Skip to content

Commit f350b68

Browse files
zsdonghaozsdonghao
committed
simple API for VGG16 VGG 19
1 parent 52d8cce commit f350b68

File tree

2 files changed

+226
-280
lines changed

2 files changed

+226
-280
lines changed

‎example/tutorial_vgg16.py‎

Lines changed: 63 additions & 144 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,7 @@
44

55
import tensorflow as tf
66
import tensorlayer as tl
7+
from tensorlayer.layers import *
78
import os
89
import sys
910
import numpy as np
@@ -60,118 +61,46 @@ def conv_layers(net_in):
6061
mean = tf.constant([123.68, 116.779, 103.939], dtype=tf.float32, shape=[1, 1, 1, 3], name='img_mean')
6162
net_in.outputs = net_in.outputs - mean
6263
""" conv1 """
63-
network = tl.layers.Conv2dLayer(net_in,
64-
act = tf.nn.relu,
65-
shape = [3, 3, 3, 64], # 64 features for each 3x3 patch
66-
strides = [1, 1, 1, 1],
67-
padding='SAME',
68-
name ='conv1_1')
69-
network = tl.layers.Conv2dLayer(network,
70-
act = tf.nn.relu,
71-
shape = [3, 3, 64, 64], # 64 features for each 3x3 patch
72-
strides = [1, 1, 1, 1],
73-
padding='SAME',
74-
name ='conv1_2')
75-
network = tl.layers.PoolLayer(network,
76-
ksize=[1, 2, 2, 1],
77-
strides=[1, 2, 2, 1],
78-
padding='SAME',
79-
pool = tf.nn.max_pool,
80-
name ='pool1')
64+
network = Conv2dLayer(net_in, act = tf.nn.relu, shape = [3, 3, 3, 64], # 64 features for each 3x3 patch
65+
strides = [1, 1, 1, 1], padding='SAME', name ='conv1_1')
66+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 64, 64], # 64 features for each 3x3 patch
67+
strides = [1, 1, 1, 1], padding='SAME', name ='conv1_2')
68+
network = PoolLayer(network, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1],
69+
padding='SAME', pool = tf.nn.max_pool, name ='pool1')
8170
""" conv2 """
82-
network = tl.layers.Conv2dLayer(network,
83-
act = tf.nn.relu,
84-
shape = [3, 3, 64, 128], # 128 features for each 3x3 patch
85-
strides = [1, 1, 1, 1],
86-
padding='SAME',
87-
name ='conv2_1')
88-
network = tl.layers.Conv2dLayer(network,
89-
act = tf.nn.relu,
90-
shape = [3, 3, 128, 128], # 128 features for each 3x3 patch
91-
strides = [1, 1, 1, 1],
92-
padding='SAME',
93-
name ='conv2_2')
94-
network = tl.layers.PoolLayer(network,
95-
ksize=[1, 2, 2, 1],
96-
strides=[1, 2, 2, 1],
97-
padding='SAME',
98-
pool = tf.nn.max_pool,
99-
name ='pool2')
71+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 64, 128], # 128 features for each 3x3 patch
72+
strides = [1, 1, 1, 1], padding='SAME', name ='conv2_1')
73+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 128, 128], # 128 features for each 3x3 patch
74+
strides = [1, 1, 1, 1], padding='SAME', name ='conv2_2')
75+
network = PoolLayer(network, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1],
76+
padding='SAME', pool = tf.nn.max_pool, name ='pool2')
10077
""" conv3 """
101-
network = tl.layers.Conv2dLayer(network,
102-
act = tf.nn.relu,
103-
shape = [3, 3, 128, 256], # 256 features for each 3x3 patch
104-
strides = [1, 1, 1, 1],
105-
padding='SAME',
106-
name ='conv3_1')
107-
network = tl.layers.Conv2dLayer(network,
108-
act = tf.nn.relu,
109-
shape = [3, 3, 256, 256], # 256 features for each 3x3 patch
110-
strides = [1, 1, 1, 1],
111-
padding='SAME',
112-
name ='conv3_2')
113-
network = tl.layers.Conv2dLayer(network,
114-
act = tf.nn.relu,
115-
shape = [3, 3, 256, 256], # 256 features for each 3x3 patch
116-
strides = [1, 1, 1, 1],
117-
padding='SAME',
118-
name ='conv3_3')
119-
network = tl.layers.PoolLayer(network,
120-
