├── AlexNet
    └── deploy.prototxt
├── BN-GoogLeNet
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── DenseNet-121
    ├── solver.prototxt
    └── train_val.prototxt
├── DenseNet-161
    ├── solver.prototxt
    └── train_val.prototxt
├── GoogLeNet
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── Inception-ResNet-v2
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── Inception-v3
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── Inception-v4
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── README.md
├── ResNet-101
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── ResNet-152
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── ResNet-50
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── SE-BN-Inception
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
├── SE-ResNet-50
    ├── solver.prototxt
    └── train_val.prototxt
├── SqueezeNet
    ├── deploy.prototxt
    ├── solver.prototxt
    ├── squeezenet_v1.1.caffemodel
    └── train_val.prototxt
├── VGG16
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt
└── VGG19
    ├── deploy.prototxt
    ├── solver.prototxt
    └── train_val.prototxt


/AlexNet/deploy.prototxt:
--------------------------------------------------------------------------------
  1 | name: "AlexNet"
  2 | layer {
  3 |   name: "data"
  4 |   type: "Input"
  5 |   top: "data"
  6 |   input_param { shape: { dim: 1 dim: 3 dim: 227 dim: 227 } }
  7 | }
  8 | layer {
  9 |   name: "conv1"
 10 |   type: "Convolution"
 11 |   bottom: "data"
 12 |   top: "conv1"
 13 |   param {
 14 |     lr_mult: 1
 15 |     decay_mult: 1
 16 |   }
 17 |   param {
 18 |     lr_mult: 2
 19 |     decay_mult: 0
 20 |   }
 21 |   convolution_param {
 22 |     num_output: 96
 23 |     kernel_size: 11
 24 |     stride: 4
 25 |   }
 26 | }
 27 | layer {
 28 |   name: "relu1"
 29 |   type: "ReLU"
 30 |   bottom: "conv1"
 31 |   top: "conv1"
 32 | }
 33 | layer {
 34 |   name: "norm1"
 35 |   type: "LRN"
 36 |   bottom: "conv1"
 37 |   top: "norm1"
 38 |   lrn_param {
 39 |     local_size: 5
 40 |     alpha: 0.0001
 41 |     beta: 0.75
 42 |   }
 43 | }
 44 | layer {
 45 |   name: "pool1"
 46 |   type: "Pooling"
 47 |   bottom: "norm1"
 48 |   top: "pool1"
 49 |   pooling_param {
 50 |     pool: MAX
 51 |     kernel_size: 3
 52 |     stride: 2
 53 |   }
 54 | }
 55 | layer {
 56 |   name: "conv2"
 57 |   type: "Convolution"
 58 |   bottom: "pool1"
 59 |   top: "conv2"
 60 |   param {
 61 |     lr_mult: 1
 62 |     decay_mult: 1
 63 |   }
 64 |   param {
 65 |     lr_mult: 2
 66 |     decay_mult: 0
 67 |   }
 68 |   convolution_param {
 69 |     num_output: 256
 70 |     pad: 2
 71 |     kernel_size: 5
 72 |     group: 2
 73 |   }
 74 | }
 75 | layer {
 76 |   name: "relu2"
 77 |   type: "ReLU"
 78 |   bottom: "conv2"
 79 |   top: "conv2"
 80 | }
 81 | layer {
 82 |   name: "norm2"
 83 |   type: "LRN"
 84 |   bottom: "conv2"
 85 |   top: "norm2"
 86 |   lrn_param {
 87 |     local_size: 5
 88 |     alpha: 0.0001
 89 |     beta: 0.75
 90 |   }
 91 | }
 92 | layer {
 93 |   name: "pool2"
 94 |   type: "Pooling"
 95 |   bottom: "norm2"
 96 |   top: "pool2"
 97 |   pooling_param {
 98 |     pool: MAX
 99 |     kernel_size: 3
100 |     stride: 2
101 |   }
102 | }
103 | layer {
104 |   name: "conv3"
105 |   type: "Convolution"
106 |   bottom: "pool2"
