googlenet.py
Go to the documentation of this file.
1 '''
2 Based on https://gist.github.com/joelouismarino/a2ede9ab3928f999575423b9887abd14
3 '''
4 
5 from scipy.misc import imread, imresize
6 
7 from keras.layers import Input, Dense, Convolution2D, MaxPooling2D, AveragePooling2D, ZeroPadding2D, Dropout, Flatten, merge, Reshape, Activation
8 from keras.models import Model
9 from keras.regularizers import l2
10 from keras.optimizers import SGD
11 from googlenet_custom_layers import PoolHelper,LRN
12 
13 
14 def googlenet(input_shape=[500,500,3], classes=13, weights_path=None):
15  # creates GoogLeNet a.k.a. Inception v1 (Szegedy, 2015)
16 
17  input = Input(shape=input_shape)
18 
19  conv1_7x7_s2 = Convolution2D(64,7,7,subsample=(2,2),border_mode='same',activation='relu',name='conv1/7x7_s2',W_regularizer=l2(0.0002))(input)
20 
21  conv1_zero_pad = ZeroPadding2D(padding=(1, 1))(conv1_7x7_s2)
22 
23  pool1_helper = PoolHelper()(conv1_zero_pad)
24 
25  pool1_3x3_s2 = MaxPooling2D(pool_size=(3,3),strides=(2,2),border_mode='valid',name='pool1/3x3_s2')(pool1_helper)
26 
27  pool1_norm1 = LRN(name='pool1/norm1')(pool1_3x3_s2)
28 
29  conv2_3x3_reduce = Convolution2D(64,1,1,border_mode='same',activation='relu',name='conv2/3x3_reduce',W_regularizer=l2(0.0002))(pool1_norm1)
30 
31  conv2_3x3 = Convolution2D(192,3,3,border_mode='same',activation='relu',name='conv2/3x3',W_regularizer=l2(0.0002))(conv2_3x3_reduce)
32 
33  conv2_norm2 = LRN(name='conv2/norm2')(conv2_3x3)
34 
35  conv2_zero_pad = ZeroPadding2D(padding=(1, 1))(conv2_norm2)
36 
37  pool2_helper = PoolHelper()(conv2_zero_pad)
38 
39  pool2_3x3_s2 = MaxPooling2D(pool_size=(3,3),strides=(2,2),border_mode='valid',name='pool2/3x3_s2')(pool2_helper)
40 
41 
42  inception_3a_1x1 = Convolution2D(64,1,1,border_mode='same',activation='relu',name='inception_3a/1x1',W_regularizer=l2(0.0002))(pool2_3x3_s2)
43 
44  inception_3a_3x3_reduce = Convolution2D(96,1,1,border_mode='same',activation='relu',name='inception_3a/3x3_reduce',W_regularizer=l2(0.0002))(pool2_3x3_s2)
45 
46  inception_3a_3x3 = Convolution2D(128,3,3,border_mode='same',activation='relu',name='inception_3a/3x3',W_regularizer=l2(0.0002))(inception_3a_3x3_reduce)
47 
48  inception_3a_5x5_reduce = Convolution2D(16,1,1,border_mode='same',activation='relu',name='inception_3a/5x5_reduce',W_regularizer=l2(0.0002))(pool2_3x3_s2)
49 
50  inception_3a_5x5 = Convolution2D(32,5,5,border_mode='same',activation='relu',name='inception_3a/5x5',W_regularizer=l2(0.0002))(inception_3a_5x5_reduce)
51 
52  inception_3a_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_3a/pool')(pool2_3x3_s2)
53 
54  inception_3a_pool_proj = Convolution2D(32,1,1,border_mode='same',activation='relu',name='inception_3a/pool_proj',W_regularizer=l2(0.0002))(inception_3a_pool)
55 
56  inception_3a_output = merge([inception_3a_1x1,inception_3a_3x3,inception_3a_5x5,inception_3a_pool_proj],mode='concat',concat_axis=1,name='inception_3a/output')
57 
58 
59  inception_3b_1x1 = Convolution2D(128,1,1,border_mode='same',activation='relu',name='inception_3b/1x1',W_regularizer=l2(0.0002))(inception_3a_output)
60 