ksize=[1, 2, 2, 1],
121-
strides=[1, 2, 2, 1],
122-
padding='SAME',
123-
pool = tf.nn.max_pool,
124-
name ='pool3')
78+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 128, 256], # 256 features for each 3x3 patch
79+
strides = [1, 1, 1, 1], padding='SAME', name ='conv3_1')
80+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 256, 256], # 256 features for each 3x3 patch
81+
strides = [1, 1, 1, 1], padding='SAME', name ='conv3_2')
82+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 256, 256], # 256 features for each 3x3 patch
83+
strides = [1, 1, 1, 1], padding='SAME', name ='conv3_3')
84+
network = PoolLayer(network, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1],
85+
padding='SAME', pool = tf.nn.max_pool, name ='pool3')
12586
""" conv4 """
126-
network = tl.layers.Conv2dLayer(network,
127-
act = tf.nn.relu,
128-
shape = [3, 3, 256, 512], # 512 features for each 3x3 patch
129-
strides = [1, 1, 1, 1],
130-
padding='SAME',
131-
name ='conv4_1')
132-
network = tl.layers.Conv2dLayer(network,
133-
act = tf.nn.relu,
134-
shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
135-
strides = [1, 1, 1, 1],
136-
padding='SAME',
137-
name ='conv4_2')
138-
network = tl.layers.Conv2dLayer(network,
139-
act = tf.nn.relu,
140-
shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
141-
strides = [1, 1, 1, 1],
142-
padding='SAME',
143-
name ='conv4_3')
144-
network = tl.layers.PoolLayer(network,
145-
ksize=[1, 2, 2, 1],
146-
strides=[1, 2, 2, 1],
147-
padding='SAME',
148-
pool = tf.nn.max_pool,
149-
name ='pool4')
87+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 256, 512], # 512 features for each 3x3 patch
88+
strides = [1, 1, 1, 1], padding='SAME', name ='conv4_1')
89+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
90+
strides = [1, 1, 1, 1], padding='SAME', name ='conv4_2')
91+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
92+
strides = [1, 1, 1, 1], padding='SAME', name ='conv4_3')
93+
network = PoolLayer(network, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1],
94+
padding='SAME', pool = tf.nn.max_pool, name ='pool4')
15095
""" conv5 """
151-
network = tl.layers.Conv2dLayer(network,
152-
act = tf.nn.relu,
153-
shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
154-
strides = [1, 1, 1, 1],
155-
padding='SAME',
156-
name ='conv5_1')
157-
network = tl.layers.Conv2dLayer(network,
158-
act = tf.nn.relu,
159-
shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
160-
strides = [1, 1, 1, 1],
161-
padding='SAME',
162-
name ='conv5_2')
163-
network = tl.layers.Conv2dLayer(network,
164-
act = tf.nn.relu,
165-
shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
166-
strides = [1, 1, 1, 1],
167-
padding='SAME',
168-
name ='conv5_3')
169-
network = tl.layers.PoolLayer(network,
170-
ksize=[1, 2, 2, 1],
171-
strides=[1, 2, 2, 1],
172-
padding='SAME',
173-
pool = tf.nn.max_pool,
174-
name ='pool5')
96+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
97+
strides = [1, 1, 1, 1], padding='SAME', name ='conv5_1')
98+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
99+
strides = [1, 1, 1, 1], padding='SAME', name ='conv5_2')
100+
network = Conv2dLayer(network, act = tf.nn.relu, shape = [3, 3, 512, 512], # 512 features for each 3x3 patch
101+
strides = [1, 1, 1, 1], padding='SAME', name ='conv5_3')
102+
network = PoolLayer(network, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1],
103+
padding='SAME', pool = tf.nn.max_pool, name ='pool5')
175104
return network
176105