107 |   top: "conv3"
108 |   param {
109 |     lr_mult: 1
110 |     decay_mult: 1
111 |   }
112 |   param {
113 |     lr_mult: 2
114 |     decay_mult: 0
115 |   }
116 |   convolution_param {
117 |     num_output: 384
118 |     pad: 1
119 |     kernel_size: 3
120 |   }
121 | }
122 | layer {
123 |   name: "relu3"
124 |   type: "ReLU"
125 |   bottom: "conv3"
126 |   top: "conv3"
127 | }
128 | layer {
129 |   name: "conv4"
130 |   type: "Convolution"
131 |   bottom: "conv3"
132 |   top: "conv4"
133 |   param {
134 |     lr_mult: 1
135 |     decay_mult: 1
136 |   }
137 |   param {
138 |     lr_mult: 2
139 |     decay_mult: 0
140 |   }
141 |   convolution_param {
142 |     num_output: 384
143 |     pad: 1
144 |     kernel_size: 3
145 |     group: 2
146 |   }
147 | }
148 | layer {
149 |   name: "relu4"
150 |   type: "ReLU"
151 |   bottom: "conv4"
152 |   top: "conv4"
153 | }
154 | layer {
155 |   name: "conv5"
156 |   type: "Convolution"
157 |   bottom: "conv4"
158 |   top: "conv5"
159 |   param {
160 |     lr_mult: 1
161 |     decay_mult: 1
162 |   }
163 |   param {
164 |     lr_mult: 2
165 |     decay_mult: 0
166 |   }
167 |   convolution_param {
168 |     num_output: 256
169 |     pad: 1
170 |     kernel_size: 3
171 |     group: 2
172 |   }
173 | }
174 | layer {
175 |   name: "relu5"
176 |   type: "ReLU"
177 |   bottom: "conv5"
178 |   top: "conv5"
179 | }
180 | layer {
181 |   name: "pool5"
182 |   type: "Pooling"
183 |   bottom: "conv5"
184 |   top: "pool5"
185 |   pooling_param {
186 |     pool: MAX
187 |     kernel_size: 3
188 |     stride: 2
189 |   }
190 | }
191 | layer {
192 |   name: "fc6"
193 |   type: "InnerProduct"
194 |   bottom: "pool5"
195 |   top: "fc6"
196 |   param {
197 |     lr_mult: 1
198 |     decay_mult: 1
199 |   }
200 |   param {
201 |     lr_mult: 2
202 |     decay_mult: 0
203 |   }
204 |   inner_product_param {
205 |     num_output: 4096
206 |   }
207 | }
208 | layer {
209 |   name: "relu6"
210 |   type: "ReLU"
211 |   bottom: "fc6"
212 |   top: "fc6"
213 | }
214 | layer {
215 |   name: "drop6"
216 |   type: "Dropout"
217 |   bottom: "fc6"
218 |   top: "fc6"
219 |   dropout_param {
220 |     dropout_ratio: 0.5
221 |   }
222 | }
223 | layer {
224 |   name: "fc7"
225 |   type: "InnerProduct"
226 |   bottom: "fc6"
227 |   top: "fc7"
228 |   param {
229 |     lr_mult: 1
230 |     decay_mult: 1
231 |   }
232 |   param {
233 |     lr_mult: 2
234 |     decay_mult: 0
235 |   }
236 |   inner_product_param {
237 |     num_output: 4096
238 |   }
239 | }
240 | layer {
241 |   name: "relu7"
242 |   type: "ReLU"
243 |   bottom: "fc7"
244 |   top: "fc7"
245 | }
246 | layer {
247 |   name: "drop7"
248 |   type: "Dropout"
249 |   bottom: "fc7"
250 |   top: "fc7"
251 |   dropout_param {
252 |     dropout_ratio: 0.5
253 |   }
254 | }
255 | layer {
256 |   name: "fc8"
257 |   type: "InnerProduct"
258 |   bottom: "fc7"
259 |   top: "fc8"
260 |   param {
261 |     lr_mult: 1
262 |     decay_mult: 1
263 |   }
264 |   param {
265 |     lr_mult: 2
266 |     decay_mult: 0
267 |   }
268 |   inner_product_param {
269 |     num_output: 1000
270 |   }
271 | }
272 | layer {
273 |   name: "prob"
274 |   type: "Softmax"
275 |   bottom: "fc8"
276 |   top: "prob"
277 | }
278 | 


--------------------------------------------------------------------------------
