-
Notifications
You must be signed in to change notification settings - Fork 1
/
utils.py
230 lines (178 loc) · 8.77 KB
/
utils.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
import numpy as np
import sys
import os
import urllib
from ops import relu, conv2d, deconv2d
from scipy.io import loadmat
import numpy as np
import tensorflow as tf
from scipy.misc import imsave,imread
download_url="http://www.vlfeat.org/matconvnet/models/imagenet-vgg-verydeep-19.mat"
def download_model(download_link, file_name, expected_bytes=0):
if os.path.exists(file_name):
print "VGG19 model already exists"
return
file_name,_ = urllib.urlretrieve(download_link, file_name)
#if os.stat(file_name).st_size == expected_bytes :
# print "Download finished"
# return
#print "Download fail"
return
def _conv2d_relu(vgg_layers, prev_layer, layer, name):
W , b = _get_layer(vgg_layers, layer, name)
with tf.variable_scope(name):
W = tf.constant(W , name='weights')
b = tf.constant(b , name='bias')
conv = tf.nn.conv2d(prev_layer, W, strides=[1, 1, 1, 1], padding='SAME')
conv = tf.nn.bias_add(conv, b)
return relu(conv)
def _max_pool(prev_layer, name):
return tf.nn.max_pool(prev_layer, ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1], padding='SAME', name=name)
def _get_layer(vgg_layers, layer, expected_name):
W = vgg_layers[0][layer][0][0][2][0][0]
b = vgg_layers[0][layer][0][0][2][0][1]
name = vgg_layers[0][layer][0][0][0][0]
#assert name == expected_name
return W , b.reshape(b.size)
def vgg19(input_image, mat_file='vgg19model.mat'):
"""
Build vgg19 model with pretrained weight
"""
download_model(download_url, mat_file)
vgg_layers = loadmat(mat_file)['layers']
vgg_model = {}
vgg_model['conv1_1'] = _conv2d_relu(vgg_layers, input_image, 0, 'conv1_1')
vgg_model['conv1_2'] = _conv2d_relu(vgg_layers, vgg_model['conv1_1'], 2, 'conv1_2')
vgg_model['maxpool_1'] = _max_pool(vgg_model['conv1_2'], 'conv1_2')
vgg_model['conv2_1'] = _conv2d_relu(vgg_layers, vgg_model['maxpool_1'], 5, 'conv2_1')
vgg_model['conv2_2'] = _conv2d_relu(vgg_layers, vgg_model['conv2_1'], 7, 'conv2_2')
vgg_model['maxpool_2'] = _max_pool(vgg_model['conv2_2'], 'maxpool_2')
vgg_model['conv3_1'] = _conv2d_relu(vgg_layers, vgg_model['maxpool_2'], 10, 'conv3_1')
vgg_model['conv3_2'] = _conv2d_relu(vgg_layers, vgg_model['conv3_1'], 12, 'conv3_2')
vgg_model['conv3_3'] = _conv2d_relu(vgg_layers, vgg_model['conv3_2'], 14, 'conv3_3')
vgg_model['conv3_4'] = _conv2d_relu(vgg_layers, vgg_model['conv3_3'], 16, 'conv3_4')
vgg_model['maxpool_3'] = _max_pool(vgg_model['conv3_4'], 'maxpool_3')
vgg_model['conv4_1'] = _conv2d_relu(vgg_layers, vgg_model['maxpool_3'], 19, 'conv4_1')
vgg_model['conv4_2'] = _conv2d_relu(vgg_layers, vgg_model['conv4_1'], 21, 'conv4_1')
vgg_model['conv4_3'] = _conv2d_relu(vgg_layers, vgg_model['conv4_2'], 23, 'conv4_3')
vgg_model['conv4_4'] = _conv2d_relu(vgg_layers, vgg_model['conv4_3'], 25, 'conv4_4')
vgg_model['maxpool_4'] = _max_pool(vgg_model['conv4_4'], 'maxpool_4')
vgg_model['conv5_1'] = _conv2d_relu(vgg_layers, vgg_model['maxpool_4'], 28, 'conv5_1')
vgg_model['conv5_2'] = _conv2d_relu(vgg_layers, vgg_model['conv5_1'], 30, 'conv5_2')
vgg_model['conv5_3'] = _conv2d_relu(vgg_layers, vgg_model['conv5_2'], 32, 'conv5_3')
vgg_model['conv5_4'] = _conv2d_relu(vgg_layers, vgg_model['conv5_3'], 34, 'conv5_4')
vgg_model['maxpool_5'] = _max_pool(vgg_model['conv5_3'], 'maxpool_5')
return vgg_model
IMAGE_NET_MEAN = [123.68, 116.779, 103.939]
def fcn32s(input_image, batch_size):
mean = tf.constant(IMAGE_NET_MEAN)
input_image -= mean
fcn = vgg19(input_image)
fcn['fc6'] = conv2d(fcn['maxpool_5'], 4096, name='fc6', k_w=7, k_h=7)
fcn['relu6'] = relu(fcn['fc6'])
fcn['drop6'] = tf.nn.dropout(fcn['relu6'], 0.5)
fcn['fc7'] = conv2d(fcn['drop6'], 4096, name='fc7', k_w=1, k_h=1)
fcn['relu7'] = relu(fcn['fc7'])
#num_class = 6
fcn['score_fr'] = conv2d(fcn['relu6'], 6, name='score_fr', k_w=1, k_h=1)
fcn['upscore'] = deconv2d(fcn['score_fr'], [batch_size, 224, 224, 6], name='upscore')
fcn['predict'] = tf.argmax(fcn['upscore'], axis=3, name='predict')