177106

@@ -185,82 +114,72 @@ def conv_layers_simple_api(net_in):
185114
mean = tf.constant([123.68, 116.779, 103.939], dtype=tf.float32, shape=[1, 1, 1, 3], name='img_mean')
186115
net_in.outputs = net_in.outputs - mean
187116
""" conv1 """
188-
network = tl.layers.Conv2d(net_in, n_filter=64, filter_size=(3, 3),
117+
network = Conv2d(net_in, n_filter=64, filter_size=(3, 3),
189118
strides=(1, 1), act=tf.nn.relu,padding='SAME', name='conv1_1')
190-
network = tl.layers.Conv2d(network, n_filter=64, filter_size=(3, 3),
119+
network = Conv2d(network, n_filter=64, filter_size=(3, 3),
191120
strides=(1, 1), act=tf.nn.relu,padding='SAME', name='conv1_2')
192-
network = tl.layers.MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
121+
network = MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
193122
padding='SAME', name='pool1')
194123
""" conv2 """
195-
network = tl.layers.Conv2d(network, n_filter=128, filter_size=(3, 3),
124+
network = Conv2d(network, n_filter=128, filter_size=(3, 3),
196125
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv2_1')
197-
network = tl.layers.Conv2d(network,n_filter=128, filter_size=(3, 3),
126+
network = Conv2d(network,n_filter=128, filter_size=(3, 3),
198127
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv2_2')
199-
network = tl.layers.MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
128+
network = MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
200129
padding='SAME', name='pool2')
201130
""" conv3 """
202-
network = tl.layers.Conv2d(network, n_filter=256, filter_size=(3, 3),
131+
network = Conv2d(network, n_filter=256, filter_size=(3, 3),
203132
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv3_1')
204-
network = tl.layers.Conv2d(network, n_filter=256, filter_size=(3, 3),
133+
network = Conv2d(network, n_filter=256, filter_size=(3, 3),
205134
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv3_2')
206-
network = tl.layers.Conv2d(network, n_filter=256, filter_size=(3, 3),
135+
network = Conv2d(network, n_filter=256, filter_size=(3, 3),
207136
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv3_3')
208-
network = tl.layers.MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
137+
network = MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
209138
padding='SAME', name='pool3')
210139
""" conv4 """
211-
network = tl.layers.Conv2d(network, n_filter=512, filter_size=(3, 3),
140+
network = Conv2d(network, n_filter=512, filter_size=(3, 3),
212141
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv4_1')
213-
network = tl.layers.Conv2d(network, n_filter=512, filter_size=(3, 3),
142+
network = Conv2d(network, n_filter=512, filter_size=(3, 3),
214143
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv4_2')
215-
network = tl.layers.Conv2d(network, n_filter=512, filter_size=(3, 3),
144+
network = Conv2d(network, n_filter=512, filter_size=(3, 3),
216145
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv4_3')
217-
network = tl.layers.MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
146+
network = MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
218147
padding='SAME', name='pool4')
219148
""" conv5 """
220-
network = tl.layers.Conv2d(network, n_filter=512, filter_size=(3, 3),
149+
network = Conv2d(network, n_filter=512, filter_size=(3, 3),
221150
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv5_1')
222-
network = tl.layers.Conv2d(network, n_filter=512, filter_size=(3, 3),
151+
network = Conv2d(network, n_filter=512, filter_size=(3, 3),
223152
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv5_2')
224-
network = tl.layers.Conv2d(network, n_filter=512, filter_size=(3, 3),
153+
network = Conv2d(network, n_filter=512, filter_size=(3, 3),
225154
strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv5_3')
226-
network = tl.layers.MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
155+
network = MaxPool2d(network, filter_size=(2, 2), strides=(2, 2),
227156
padding='SAME', name='pool5')
228157
return network
229158

230-
231159
def fc_layers(net):
232-
network = tl.layers.FlattenLayer(net, name='flatten')
233-
network = tl.layers.DenseLayer(network, n_units=4096,
234-
act = tf.nn.relu,
235-
name = 'fc1_relu')
236-
network = tl.layers.DenseLayer(network, n_units=4096,
237-
act = tf.nn.relu,
238-
name = 'fc2_relu')
239-
network = tl.layers.DenseLayer(network, n_units=1000,
240-
act = tf.identity,
241-
name = 'fc3_relu')
160+
network = FlattenLayer(net, name='flatten')
161+
network = DenseLayer(network, n_units=4096, act=tf.nn.relu, name='fc1_relu')
162+
network = DenseLayer(network, n_units=4096, act=tf.nn.relu, name='fc2_relu')
163+
network = DenseLayer(network, n_units=1000, act=tf.identity, name='fc3_relu')
242164
return network
243165

244-
245166
sess = tf.InteractiveSession()
246167

247168
x = tf.placeholder(tf.float32, [None, 224, 224, 3])
248-
y_ = tf.placeholder(tf.int32, shape=[None, ], name='y_')
169+
# y_ = tf.placeholder(tf.int32, shape=[None, ], name='y_')
249170

250-
net_in = tl.layers.InputLayer(x, name='input')
251-
net_cnn = conv_layers(net_in) # professional CNN APIs
252-
# net_cnn = conv_layers_simple_api(net_in) # simplified CNN APIs
171+
net_in = InputLayer(x, name='input')
172+
# net_cnn = conv_layers(net_in) # professional CNN APIs
173+
net_cnn = conv_layers_simple_api(net_in) # simplified CNN APIs
253174
network = fc_layers(net_cnn)
254175

255176
y = network.outputs
256177
probs = tf.nn.softmax(y)
257178
# y_op = tf.argmax(tf.nn.softmax(y), 1)
258179
# cost = tl.cost.cross_entropy(y, y_, name='cost')
259-
#
260180
# correct_prediction = tf.equal(tf.cast(tf.argmax(y, 1), tf.float32), tf.cast(y_, tf.float32))
261181
# acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
262182

263-
# sess.run(tf.initialize_all_variables())
264183
tl.layers.initialize_global_variables(sess)
265184
network.print_params()
266185
network.print_layers()

0 commit comments

Comments
 (0)