-
Notifications
You must be signed in to change notification settings - Fork 0
/
build_resnet_graph_potsdam.py
232 lines (187 loc) · 12 KB
/
build_resnet_graph_potsdam.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
231
232
from os import environ
from sys import argv
import tensor_utils_5_channels as utils
import numpy as np
import tensorflow as tf
from scipy.io import loadmat
from scipy.misc import imread
from skimage.transform import resize
def _input():
input_tensor = tf.placeholder(dtype=tf.float32, shape=[None, 224, 224, 3], name='input')
keep_prob = tf.placeholder(tf.float32, name="keep_prob")
is_training = tf.placeholder(tf.bool, name="is_training")
return input_tensor, keep_prob, is_training
def create_branch_layer_names(res_num, res_type, branch_type, branch_order=''):
if branch_type == 1 or branch_order == 'c':
name_list = [None] * 2
name_list[0] = 'res' + str(res_num) + res_type + '_branch' + str(branch_type) + branch_order
name_list[1] = 'bn' + str(res_num) + res_type + '_branch' + str(branch_type) + branch_order
else:
name_list = [None] * 3
name_list[0] = 'res' + str(res_num) + res_type + '_branch' + str(branch_type) + branch_order
name_list[1] = 'bn' + str(res_num) + res_type + '_branch' + str(branch_type) + branch_order
name_list[2] = 'res' + str(res_num) + res_type + '_branch' + str(branch_type) + branch_order + '_relu'
return name_list
def create_param_names_from_layers(layer_names):
param_list = [None] * 4
param_list[0] = layer_names[0] + '_filter'
param_list[1] = layer_names[1] + '_mult'
param_list[2] = layer_names[1] + '_bias'
param_list[3] = layer_names[1] + '_moments'
return param_list
def construct_batch_normalisation_block(current, net, weights, start_weight_index, param_names, layer_names, is_training):
scale = weights[start_weight_index + 1][1].reshape(-1)
scale = utils.get_variable(scale, name=param_names[1])
offset = weights[start_weight_index + 2][1].reshape(-1)
offset = utils.get_variable(offset, name=param_names[2])
mean = weights[start_weight_index + 3][1][:, 0].reshape(-1)
mean = utils.get_variable(mean, name=param_names[3] + '_mean')
variance = weights[start_weight_index + 3][1][:, 1].reshape(-1)
variance = utils.get_variable(variance * variance, name=param_names[3] + '_variance')
batch_mean, batch_var = tf.nn.moments(current, [0, 1, 2], name='batch_moments')
decay = 1 - weights[start_weight_index + 3][2][0][0]
ema = tf.train.ExponentialMovingAverage(decay=decay)
# decay = 0.9999
def mean_var_with_update():
ema_apply_op = ema.apply([batch_mean, batch_var])
with tf.control_dependencies([ema_apply_op]):
return tf.identity(batch_mean), tf.identity(batch_var)
mean, variance = tf.cond(is_training, mean_var_with_update, lambda: (ema.average(batch_mean), ema.average(batch_var)))
current = tf.nn.batch_normalization(current, mean, variance, offset, scale, 1e-5, name=layer_names[1])
net[layer_names[1]] = current
return current, net
def construct_conv_bn_block(current, net, weights, start_weight_index, param_names, layer_names, stride, keep_prob, is_training):
# conv
kernel = weights[start_weight_index][1]
if param_names[0] == 'conv1_filter':
np.random.seed(3796)
appended_kernel = np.random.normal(loc=0, scale=0.02, size=(7, 7, 3, 64)) # for 6 channels of Potsdam
kernel = np.concatenate((kernel, appended_kernel), axis=2)
else:
kernel = utils.get_variable(kernel, name=param_names[0])
if stride == 1:
current = tf.nn.conv2d(current, kernel, strides=[1, 1, 1, 1], padding='SAME', name=layer_names[0])
else:
kernel_size = kernel.shape[0]
padding = int((int(kernel_size) - 1) / 2)
current = tf.nn.conv2d(
tf.pad(current, tf.constant([[0, 0], [padding, padding], [padding, padding], [0, 0]])),
kernel, strides=[1, stride, stride, 1], padding='VALID', name=layer_names[0])