61  inception_3b_3x3_reduce = Convolution2D(128,1,1,border_mode='same',activation='relu',name='inception_3b/3x3_reduce',W_regularizer=l2(0.0002))(inception_3a_output)
62 
63  inception_3b_3x3 = Convolution2D(192,3,3,border_mode='same',activation='relu',name='inception_3b/3x3',W_regularizer=l2(0.0002))(inception_3b_3x3_reduce)
64 
65  inception_3b_5x5_reduce = Convolution2D(32,1,1,border_mode='same',activation='relu',name='inception_3b/5x5_reduce',W_regularizer=l2(0.0002))(inception_3a_output)
66 
67  inception_3b_5x5 = Convolution2D(96,5,5,border_mode='same',activation='relu',name='inception_3b/5x5',W_regularizer=l2(0.0002))(inception_3b_5x5_reduce)
68 
69  inception_3b_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_3b/pool')(inception_3a_output)
70 
71  inception_3b_pool_proj = Convolution2D(64,1,1,border_mode='same',activation='relu',name='inception_3b/pool_proj',W_regularizer=l2(0.0002))(inception_3b_pool)
72 
73  inception_3b_output = merge([inception_3b_1x1,inception_3b_3x3,inception_3b_5x5,inception_3b_pool_proj],mode='concat',concat_axis=1,name='inception_3b/output')
74 
75 
76  inception_3b_output_zero_pad = ZeroPadding2D(padding=(1, 1))(inception_3b_output)
77 
78  pool3_helper = PoolHelper()(inception_3b_output_zero_pad)
79 
80  pool3_3x3_s2 = MaxPooling2D(pool_size=(3,3),strides=(2,2),border_mode='valid',name='pool3/3x3_s2')(pool3_helper)
81 
82 
83  inception_4a_1x1 = Convolution2D(192,1,1,border_mode='same',activation='relu',name='inception_4a/1x1',W_regularizer=l2(0.0002))(pool3_3x3_s2)
84 
85  inception_4a_3x3_reduce = Convolution2D(96,1,1,border_mode='same',activation='relu',name='inception_4a/3x3_reduce',W_regularizer=l2(0.0002))(pool3_3x3_s2)
86 
87  inception_4a_3x3 = Convolution2D(208,3,3,border_mode='same',activation='relu',name='inception_4a/3x3',W_regularizer=l2(0.0002))(inception_4a_3x3_reduce)
88 
89  inception_4a_5x5_reduce = Convolution2D(16,1,1,border_mode='same',activation='relu',name='inception_4a/5x5_reduce',W_regularizer=l2(0.0002))(pool3_3x3_s2)
90 
91  inception_4a_5x5 = Convolution2D(48,5,5,border_mode='same',activation='relu',name='inception_4a/5x5',W_regularizer=l2(0.0002))(inception_4a_5x5_reduce)
92 
93  inception_4a_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_4a/pool')(pool3_3x3_s2)
94 
95  inception_4a_pool_proj = Convolution2D(64,1,1,border_mode='same',activation='relu',name='inception_4a/pool_proj',W_regularizer=l2(0.0002))(inception_4a_pool)
96 
97  inception_4a_output = merge([inception_4a_1x1,inception_4a_3x3,inception_4a_5x5,inception_4a_pool_proj],mode='concat',concat_axis=1,name='inception_4a/output')
98 
99 
100  loss1_ave_pool = AveragePooling2D(pool_size=(5,5),strides=(3,3),name='loss1/ave_pool')(inception_4a_output)
101 
102  loss1_conv = Convolution2D(128,1,1,border_mode='same',activation='relu',name='loss1/conv',W_regularizer=l2(0.0002))(loss1_ave_pool)
103 
104  loss1_flat = Flatten()(loss1_conv)
105 
106  loss1_fc = Dense(1024,activation='relu',name='loss1/fc',W_regularizer=l2(0.0002))(loss1_flat)
107 
108  loss1_drop_fc = Dropout(0.7)(loss1_fc)
109 
110  loss1_classifier = Dense(1000,name='loss1/classifier',W_regularizer=l2(0.0002))(loss1_drop_fc)
111 
112  loss1_classifier_act = Activation('softmax')(loss1_classifier)
