-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
222 lines (168 loc) · 11.5 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
import os
import sys
import tensorflow as tf
import tensorflow.contrib.slim as slim
import CVCMFFNet
import scipy.io as scio
import time
import glob
import numpy as np
from PIL import Image
import math
from time import sleep
from PIL import ImageEnhance
from tqdm import tqdm
from tqdm import trange
import warnings
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
WORKING_DIR = os.getcwd()
TRAINING_DIR = os.path.join(WORKING_DIR, 'Data', 'Training')
TEST_DIR = os.path.join(WORKING_DIR, 'Data', 'Test1')
ROOT_LOG_DIR = os.path.join(WORKING_DIR, 'Output')
RUN_NAME = "model"
LOG_DIR = os.path.join(ROOT_LOG_DIR, RUN_NAME)
TRAIN_WRITER_DIR = os.path.join(LOG_DIR, 'Training')
TEST_WRITER_DIR = os.path.join(LOG_DIR, 'Test1')
CHECKPOINT_FN = 'model.ckpt'
CHECKPOINT_FL = os.path.join(LOG_DIR, CHECKPOINT_FN)
BATCH_NORM_DECAY = 0.95
MAX_STEPS = 30000
BATCH_SIZE = 8
SAVE_INTERVAL = 200
def main():
training_data = CVCMFFNet.GetData(TRAINING_DIR)
test_data = CVCMFFNet.GetData(TEST_DIR)
g = tf.Graph()
with g.as_default():
images1_real, images1_imag, images2_real, images2_imag, ang, labels, is_training = CVCMFFNet.placeholder_inputs(batch_size=BATCH_SIZE)
arg_scope = CVCMFFNet.inference_scope(is_training=True, batch_norm_decay=BATCH_NORM_DECAY)
with slim.arg_scope(arg_scope):
#logits, freal_layer1, fimag_layer1, freal_layer2, fimag_layer2, freal_layer3, fimag_layer3, freal_layer4, fimag_layer4, freal_layer5, fimag_layer5, freal_conv_1x1, fimag_conv_1x1, freal_conv_3x3_1, fimag_conv_3x3_1, freal_conv_3x3_2, fimag_conv_3x3_2, freal_conv_3x3_3, fimag_conv_3x3_3, freal_mix, fimag_mix, fang_2, fang_4, fang_6, fang_8, fang_out = SegNet.inference(images1_real, images1_imag, images2_real, images2_imag, ang, class_inc_bg=3)
logits = CVCMFFNet.inference(images1_real, images1_imag, images2_real, images2_imag, ang, class_inc_bg=3)
CVCMFFNet.add_output_images(images_real=images1_real, images_imag=images1_imag, logits=logits, labels=labels)
loss = CVCMFFNet.loss_calc(logits=logits, labels=labels)
train_op, global_step = CVCMFFNet.training(loss=loss, learning_rate=1e-05)
total_parameters = CVCMFFNet.count()
flops = CVCMFFNet.count_flops(g)
accuracy, out_label = CVCMFFNet.evaluation(logits=logits, labels=labels)
iou_class0, iou_class1, iou_class2, pa_0, pa_1, pa_2 = CVCMFFNet.iou_and_pa(logits=logits, labels=labels)
summary = tf.summary.merge_all()
init = tf.global_variables_initializer()
saver = tf.train.Saver([x for x in tf.global_variables() if 'Adam' not in x.name])
sm = tf.train.SessionManager()
with sm.prepare_session("", init_op=init, saver=saver, checkpoint_dir=LOG_DIR) as sess:
sess.run(tf.variables_initializer([x for x in tf.global_variables() if 'Adam' in x.name]))
train_writer = tf.summary.FileWriter(TRAIN_WRITER_DIR, sess.graph)
test_writer = tf.summary.FileWriter(TEST_WRITER_DIR)
global_step_value, = sess.run([global_step])
print("CVCMFF Net is Ready, Last Trained Iteration was:", global_step_value)
print("Train Start...")
