os.path.isdir("%s/%04d" % (task, epoch)):
continue
cnt = 0
for id in np.random.permutation(num_train):
st = time.time()
if input_images_ids[id] is None:
_id = id % len(input_images_path)
running_idx = (epoch - 1) * num_train + cnt
magic = np.random.rand()
current_img_id = ''
inputimg = cv2.imread(input_images_path[_id], -1)
neww = np.random.randint(256, 300) # w is the longer width[]
newh = round((neww / inputimg.shape[1]) * inputimg.shape[0])
if magic < ARGS.use_da: # choose from fake images
is_syn = True
current_img_id = random.sample(syn_images, 1)[0]
iminput, imtarget, maskgt = parpare_image_syn(current_img_id, (neww, newh), da=True, stage='synC')
else:
is_syn = False
current_img_id = input_images_path[_id]
iminput, imtarget, maskgt = parpare_image(current_img_id, (neww, newh), da=True,
stage=['_M', '_C', '_B'])
# alternate training, update discriminator every two iterations
if cnt % 2 == 0:
fetch_list = [d_opt]
# update D
_ = sess.run(fetch_list, feed_dict={input: iminput, target: imtarget, gtmask: maskgt})
# update G
fetch_list = [g_opt, shadow_free_image, d_loss, g_loss, loss, loss_percep]
_, imoutput, current_d, current_g, current, current_percep = \
sess.run(fetch_list, feed_dict={input: iminput, target: imtarget, gtmask: maskgt})
all_l[id] = current
all_percep[id] = current_percep
all_g[id] = current_g
g_mean = np.mean(all_g[np.where(all_g)])
if running_idx % 5 == 0:
print("iter: %d %d || D: %.2f || G: %.2f %.2f || mean all: %.2f || percp: %.2f %.2f || time: %.2f" %
(epoch, cnt, current_d, current_g, g_mean,
np.mean(all_l[np.where(all_l)]),
current_percep, np.mean(all_percep[np.where(all_percep)]),
time.time() - st))
fileid = os.path.splitext(os.path.basename(input_images_path[_id]))[0]
imoutput = decode_image(imoutput)
iminput = decode_image(iminput)
imtarget = decode_image(imtarget)
cv2.imwrite("%s/%s_%s.png" % (task, running_idx, fileid),
np.concatenate((iminput, imoutput, imtarget), axis=1))
cnt += 1
input_images_ids[id] = 1.
target_images_ids[id] = 1.
print('epoch %s use %s' % (epoch, time.time() - epoch_st))
# save model and images every epoch
if epoch % ARGS.save_model_freq == 0:
saver.save(sess, "%s/lasted_model.ckpt" % task)
sys.stdout.flush()