113 
114 
115  inception_4b_1x1 = Convolution2D(160,1,1,border_mode='same',activation='relu',name='inception_4b/1x1',W_regularizer=l2(0.0002))(inception_4a_output)
116 
117  inception_4b_3x3_reduce = Convolution2D(112,1,1,border_mode='same',activation='relu',name='inception_4b/3x3_reduce',W_regularizer=l2(0.0002))(inception_4a_output)
118 
119  inception_4b_3x3 = Convolution2D(224,3,3,border_mode='same',activation='relu',name='inception_4b/3x3',W_regularizer=l2(0.0002))(inception_4b_3x3_reduce)
120 
121  inception_4b_5x5_reduce = Convolution2D(24,1,1,border_mode='same',activation='relu',name='inception_4b/5x5_reduce',W_regularizer=l2(0.0002))(inception_4a_output)
122 
123  inception_4b_5x5 = Convolution2D(64,5,5,border_mode='same',activation='relu',name='inception_4b/5x5',W_regularizer=l2(0.0002))(inception_4b_5x5_reduce)
124 
125  inception_4b_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_4b/pool')(inception_4a_output)
126 
127  inception_4b_pool_proj = Convolution2D(64,1,1,border_mode='same',activation='relu',name='inception_4b/pool_proj',W_regularizer=l2(0.0002))(inception_4b_pool)
128 
129  inception_4b_output = merge([inception_4b_1x1,inception_4b_3x3,inception_4b_5x5,inception_4b_pool_proj],mode='concat',concat_axis=1,name='inception_4b_output')
130 
131 
132  inception_4c_1x1 = Convolution2D(128,1,1,border_mode='same',activation='relu',name='inception_4c/1x1',W_regularizer=l2(0.0002))(inception_4b_output)
133 
134  inception_4c_3x3_reduce = Convolution2D(128,1,1,border_mode='same',activation='relu',name='inception_4c/3x3_reduce',W_regularizer=l2(0.0002))(inception_4b_output)
135 
136  inception_4c_3x3 = Convolution2D(256,3,3,border_mode='same',activation='relu',name='inception_4c/3x3',W_regularizer=l2(0.0002))(inception_4c_3x3_reduce)
137 
138  inception_4c_5x5_reduce = Convolution2D(24,1,1,border_mode='same',activation='relu',name='inception_4c/5x5_reduce',W_regularizer=l2(0.0002))(inception_4b_output)
139 
140  inception_4c_5x5 = Convolution2D(64,5,5,border_mode='same',activation='relu',name='inception_4c/5x5',W_regularizer=l2(0.0002))(inception_4c_5x5_reduce)
141 
142  inception_4c_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_4c/pool')(inception_4b_output)
143 
144  inception_4c_pool_proj = Convolution2D(64,1,1,border_mode='same',activation='relu',name='inception_4c/pool_proj',W_regularizer=l2(0.0002))(inception_4c_pool)
145 
146  inception_4c_output = merge([inception_4c_1x1,inception_4c_3x3,inception_4c_5x5,inception_4c_pool_proj],mode='concat',concat_axis=1,name='inception_4c/output')
147 
148 
149  inception_4d_1x1 = Convolution2D(112,1,1,border_mode='same',activation='relu',name='inception_4d/1x1',W_regularizer=l2(0.0002))(inception_4c_output)
150 
151  inception_4d_3x3_reduce = Convolution2D(144,1,1,border_mode='same',activation='relu',name='inception_4d/3x3_reduce',W_regularizer=l2(0.0002))(inception_4c_output)
152 
153  inception_4d_3x3 = Convolution2D(288,3,3,border_mode='same',activation='relu',name='inception_4d/3x3',W_regularizer=l2(0.0002))(inception_4d_3x3_reduce)
154 
155  inception_4d_5x5_reduce = Convolution2D(32,1,1,border_mode='same',activation='relu',name='inception_4d/5x5_reduce',W_regularizer=l2(0.0002))(inception_4c_output)