/BN-GoogLeNet/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 512
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 10001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 10
26 | # save path
27 | snapshot_prefix: "inception-v2-cervix"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/DenseNet-121/solver.prototxt:
--------------------------------------------------------------------------------
 1 | train_net: "train_val.prototxt"
 2 | #test_net: "test_densenet.prototxt"
 3 | #test_iter: 200
 4 | #test_interval: 800
 5 | base_lr: 0.001
 6 | display: 1
 7 | max_iter: 2001
 8 | lr_policy: "multistep"
 9 | gamma: 0.1
10 | momentum: 0.9
11 | weight_decay: 0.0001
12 | solver_mode: GPU
13 | random_seed: 831486
14 | stepvalue: 115000
15 | stepvalue: 172500
16 | type: "Nesterov"
17 | snapshot_prefix: "densenet-121-product"
18 | snapshot: 50
19 | 


--------------------------------------------------------------------------------
/DenseNet-161/solver.prototxt:
--------------------------------------------------------------------------------
 1 | train_net: "train_val.prototxt"
 2 | #test_net: "test_densenet.prototxt"
 3 | #test_iter: 200
 4 | #test_interval: 800
 5 | base_lr: 0.001
 6 | display: 1
 7 | max_iter: 2001
 8 | lr_policy: "multistep"
 9 | gamma: 0.1
10 | momentum: 0.9
11 | weight_decay: 0.0001
12 | solver_mode: GPU
13 | random_seed: 831486
14 | stepvalue: 115000
15 | stepvalue: 172500
16 | type: "Nesterov"
17 | snapshot_prefix: "densenet-161-product"
18 | snapshot: 50
19 | 


--------------------------------------------------------------------------------
/GoogLeNet/deploy.prototxt:
--------------------------------------------------------------------------------
   1 | name: "GoogleNet"
   2 | input: "data"
   3 | input_shape {
   4 |   dim: 1
   5 |   dim: 3
   6 |   dim: 224
   7 |   dim: 224
   8 | }
   9 | layer {
  10 |   name: "conv1/7x7_s2"
  11 |   type: "Convolution"
  12 |   bottom: "data"
  13 |   top: "conv1/7x7_s2"
  14 |   param {
  15 |     lr_mult: 1
  16 |     decay_mult: 1
  17 |   }
  18 |   param {
  19 |     lr_mult: 2
  20 |     decay_mult: 0
  21 |   }
  22 |   convolution_param {
  23 |     num_output: 64
  24 |     pad: 3
  25 |     kernel_size: 7
  26 |     stride: 2
  27 |     weight_filler {
  28 |       type: "xavier"
  29 |     }
  30 |     bias_filler {
  31 |       type: "constant"
  32 |       value: 0.2
  33 |     }
  34 |   }
  35 | }
  36 | layer {
  37 |   name: "conv1/relu_7x7"
  38 |   type: "ReLU"
  39 |   bottom: "conv1/7x7_s2"
  40 |   top: "conv1/7x7_s2"
  41 | }
  42 | layer {
  43 |   name: "pool1/3x3_s2"
  44 |   type: "Pooling"
  45 |   bottom: "conv1/7x7_s2"
  46 |   top: "pool1/3x3_s2"
  47 |   pooling_param {
  48 |     pool: MAX
  49 |     kernel_size: 3
  50 |     stride: 2
  51 |   }
  52 | }
  53 | layer {
  54 |   name: "pool1/norm1"
  55 |   type: "LRN"
  56 |   bottom: "pool1/3x3_s2"
  57 |   top: "pool1/norm1"
  58 |   lrn_param {
  59 |     local_size: 5
  60 |     alpha: 0.0001
  61 |     beta: 0.75
  62 |   }
  63 | }
  64 | layer {
  65 |   name: "conv2/3x3_reduce"
  66 |   type: "Convolution"
  67 |   bottom: "pool1/norm1"
  68 |   top: "conv2/3x3_reduce"
  69 |   param {
  70 |     lr_mult: 1
  71 |     decay_mult: 1
  72 |   }
  73 |   param {
  74 |     lr_mult: 2
  75 |     decay_mult: 0