# net[layer_names[0]] = tf.nn.dropout(current, keep_prob=keep_prob)
# bn
# print(layer_names)
# print(param_names)
# print()
current, net = construct_batch_normalisation_block(current, net, weights, start_weight_index, param_names, layer_names, is_training)
return current, net
def construct_conv_bn_relu_block(current, net, weights, start_weight_index, param_names, layer_names, stride, keep_prob, is_training):
# conv_bn
current, net = construct_conv_bn_block(current, net, weights, start_weight_index, param_names, layer_names, stride, keep_prob, is_training)
# relu
current = tf.nn.relu(current, name=layer_names[2])
net[layer_names[2]] = current
return current, net
def construct_branch1_block(res_num, input_tensor, net, weights, start_param_index, first_conv_stride, keep_prob, is_training):
branch1_layer_names = create_branch_layer_names(res_num, res_type='a', branch_type=1)
branch1_param_names = create_param_names_from_layers(branch1_layer_names)
current, net = construct_conv_bn_block(input_tensor, net, weights, start_param_index, branch1_param_names, branch1_layer_names, first_conv_stride, keep_prob, is_training)
return current, net
def construct_branch2_block(res_num, res_type, input_tensor, net, weights, start_param_index, first_conv_stride, keep_prob, is_training):
branch2a_layer_names = create_branch_layer_names(res_num, res_type, branch_type=2, branch_order='a')
branch2a_param_names = create_param_names_from_layers(branch2a_layer_names)
current, net = construct_conv_bn_relu_block(input_tensor, net, weights, start_param_index, branch2a_param_names, branch2a_layer_names, first_conv_stride, keep_prob, is_training)
start_param_index += 4
branch2b_layer_names = create_branch_layer_names(res_num, res_type, branch_type=2, branch_order='b')
branch2b_param_names = create_param_names_from_layers(branch2b_layer_names)
current, net = construct_conv_bn_relu_block(current, net, weights, start_param_index, branch2b_param_names, branch2b_layer_names, 1, keep_prob, is_training)
start_param_index += 4
branch2c_layer_names = create_branch_layer_names(res_num, res_type, branch_type=2, branch_order='c')
branch2c_param_names = create_param_names_from_layers(branch2c_layer_names)
current, net = construct_conv_bn_block(current, net, weights, start_param_index, branch2c_param_names, branch2c_layer_names, 1, keep_prob, is_training)
start_param_index += 4
return current, net
def construct_res_xa_block(res_num, input_tensor, net, weights, start_param_index, keep_prob, is_training, down_sample=True):
# resxa_branch1 block
if down_sample == True:
first_conv_stride = 2
else:
first_conv_stride = 1
current, net = construct_branch1_block(res_num, input_tensor, net, weights, start_param_index, first_conv_stride, keep_prob, is_training)
bn_branch1 = current
start_param_index += 4
# resxa_branch2 block
current, net = construct_branch2_block(res_num, 'a', input_tensor, net, weights, start_param_index, first_conv_stride, keep_prob, is_training)
start_param_index += 12
current = tf.add(bn_branch1, current, name='res' + str(res_num) + 'a')
net['res' + str(res_num) + 'a'] = current
current = tf.nn.relu(current, name='res' + str(res_num) + 'a_relu')
net['res' + str(res_num) + 'a_relu'] = current
return current, net, start_param_index
def construct_res_xxx_block(res_num, res_type, input_tensor, net, weights, start_param_index, keep_prob, is_training):
current, net = construct_branch2_block(res_num, res_type, input_tensor, net, weights, start_param_index, 1, keep_prob, is_training)
start_param_index += 12
current = tf.add(input_tensor, current, name='res' + str(res_num) + res_type)
net['res' + str(res_num) + res_type] = current
current = tf.nn.relu(current, name='res' + str(res_num) + res_type + '_relu')
if 'res' + str(res_num) + res_type + '_relu' == 'res5c_relu':
net['res' + str(res_num) + res_type + '_relu'] = tf.nn.dropout(current, keep_prob=keep_prob)
else:
net['res' + str(res_num) + res_type + '_relu'] = current
return current, net, start_param_index