156 
157  inception_4d_5x5 = Convolution2D(64,5,5,border_mode='same',activation='relu',name='inception_4d/5x5',W_regularizer=l2(0.0002))(inception_4d_5x5_reduce)
158 
159  inception_4d_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_4d/pool')(inception_4c_output)
160 
161  inception_4d_pool_proj = Convolution2D(64,1,1,border_mode='same',activation='relu',name='inception_4d/pool_proj',W_regularizer=l2(0.0002))(inception_4d_pool)
162 
163  inception_4d_output = merge([inception_4d_1x1,inception_4d_3x3,inception_4d_5x5,inception_4d_pool_proj],mode='concat',concat_axis=1,name='inception_4d/output')
164 
165 
166  loss2_ave_pool = AveragePooling2D(pool_size=(5,5),strides=(3,3),name='loss2/ave_pool')(inception_4d_output)
167 
168  loss2_conv = Convolution2D(128,1,1,border_mode='same',activation='relu',name='loss2/conv',W_regularizer=l2(0.0002))(loss2_ave_pool)
169 
170  loss2_flat = Flatten()(loss2_conv)
171 
172  loss2_fc = Dense(1024,activation='relu',name='loss2/fc',W_regularizer=l2(0.0002))(loss2_flat)
173 
174  loss2_drop_fc = Dropout(0.7)(loss2_fc)
175 
176  loss2_classifier = Dense(1000,name='loss2/classifier',W_regularizer=l2(0.0002))(loss2_drop_fc)
177 
178  loss2_classifier_act = Activation('softmax')(loss2_classifier)
179 
180 
181  inception_4e_1x1 = Convolution2D(256,1,1,border_mode='same',activation='relu',name='inception_4e/1x1',W_regularizer=l2(0.0002))(inception_4d_output)
182 
183  inception_4e_3x3_reduce = Convolution2D(160,1,1,border_mode='same',activation='relu',name='inception_4e/3x3_reduce',W_regularizer=l2(0.0002))(inception_4d_output)
184 
185  inception_4e_3x3 = Convolution2D(320,3,3,border_mode='same',activation='relu',name='inception_4e/3x3',W_regularizer=l2(0.0002))(inception_4e_3x3_reduce)
186 
187  inception_4e_5x5_reduce = Convolution2D(32,1,1,border_mode='same',activation='relu',name='inception_4e/5x5_reduce',W_regularizer=l2(0.0002))(inception_4d_output)
188 
189  inception_4e_5x5 = Convolution2D(128,5,5,border_mode='same',activation='relu',name='inception_4e/5x5',W_regularizer=l2(0.0002))(inception_4e_5x5_reduce)
190 
191  inception_4e_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_4e/pool')(inception_4d_output)
192 
193  inception_4e_pool_proj = Convolution2D(128,1,1,border_mode='same',activation='relu',name='inception_4e/pool_proj',W_regularizer=l2(0.0002))(inception_4e_pool)
194 
195  inception_4e_output = merge([inception_4e_1x1,inception_4e_3x3,inception_4e_5x5,inception_4e_pool_proj],mode='concat',concat_axis=1,name='inception_4e/output')
196 
197 
198  inception_4e_output_zero_pad = ZeroPadding2D(padding=(1, 1))(inception_4e_output)
199 
200  pool4_helper = PoolHelper()(inception_4e_output_zero_pad)
201 
202  pool4_3x3_s2 = MaxPooling2D(pool_size=(3,3),strides=(2,2),border_mode='valid',name='pool4/3x3_s2')(pool4_helper)
203 
204 
205  inception_5a_1x1 = Convolution2D(256,1,1,border_mode='same',activation='relu',name='inception_5a/1x1',W_regularizer=l2(0.0002))(pool4_3x3_s2)
206 
207  inception_5a_3x3_reduce = Convolution2D(160,1,1,border_mode='same',activation='relu',name='inception_5a/3x3_reduce',W_regularizer=l2(0.0002))(pool4_3x3_s2)