  76 |   }
  77 |   convolution_param {
  78 |     num_output: 64
  79 |     kernel_size: 1
  80 |     weight_filler {
  81 |       type: "xavier"
  82 |     }
  83 |     bias_filler {
  84 |       type: "constant"
  85 |       value: 0.2
  86 |     }
  87 |   }
  88 | }
  89 | layer {
  90 |   name: "conv2/relu_3x3_reduce"
  91 |   type: "ReLU"
  92 |   bottom: "conv2/3x3_reduce"
  93 |   top: "conv2/3x3_reduce"
  94 | }
  95 | layer {
  96 |   name: "conv2/3x3"
  97 |   type: "Convolution"
  98 |   bottom: "conv2/3x3_reduce"
  99 |   top: "conv2/3x3"
 100 |   param {
 101 |     lr_mult: 1
 102 |     decay_mult: 1
 103 |   }
 104 |   param {
 105 |     lr_mult: 2
 106 |     decay_mult: 0
 107 |   }
 108 |   convolution_param {
 109 |     num_output: 192
 110 |     pad: 1
 111 |     kernel_size: 3
 112 |     weight_filler {
 113 |       type: "xavier"
 114 |     }
 115 |     bias_filler {
 116 |       type: "constant"
 117 |       value: 0.2
 118 |     }
 119 |   }
 120 | }
 121 | layer {
 122 |   name: "conv2/relu_3x3"
 123 |   type: "ReLU"
 124 |   bottom: "conv2/3x3"
 125 |   top: "conv2/3x3"
 126 | }
 127 | layer {
 128 |   name: "conv2/norm2"
 129 |   type: "LRN"
 130 |   bottom: "conv2/3x3"
 131 |   top: "conv2/norm2"
 132 |   lrn_param {
 133 |     local_size: 5
 134 |     alpha: 0.0001
 135 |     beta: 0.75
 136 |   }
 137 | }
 138 | layer {
 139 |   name: "pool2/3x3_s2"
 140 |   type: "Pooling"
 141 |   bottom: "conv2/norm2"
 142 |   top: "pool2/3x3_s2"
 143 |   pooling_param {
 144 |     pool: MAX
 145 |     kernel_size: 3
 146 |     stride: 2
 147 |   }
 148 | }
 149 | layer {
 150 |   name: "inception_3a/1x1"
 151 |   type: "Convolution"
 152 |   bottom: "pool2/3x3_s2"
 153 |   top: "inception_3a/1x1"
 154 |   param {
 155 |     lr_mult: 1
 156 |     decay_mult: 1
 157 |   }
 158 |   param {
 159 |     lr_mult: 2
 160 |     decay_mult: 0
 161 |   }
 162 |   convolution_param {
 163 |     num_output: 64
 164 |     kernel_size: 1
 165 |     weight_filler {
 166 |       type: "xavier"
 167 |     }
 168 |     bias_filler {
 169 |       type: "constant"
 170 |       value: 0.2
 171 |     }
 172 |   }
 173 | }
 174 | layer {
 175 |   name: "inception_3a/relu_1x1"
 176 |   type: "ReLU"
 177 |   bottom: "inception_3a/1x1"
 178 |   top: "inception_3a/1x1"
 179 | }
 180 | layer {
 181 |   name: "inception_3a/3x3_reduce"
 182 |   type: "Convolution"
 183 |   bottom: "pool2/3x3_s2"
 184 |   top: "inception_3a/3x3_reduce"
 185 |   param {
 186 |     lr_mult: 1
 187 |     decay_mult: 1
 188 |   }
 189 |   param {
 190 |     lr_mult: 2
 191 |     decay_mult: 0
 192 |   }
 193 |   convolution_param {
 194 |     num_output: 96
 195 |     kernel_size: 1
 196 |     weight_filler {
 197 |       type: "xavier"
 198 |     }
 199 |     bias_filler {
 200 |       type: "constant"
 201 |       value: 0.2
 202 |     }
 203 |   }
 204 | }
 205 | layer {
 206 |   name: "inception_3a/relu_3x3_reduce"
 207 |   type: "ReLU"
 208 |   bottom: "inception_3a/3x3_reduce"
 209 |   top: "inception_3a/3x3_reduce"
 210 | }
 211 | layer {
 212 |   name: "inception_3a/3x3"
 213 |   type: "Convolution"
 214 |   bottom: "inception_3a/3x3_reduce"
 215 |   top: "inception_3a/3x3"
 216 |   param {
 217 |     lr_mult: 1
 218 |     decay_mult: 1
 219 |   }
 220 |   param {
 221 |     lr_mult: 2