sleep(0.5)
ID = 0
for step in range(global_step_value + 1, global_step_value + MAX_STEPS + 1, SAVE_INTERVAL):
with trange(SAVE_INTERVAL, ncols=120, ascii='=>', mininterval=0.05) as t:
for sub_step in t:
images1_real_batch, images1_imag_batch, images2_real_batch, images2_imag_batch, ang_batch, labels_batch = training_data.next_batch(BATCH_SIZE)
train_feed_dict = {images1_real: images1_real_batch,
images1_imag: images1_imag_batch,
images2_real: images2_real_batch,
images2_imag: images2_imag_batch,
ang: ang_batch,
labels: labels_batch,
is_training: True}
#_, train_loss_value, train_accuracy_value, freal_layer1, fimag_layer1, freal_layer2, fimag_layer2, freal_layer3, fimag_layer3, freal_layer4, fimag_layer4, freal_layer5, fimag_layer5, freal_conv_1x1, fimag_conv_1x1, freal_conv_3x3_1, fimag_conv_3x3_1, freal_conv_3x3_2, fimag_conv_3x3_2, freal_conv_3x3_3, fimag_conv_3x3_3, freal_mix, fimag_mix, fang_2, fang_4, fang_6, fang_8, fang_out, train_summary_str = sess.run([train_op, loss, accuracy, freal_layer1, fimag_layer1, freal_layer2, fimag_layer2, freal_layer3, fimag_layer3, freal_layer4, fimag_layer4, freal_layer5, fimag_layer5, freal_conv_1x1, fimag_conv_1x1, freal_conv_3x3_1, fimag_conv_3x3_1, freal_conv_3x3_2, fimag_conv_3x3_2, freal_conv_3x3_3, fimag_conv_3x3_3, freal_mix, fimag_mix, fang_2, fang_4, fang_6, fang_8, fang_out, summary], feed_dict=train_feed_dict)
_, train_loss_value, train_accuracy_value, train_iou_class0, train_iou_class1, train_iou_class2, train_summary_str = sess.run([train_op, loss, accuracy, iou_class0, iou_class1, iou_class2, summary], feed_dict=train_feed_dict)
train_mean_iou = (train_iou_class0 + train_iou_class1 + train_iou_class2) / 3
step_now = step + sub_step
t.set_description('Training')
t.set_postfix(Iteration=step_now, Loss=train_loss_value, Accuarcy=train_accuracy_value, MIoU=train_mean_iou)
sleep(0.5)
train_writer.add_summary(train_summary_str, step)
train_writer.flush()
images1_real_batch, images1_imag_batch, images2_real_batch, images2_imag_batch, ang_batch, labels_batch = test_data.next_batch(BATCH_SIZE)
test_feed_dict = {images1_real: images1_real_batch,
images1_imag: images1_imag_batch,
images2_real: images2_real_batch,
images2_imag: images2_imag_batch,
ang: ang_batch,
labels: labels_batch,
is_training: False}
test_loss_value, test_accuracy_value, test_iou_class0, test_iou_class1, test_iou_class2, test_pa_0, test_pa_1, test_pa_2, test_summary_str, test_label, img_real, img_imag = sess.run([loss, accuracy, iou_class0, iou_class1, iou_class2, pa_0, pa_1, pa_2, summary, out_label, images1_real, images1_imag], feed_dict=test_feed_dict)
test_mean_iou = (test_iou_class0 + test_iou_class1 + test_iou_class2) / 3
test_mpa = (test_pa_1 + test_pa_1 + test_pa_2) / 3
'''
# Feature visualization
#scio.savemat("images1_real.mat", {"images1_real": images1_real})
#scio.savemat("images1_imag.mat", {"images1_imag": images1_imag})
#scio.savemat("images2_real.mat", {"images2_real": images2_real})
#scio.savemat("images2_imag.mat", {"images2_imag": images2_imag})
#scio.savemat("ang.mat", {"ang": ang})
scio.savemat("freal_layer1.mat", {"freal_layer1": freal_layer1})