def resnet101_net(image, weights, keep_prob, is_training):
net = {}
current = image
start_param_index = 0
# conv1 block
conv1_layer_names = ['conv1', 'bn_conv1', 'conv1_relu', 'pool1']
conv1_param_names = create_param_names_from_layers(conv1_layer_names)
current, net = construct_conv_bn_relu_block(current, net, weights, start_param_index, conv1_param_names, conv1_layer_names, 2, keep_prob, is_training)
current = tf.nn.max_pool(current, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME', name=conv1_layer_names[3])
net[conv1_layer_names[3]] = current
start_param_index += 4
current, net, start_param_index = construct_res_xa_block(2, current, net, weights, start_param_index, keep_prob, is_training, down_sample=False)
current, net, start_param_index = construct_res_xxx_block(2, 'b', current, net, weights, start_param_index, keep_prob, is_training)
current, net, start_param_index = construct_res_xxx_block(2, 'c', current, net, weights, start_param_index, keep_prob, is_training)
current, net, start_param_index = construct_res_xa_block(3, current, net, weights, start_param_index, keep_prob, is_training, down_sample=True)
for i in range(1, 4):
current, net, start_param_index = construct_res_xxx_block(3, 'b' + str(i), current, net, weights, start_param_index, keep_prob, is_training)
current, net, start_param_index = construct_res_xa_block(4, current, net, weights, start_param_index, keep_prob, is_training, down_sample=True)
for i in range(1, 23):
current, net, start_param_index = construct_res_xxx_block(4, 'b' + str(i), current, net, weights, start_param_index, keep_prob, is_training)
current, net, start_param_index = construct_res_xa_block(5, current, net, weights, start_param_index, keep_prob, is_training, down_sample=True)
current, net, start_param_index = construct_res_xxx_block(5, 'b', current, net, weights, start_param_index, keep_prob, is_training)
current, net, start_param_index = construct_res_xxx_block(5, 'c', current, net, weights, start_param_index, keep_prob, is_training)
current = tf.nn.avg_pool(current, ksize=[1, 7, 7, 1], strides=[1, 1, 1, 1], padding='VALID', name='pool5')
net['pool5'] = current
fc1000_kernel = utils.get_variable(weights[start_param_index][1], name='fc1000_filter')
fc1000_bias = utils.get_variable(weights[start_param_index + 1][1].reshape(-1), name='fc1000_bias')
current = tf.nn.bias_add(tf.nn.conv2d(current, fc1000_kernel, strides=[1, 1, 1, 1], padding="VALID"), fc1000_bias, name='fc1000')
net['fc1000'] = current
current = tf.nn.softmax(current, name='prob')
net['prob'] = current
return net
def inference(x, weights, keep_prob, is_training):
with tf.variable_scope("inference"):
image_net = resnet101_net(x, weights, keep_prob, is_training)
prediction = tf.argmax(image_net['prob'][0][0][0])
return prediction, image_net
def main(argv=None):
environ["CUDA_VISIBLE_DEVICES"] = argv[1]
resnet101_net = utils.get_model_data('../pretrained_models/imagenet-resnet-101-dag.mat')
weights = np.squeeze(resnet101_net['params'])
img = imread(argv[2])
mean = resnet101_net['meta'][0][0][2][0][0][2]
resized_img = resize(img, (224, 224), preserve_range=True, mode='reflect')
normalised_img = utils.process_image(resized_img, mean)
input_tensor, keep_prob, is_training = _input()
predicted_class, image_net = inference(input_tensor, weights, keep_prob, is_training)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
score, category = sess.run([tf.reduce_max(image_net['prob'][0][0][0]), predicted_class],
feed_dict={input_tensor:normalised_img[np.newaxis, :, :, :].astype(np.float32), keep_prob: 1.0, is_training: False})
print('Category:', resnet101_net['meta'][0][0][1][0][0][1][0][category][0])
print('Score:', score)
# shape = sess.run(image_net['res5c_relu'], feed_dict={x:normalised_img[np.newaxis, :, :, :].astype(np.float32)}).shape
# print(shape)
if __name__ == "__main__":
tf.app.run()