 222 |     decay_mult: 0
 223 |   }
 224 |   convolution_param {
 225 |     num_output: 128
 226 |     pad: 1
 227 |     kernel_size: 3
 228 |     weight_filler {
 229 |       type: "xavier"
 230 |     }
 231 |     bias_filler {
 232 |       type: "constant"
 233 |       value: 0.2
 234 |     }
 235 |   }
 236 | }
 237 | layer {
 238 |   name: "inception_3a/relu_3x3"
 239 |   type: "ReLU"
 240 |   bottom: "inception_3a/3x3"
 241 |   top: "inception_3a/3x3"
 242 | }
 243 | layer {
 244 |   name: "inception_3a/5x5_reduce"
 245 |   type: "Convolution"
 246 |   bottom: "pool2/3x3_s2"
 247 |   top: "inception_3a/5x5_reduce"
 248 |   param {
 249 |     lr_mult: 1
 250 |     decay_mult: 1
 251 |   }
 252 |   param {
 253 |     lr_mult: 2
 254 |     decay_mult: 0
 255 |   }
 256 |   convolution_param {
 257 |     num_output: 16
 258 |     kernel_size: 1
 259 |     weight_filler {
 260 |       type: "xavier"
 261 |     }
 262 |     bias_filler {
 263 |       type: "constant"
 264 |       value: 0.2
 265 |     }
 266 |   }
 267 | }
 268 | layer {
 269 |   name: "inception_3a/relu_5x5_reduce"
 270 |   type: "ReLU"
 271 |   bottom: "inception_3a/5x5_reduce"
 272 |   top: "inception_3a/5x5_reduce"
 273 | }
 274 | layer {
 275 |   name: "inception_3a/5x5"
 276 |   type: "Convolution"
 277 |   bottom: "inception_3a/5x5_reduce"
 278 |   top: "inception_3a/5x5"
 279 |   param {
 280 |     lr_mult: 1
 281 |     decay_mult: 1
 282 |   }
 283 |   param {
 284 |     lr_mult: 2
 285 |     decay_mult: 0
 286 |   }
 287 |   convolution_param {
 288 |     num_output: 32
 289 |     pad: 2
 290 |     kernel_size: 5
 291 |     weight_filler {
 292 |       type: "xavier"
 293 |     }
 294 |     bias_filler {
 295 |       type: "constant"
 296 |       value: 0.2
 297 |     }
 298 |   }
 299 | }
 300 | layer {
 301 |   name: "inception_3a/relu_5x5"
 302 |   type: "ReLU"
 303 |   bottom: "inception_3a/5x5"
 304 |   top: "inception_3a/5x5"
 305 | }
 306 | layer {
 307 |   name: "inception_3a/pool"
 308 |   type: "Pooling"
 309 |   bottom: "pool2/3x3_s2"
 310 |   top: "inception_3a/pool"
 311 |   pooling_param {
 312 |     pool: MAX
 313 |     kernel_size: 3
 314 |     stride: 1
 315 |     pad: 1
 316 |   }
 317 | }
 318 | layer {
 319 |   name: "inception_3a/pool_proj"
 320 |   type: "Convolution"
 321 |   bottom: "inception_3a/pool"
 322 |   top: "inception_3a/pool_proj"
 323 |   param {
 324 |     lr_mult: 1
 325 |     decay_mult: 1
 326 |   }
 327 |   param {
 328 |     lr_mult: 2
 329 |     decay_mult: 0
 330 |   }
 331 |   convolution_param {
 332 |     num_output: 32
 333 |     kernel_size: 1
 334 |     weight_filler {
 335 |       type: "xavier"
 336 |     }
 337 |     bias_filler {
 338 |       type: "constant"
 339 |       value: 0.2
 340 |     }
 341 |   }
 342 | }
 343 | layer {
 344 |   name: "inception_3a/relu_pool_proj"
 345 |   type: "ReLU"
 346 |   bottom: "inception_3a/pool_proj"
 347 |   top: "inception_3a/pool_proj"
 348 | }
 349 | layer {
 350 |   name: "inception_3a/output"
 351 |   type: "Concat"
 352 |   bottom: "inception_3a/1x1"
 353 |   bottom: "inception_3a/3x3"
 354 |   bottom: "inception_3a/5x5"
 355 |   bottom: "inception_3a/pool_proj"
 356 |   top: "inception_3a/output"
 357 | }
 358 | layer {
 359 |   name: "inception_3b/1x1"
 360 |   type: "Convolution"