208 
209  inception_5a_3x3 = Convolution2D(320,3,3,border_mode='same',activation='relu',name='inception_5a/3x3',W_regularizer=l2(0.0002))(inception_5a_3x3_reduce)
210 
211  inception_5a_5x5_reduce = Convolution2D(32,1,1,border_mode='same',activation='relu',name='inception_5a/5x5_reduce',W_regularizer=l2(0.0002))(pool4_3x3_s2)
212 
213  inception_5a_5x5 = Convolution2D(128,5,5,border_mode='same',activation='relu',name='inception_5a/5x5',W_regularizer=l2(0.0002))(inception_5a_5x5_reduce)
214 
215  inception_5a_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_5a/pool')(pool4_3x3_s2)
216 
217  inception_5a_pool_proj = Convolution2D(128,1,1,border_mode='same',activation='relu',name='inception_5a/pool_proj',W_regularizer=l2(0.0002))(inception_5a_pool)
218 
219  inception_5a_output = merge([inception_5a_1x1,inception_5a_3x3,inception_5a_5x5,inception_5a_pool_proj],mode='concat',concat_axis=1,name='inception_5a/output')
220 
221 
222  inception_5b_1x1 = Convolution2D(384,1,1,border_mode='same',activation='relu',name='inception_5b/1x1',W_regularizer=l2(0.0002))(inception_5a_output)
223 
224  inception_5b_3x3_reduce = Convolution2D(192,1,1,border_mode='same',activation='relu',name='inception_5b/3x3_reduce',W_regularizer=l2(0.0002))(inception_5a_output)
225 
226  inception_5b_3x3 = Convolution2D(384,3,3,border_mode='same',activation='relu',name='inception_5b/3x3',W_regularizer=l2(0.0002))(inception_5b_3x3_reduce)
227 
228  inception_5b_5x5_reduce = Convolution2D(48,1,1,border_mode='same',activation='relu',name='inception_5b/5x5_reduce',W_regularizer=l2(0.0002))(inception_5a_output)
229 
230  inception_5b_5x5 = Convolution2D(128,5,5,border_mode='same',activation='relu',name='inception_5b/5x5',W_regularizer=l2(0.0002))(inception_5b_5x5_reduce)
231 
232  inception_5b_pool = MaxPooling2D(pool_size=(3,3),strides=(1,1),border_mode='same',name='inception_5b/pool')(inception_5a_output)
233 
234  inception_5b_pool_proj = Convolution2D(128,1,1,border_mode='same',activation='relu',name='inception_5b/pool_proj',W_regularizer=l2(0.0002))(inception_5b_pool)
235 
236  inception_5b_output = merge([inception_5b_1x1,inception_5b_3x3,inception_5b_5x5,inception_5b_pool_proj],mode='concat',concat_axis=1,name='inception_5b/output')
237 
238 
239  pool5_7x7_s1 = AveragePooling2D(pool_size=(7,7),strides=(1,1),name='pool5/7x7_s2')(inception_5b_output)
240 
241  loss3_flat = Flatten()(pool5_7x7_s1)
242 
243  pool5_drop_7x7_s1 = Dropout(0.4)(loss3_flat)
244 
245  loss3_classifier = Dense(classes,name='loss3/classifier',W_regularizer=l2(0.0002))(pool5_drop_7x7_s1)
246 
247  loss3_classifier_act = Activation('softmax',name='prob')(loss3_classifier)
248 
249 
250  model = Model(input=input, output=[loss1_classifier_act,loss2_classifier_act,loss3_classifier_act])
251 
252  if weights_path:
253  model.load_weights(weights_path)
254 
255  return model
googlenet_custom_layers.PoolHelper
Definition: googlenet_custom_layers.py:36
googlenet_custom_layers.LRN
Definition: googlenet_custom_layers.py:4
googlenet.googlenet
def googlenet(input_shape=[500, classes=13, weights_path=None)
Definition: googlenet.py:14
Generated by   doxygen 1.8.11