scio.savemat("fimag_layer1.mat", {"fimag_layer1": fimag_layer1})
scio.savemat("freal_layer2.mat", {"freal_layer2": freal_layer2})
scio.savemat("fimag_layer2.mat", {"fimag_layer2": fimag_layer2})
scio.savemat("freal_layer3.mat", {"freal_layer3": freal_layer3})
scio.savemat("fimag_layer3.mat", {"fimag_layer3": fimag_layer3})
scio.savemat("freal_layer4.mat", {"freal_layer4": freal_layer4})
scio.savemat("fimag_layer4.mat", {"fimag_layer4": fimag_layer4})
scio.savemat("freal_layer5.mat", {"freal_layer5": freal_layer5})
scio.savemat("fimag_layer5.mat", {"fimag_layer5": fimag_layer5})
scio.savemat("freal_conv_1x1.mat", {"freal_conv_1x1": freal_conv_1x1})
scio.savemat("fimag_conv_1x1.mat", {"fimag_conv_1x1": fimag_conv_1x1})
scio.savemat("freal_conv_3x3_1.mat", {"freal_conv_3x3_1": freal_conv_3x3_1})
scio.savemat("fimag_conv_3x3_1.mat", {"fimag_conv_3x3_1": fimag_conv_3x3_1})
scio.savemat("freal_conv_3x3_2.mat", {"freal_conv_3x3_2": freal_conv_3x3_2})
scio.savemat("fimag_conv_3x3_2.mat", {"fimag_conv_3x3_2": fimag_conv_3x3_2})
scio.savemat("freal_conv_3x3_3.mat", {"freal_conv_3x3_3": freal_conv_3x3_3})
scio.savemat("fimag_conv_3x3_3.mat", {"fimag_conv_3x3_3": fimag_conv_3x3_3})
scio.savemat("freal_mix.mat", {"freal_mix": freal_mix})
scio.savemat("fimag_mix.mat", {"fimag_mix": fimag_mix})
scio.savemat("fang_2.mat", {"fang_2": fang_2})
scio.savemat("fang_4.mat", {"fang_4": fang_4})
scio.savemat("fang_6.mat", {"fang_6": fang_6})
scio.savemat("fang_8.mat", {"fang_8": fang_8})
scio.savemat("fang_out.mat", {"fang_out": fang_out})
'''
print("Test Loss:", test_loss_value, "Test accuracy:", test_accuracy_value)
print("Test IoU_class0:", test_iou_class0, "Test IoU_class1:", test_iou_class1, "Test IoU_class2:", test_iou_class2, "Test MIoU:", test_mean_iou)
print("Test Pix_accuary0:", test_pa_0, "Test Pix_accuary1:", test_pa_1, "Test Pix_accuary2:", test_pa_2, "Test MPA:", test_mpa)
scio.savemat("{ID}".format(ID=ID) + ".mat", {"label": test_label})
test_writer.add_summary(test_summary_str, step)
test_writer.flush()
# testlabel2outtag
data_path = "{ID}".format(ID=ID) + ".mat"
output_path = "./outtag/"
data = scio.loadmat(data_path)
label = np.array(data['label'])
R = np.zeros([256, 256, 2])
G = np.zeros([256, 256, 2])
B = np.zeros([256, 256, 2])
for k in range(2):
for i in range(256):
for j in range(256):
if (label[i, j, k] == 0):
[R[i, j, k], G[i, j, k], B[i, j, k]] = [0, 0, 0]
if (label[i, j, k] == 1):
[R[i, j, k], G[i, j, k], B[i, j, k]] = [128, 0, 0]
if (label[i, j, k] == 2):
[R[i, j, k], G[i, j, k], B[i, j, k]] = [0, 128, 0]
# save outtag as png
for k in range(2):
R1 = Image.fromarray(R[:, :, k]).convert('L')
G1 = Image.fromarray(G[:, :, k]).convert('L')
B1 = Image.fromarray(B[:, :, k]).convert('L')
image = Image.merge("RGB", (R1, G1, B1))
image.save(output_path + str(ID + k) + ".png", 'png')
print("Outtag is OK")
ID = ID + 2
# delete tag.mat
for root, dirs, files in os.walk(WORKING_DIR):
for name in files:
if name.endswith(".mat"):
os.remove(os.path.join(root, name))
#print("Delete File: " + os.path.join(root, name))
saver.save(sess, CHECKPOINT_FL, global_step=step)
print("CHECKPOINT Saved")
print("================")
if __name__ == '__main__':
main()