 361 |   bottom: "inception_3a/output"
 362 |   top: "inception_3b/1x1"
 363 |   param {
 364 |     lr_mult: 1
 365 |     decay_mult: 1
 366 |   }
 367 |   param {
 368 |     lr_mult: 2
 369 |     decay_mult: 0
 370 |   }
 371 |   convolution_param {
 372 |     num_output: 128
 373 |     kernel_size: 1
 374 |     weight_filler {
 375 |       type: "xavier"
 376 |     }
 377 |     bias_filler {
 378 |       type: "constant"
 379 |       value: 0.2
 380 |     }
 381 |   }
 382 | }
 383 | layer {
 384 |   name: "inception_3b/relu_1x1"
 385 |   type: "ReLU"
 386 |   bottom: "inception_3b/1x1"
 387 |   top: "inception_3b/1x1"
 388 | }
 389 | layer {
 390 |   name: "inception_3b/3x3_reduce"
 391 |   type: "Convolution"
 392 |   bottom: "inception_3a/output"
 393 |   top: "inception_3b/3x3_reduce"
 394 |   param {
 395 |     lr_mult: 1
 396 |     decay_mult: 1
 397 |   }
 398 |   param {
 399 |     lr_mult: 2
 400 |     decay_mult: 0
 401 |   }
 402 |   convolution_param {
 403 |     num_output: 128
 404 |     kernel_size: 1
 405 |     weight_filler {
 406 |       type: "xavier"
 407 |     }
 408 |     bias_filler {
 409 |       type: "constant"
 410 |       value: 0.2
 411 |     }
 412 |   }
 413 | }
 414 | layer {
 415 |   name: "inception_3b/relu_3x3_reduce"
 416 |   type: "ReLU"
 417 |   bottom: "inception_3b/3x3_reduce"
 418 |   top: "inception_3b/3x3_reduce"
 419 | }
 420 | layer {
 421 |   name: "inception_3b/3x3"
 422 |   type: "Convolution"
 423 |   bottom: "inception_3b/3x3_reduce"
 424 |   top: "inception_3b/3x3"
 425 |   param {
 426 |     lr_mult: 1
 427 |     decay_mult: 1
 428 |   }
 429 |   param {
 430 |     lr_mult: 2
 431 |     decay_mult: 0
 432 |   }
 433 |   convolution_param {
 434 |     num_output: 192
 435 |     pad: 1
 436 |     kernel_size: 3
 437 |     weight_filler {
 438 |       type: "xavier"
 439 |     }
 440 |     bias_filler {
 441 |       type: "constant"
 442 |       value: 0.2
 443 |     }
 444 |   }
 445 | }
 446 | layer {
 447 |   name: "inception_3b/relu_3x3"
 448 |   type: "ReLU"
 449 |   bottom: "inception_3b/3x3"
 450 |   top: "inception_3b/3x3"
 451 | }
 452 | layer {
 453 |   name: "inception_3b/5x5_reduce"
 454 |   type: "Convolution"
 455 |   bottom: "inception_3a/output"
 456 |   top: "inception_3b/5x5_reduce"
 457 |   param {
 458 |     lr_mult: 1
 459 |     decay_mult: 1
 460 |   }
 461 |   param {
 462 |     lr_mult: 2
 463 |     decay_mult: 0
 464 |   }
 465 |   convolution_param {
 466 |     num_output: 32
 467 |     kernel_size: 1
 468 |     weight_filler {
 469 |       type: "xavier"
 470 |     }
 471 |     bias_filler {
 472 |       type: "constant"
 473 |       value: 0.2
 474 |     }
 475 |   }
 476 | }
 477 | layer {
 478 |   name: "inception_3b/relu_5x5_reduce"
 479 |   type: "ReLU"
 480 |   bottom: "inception_3b/5x5_reduce"
 481 |   top: "inception_3b/5x5_reduce"
 482 | }
 483 | layer {
 484 |   name: "inception_3b/5x5"
 485 |   type: "Convolution"
 486 |   bottom: "inception_3b/5x5_reduce"
 487 |   top: "inception_3b/5x5"
 488 |   param {
 489 |     lr_mult: 1
 490 |     decay_mult: 1
 491 |   }
 492 |   param {
 493 |     lr_mult: 2
 494 |     decay_mult: 0
 495 |   }
 496 |   convolution_param {
 497 |     num_output: 96
 498 |     pad: 2
 499 |     kernel_size: 5
 500 |     weight_filler {
 501 |       type: "xavier"
 502 |     }