fcn['predict'] = tf.reshape(fcn['predict'], shape=[batch_size, 224, 224, 1])
return fcn
def fcn8s(input_image, batch_size, ground_truth, learning_rate, IMAGE_SIZE=224):
input_height = input_width = IMAGE_SIZE
mean = tf.constant(IMAGE_NET_MEAN)
input_image -= mean
fcn = vgg19(input_image)
fcn['fc6'] = conv2d(fcn['maxpool_5'], 4096, name='fc6', k_w=7, k_h=7)
fcn['relu6'] = relu(fcn['fc6'])
fcn['drop6'] = tf.nn.dropout(fcn['relu6'], 0.5)
fcn['fc7'] = conv2d(fcn['drop6'], 4096, name='fc7', k_w=1, k_h=1)
fcn['relu7'] = relu(fcn['fc7'])
#num_class = 6
fcn['score_fr'] = conv2d(fcn['relu6'], 6, name='score_fr', k_w=1, k_h=1)
#fcn['upscore'] = deconv2d(fcn['score_fr'], [batch_size, 224, 224, 6], name='upscore')
#fcn['predict'] = tf.argmax(fcn['upscore'], axis=3, name='predict')
#fcn['predict'] = tf.reshape(fcn['predict'], shape=[batch_size, 224, 224, 1])
#print "building"
fcn['upscore2'] = deconv2d(fcn['score_fr'], [batch_size, 14, 14, 6], name='upscore2', stride_h=2, stride_w=2, k_h=4, k_w=4)
fcn['score_pool4'] = conv2d(fcn['maxpool_4'], 6, name='score_pool4', k_w=1, k_h=1)
fcn['fuse_pool4'] = fcn['upscore2'] + fcn['score_pool4']
#print "building2"
fcn['upscore_pool4'] = deconv2d(fcn['fuse_pool4'], [batch_size, 28, 28, 6], name='upscore_pool4', stride_h=2, stride_w=2, k_h=4, k_w=4)
fcn['score_pool3'] = conv2d(fcn['maxpool_3'], 6, name='score_pool3', k_w=1, k_h=1)
fcn['fuse_pool3'] = fcn['score_pool3'] + fcn['upscore_pool4']
#print "building22"
fcn['upscore'] = deconv2d(fcn['fuse_pool3'], [batch_size, 224, 224, 6], name='upscore8', stride_h=8, stride_w=8, k_h=16, k_w=16)
fcn['predict'] = tf.argmax(fcn['upscore'], axis=3, name='predict8')
fcn['predict'] = tf.reshape(fcn['predict'], shape=[batch_size, 224, 224, 1])
#print "buidling22222"
gts = tf.reshape(ground_truth, shape=[batch_size, input_height, input_width])
gts = tf.cast(gts, dtype=tf.int32)
fcn['loss'] = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=gts, logits=fcn['upscore']))
fcn['optimizer'] = tf.train.AdamOptimizer(self.learning_rate, beta1=0.005).minimize(fcn['loss'])
fcn['var_list'] = tf.trainable_variables()
fcn['summary_op'] = tf.summary.scalar('loss', self.model['loss'])
return fcn
#print vgg19('vgg19model.mat')
#download_model(download_url, "vgg19model.mat", 534904783)
COLOR_2_LABEL = np.array([[255, 255, 255] , [0, 0, 255], [0, 255, 255], [0, 255, 0], [255, 255, 0], [255, 0, 0]])
def get_label_id(rgb):
for class_id in xrange(0, 6):
if np.array_equal(rgb, COLOR_2_LABEL[class_id]):
return class_id
def get_rgb(labels):
return COLOR_2_LABEL[labels]
def labels_2_rgb(label_image):
"Label image size [width, height, 1]"
r = np.zeros_like(label_image)
g = np.zeros_like(label_image)
b = np.zeros_like(label_image)
for i in xrange(0, 6):
r[label_image == i] = COLOR_2_LABEL[i , 0]
g[label_image == i] = COLOR_2_LABEL[i , 1]
b[label_image == i] = COLOR_2_LABEL[i , 2]
#rgb = np.zeros((r.shape[0], r.shape[1], 3)).astype(np.uint8)
return np.concatenate((r, g, b), axis=2)
def rgb_2_labels(rgb_image):
rgb_image = np.asarray(rgb_image)
shape = rgb_image.shape
labels_tensor = np.zeros((shape[0], shape[1], 1))
for r in xrange(shape[0]):
for c in xrange(shape[1]):
label_id = get_label_id(rgb_image[r][c])
labels_tensor[r][c][0] = label_id
labels_tensor.astype(np.float32)
return labels_tensor
def get_score(ground_truth, prediction):
#only Overall Accurracy
ground_truth = np.asarray(ground_truth)
prediction = np.asarray(prediction)
temp = ground_truth.shape
sz = 1
for i in temp:
sz = sz * i
#print sz, np.sum((ground_truth == prediction))
return np.sum((ground_truth == prediction)) * 1.0 / sz
def image_save(imgs , path):
imsave(path, imgs)
def softmax(arr, axis=-1):
"""
Return softmax probabilties for given arr and axis
"""
arr = arr - np.expand_dims(np.max(arr, axis), axis)
arr = np.exp(arr)
arr_sum = np.expand_dims(np.sum(arr, axis), axis)
return arr/arr_sum
if __name__ == '__main__' :
"""
a = imread('test_sample.tif')
labels = rgb_2_labels(a)
rev_a = labels_2_rgb(labels)
imsave('rev_test_sample.tif', rev_a)
"""
a = np.array([[1 , 2], [2 , 3]])
b = np.array([[1 , 3], [3 , 4]])
print get_score(a , b)