 503 |     bias_filler {
 504 |       type: "constant"
 505 |       value: 0.2
 506 |     }
 507 |   }
 508 | }
 509 | layer {
 510 |   name: "inception_3b/relu_5x5"
 511 |   type: "ReLU"
 512 |   bottom: "inception_3b/5x5"
 513 |   top: "inception_3b/5x5"
 514 | }
 515 | layer {
 516 |   name: "inception_3b/pool"
 517 |   type: "Pooling"
 518 |   bottom: "inception_3a/output"
 519 |   top: "inception_3b/pool"
 520 |   pooling_param {
 521 |     pool: MAX
 522 |     kernel_size: 3
 523 |     stride: 1
 524 |     pad: 1
 525 |   }
 526 | }
 527 | layer {
 528 |   name: "inception_3b/pool_proj"
 529 |   type: "Convolution"
 530 |   bottom: "inception_3b/pool"
 531 |   top: "inception_3b/pool_proj"
 532 |   param {
 533 |     lr_mult: 1
 534 |     decay_mult: 1
 535 |   }
 536 |   param {
 537 |     lr_mult: 2
 538 |     decay_mult: 0
 539 |   }
 540 |   convolution_param {
 541 |     num_output: 64
 542 |     kernel_size: 1
 543 |     weight_filler {
 544 |       type: "xavier"
 545 |     }
 546 |     bias_filler {
 547 |       type: "constant"
 548 |       value: 0.2
 549 |     }
 550 |   }
 551 | }
 552 | layer {
 553 |   name: "inception_3b/relu_pool_proj"
 554 |   type: "ReLU"
 555 |   bottom: "inception_3b/pool_proj"
 556 |   top: "inception_3b/pool_proj"
 557 | }
 558 | layer {
 559 |   name: "inception_3b/output"
 560 |   type: "Concat"
 561 |   bottom: "inception_3b/1x1"
 562 |   bottom: "inception_3b/3x3"
 563 |   bottom: "inception_3b/5x5"
 564 |   bottom: "inception_3b/pool_proj"
 565 |   top: "inception_3b/output"
 566 | }
 567 | layer {
 568 |   name: "pool3/3x3_s2"
 569 |   type: "Pooling"
 570 |   bottom: "inception_3b/output"
 571 |   top: "pool3/3x3_s2"
 572 |   pooling_param {
 573 |     pool: MAX
 574 |     kernel_size: 3
 575 |     stride: 2
 576 |   }
 577 | }
 578 | layer {
 579 |   name: "inception_4a/1x1"
 580 |   type: "Convolution"
 581 |   bottom: "pool3/3x3_s2"
 582 |   top: "inception_4a/1x1"
 583 |   param {
 584 |     lr_mult: 1
 585 |     decay_mult: 1
 586 |   }
 587 |   param {
 588 |     lr_mult: 2
 589 |     decay_mult: 0
 590 |   }
 591 |   convolution_param {
 592 |     num_output: 192
 593 |     kernel_size: 1
 594 |     weight_filler {
 595 |       type: "xavier"
 596 |     }
 597 |     bias_filler {
 598 |       type: "constant"
 599 |       value: 0.2
 600 |     }
 601 |   }
 602 | }
 603 | layer {
 604 |   name: "inception_4a/relu_1x1"
 605 |   type: "ReLU"
 606 |   bottom: "inception_4a/1x1"
 607 |   top: "inception_4a/1x1"
 608 | }
 609 | layer {
 610 |   name: "inception_4a/3x3_reduce"
 611 |   type: "Convolution"
 612 |   bottom: "pool3/3x3_s2"
 613 |   top: "inception_4a/3x3_reduce"
 614 |   param {
 615 |     lr_mult: 1
 616 |     decay_mult: 1
 617 |   }
 618 |   param {
 619 |     lr_mult: 2
 620 |     decay_mult: 0
 621 |   }
 622 |   convolution_param {
 623 |     num_output: 96
 624 |     kernel_size: 1
 625 |     weight_filler {
 626 |       type: "xavier"
 627 |     }
 628 |     bias_filler {
 629 |       type: "constant"
 630 |       value: 0.2
 631 |     }
 632 |   }
 633 | }
 634 | layer {
 635 |   name: "inception_4a/relu_3x3_reduce"
 636 |   type: "ReLU"
 637 |   bottom: "inception_4a/3x3_reduce"
 638 |   top: "inception_4a/3x3_reduce"
 639 | }
 640 | layer {
 641 |   name: "inception_4a/3x3"
 642 |   type: "Convolution"
 643 |   bottom: "inception_4a/3x3_reduce"
 644 |   top: "inception_4a/3x3"
 645 |   param {
 646 |     lr_mult: 1
 647 |     decay_mult: 1
 648 |   }
 649 |   param {
 650 |     lr_mult: 2
 651 |     decay_mult: 0
 652 |   }
 653 |   convolution_param {
 654 |     num_output: 208
 655 |     pad: 1
 656 |     kernel_size: 3
 657 |     weight_filler {
 658 |       type: "xavier"
 659 |     }
 660 |     bias_filler {
 661 |       type: "constant"
 662 |       value: 0.2
 663 |     }
 664 |   }
 665 | }
 666 | layer {
 667 |   name: "inception_4a/relu_3x3"
 668 |   type: "ReLU"
 669 |   bottom: "inception_4a/3x3"
 670 |   top: "inception_4a/3x3"
 671 | }
 672 | layer {
 673 |   name: "inception_4a/5x5_reduce"
 674 |   type: "Convolution"
 675 |   bottom: "pool3/3x3_s2"
 676 |   top: "inception_4a/5x5_reduce"
 677 |   param {
 678 |     lr_mult: 1
 679 |     decay_mult: 1
 680 |   }
 681 |   param {
 682 |     lr_mult: 2
 683 |     decay_mult: 0
 684 |   }
 685 |   convolution_param {
 686 |     num_output: 16
 687 |     kernel_size: 1
 688 |     weight_filler {
 689 |       type: "xavier"
 690 |     }
 691 |     bias_filler {
 692 |       type: "constant"
 693 |       value: 0.2
 694 |     }
 695 |   }
 696 | }
 697 | layer {
 698 |   name: "inception_4a/relu_5x5_reduce"
 699 |   type: "ReLU"
 700 |   bottom: "inception_4a/5x5_reduce"
 701 |   top: "inception_4a/5x5_reduce"
 702 | }
 703 | layer {
 704 |   name: "inception_4a/5x5"
 705 |   type: "Convolution"
 706 |   bottom: "inception_4a/5x5_reduce"
 707 |   top: "inception_4a/5x5"
 708 |   param {
 709 |     lr_mult: 1
 710 |     decay_mult: 1
 711 |   }
 712 |   param {
 713 |     lr_mult: 2
 714 |     decay_mult: 0
 715 |   }
 716 |   convolution_param {
 717 |     num_output: 48
 718 |     pad: 2
 719 |     kernel_size: 5
 720 |     weight_filler {
 721 |       type: "xavier"
 722 |     }
 723 |     bias_filler {
 724 |       type: "constant"
 725 |       value: 0.2
 726 |     }
 727 |   }
 728 | }
 729 | layer {
 730 |   name: "inception_4a/relu_5x5"
 731 |   type: "ReLU"
 732 |   bottom: "inception_4a/5x5"
 733 |   top: "inception_4a/5x5"
 734 | }
 735 | layer {
 736 |   name: "inception_4a/pool"
 737 |   type: "Pooling"
 738 |   bottom: "pool3/3x3_s2"
 739 |   top: "inception_4a/pool"
 740 |   pooling_param {
 741 |     pool: MAX
 742 |     kernel_size: 3
 743 |     stride: 1
 744 |     pad: 1
 745 |   }
 746 | }
 747 | layer {
 748 |   name: "inception_4a/pool_proj"
 749 |   type: "Convolution"
 750 |   bottom: "inception_4a/pool"
 751 |   top: "inception_4a/pool_proj"
 752 |   param {
 753 |     lr_mult: 1
 754 |     decay_mult: 1
 755 |   }
 756 |   param {
 757 |     lr_mult: 2
 758 |     decay_mult: 0
 759 |   }
 760 |   convolution_param {
 761 |     num_output: 64
 762 |     kernel_size: 1
 763 |     weight_filler {
 764 |       type: "xavier"
 765 |     }
 766 |     bias_filler {
 767 |       type: "constant"
 768 |       value: 0.2
 769 |     }
 770 |   }
 771 | }
 772 | layer {
 773 |   name: "inception_4a/relu_pool_proj"
 774 |   type: "ReLU"
 775 |   bottom: "inception_4a/pool_proj"
 776 |   top: "inception_4a/pool_proj"
 777 | }
 778 | layer {
 779 |   name: "inception_4a/output"
 780 |   type: "Concat"
 781 |   bottom: "inception_4a/1x1"
 782 |   bottom: "inception_4a/3x3"
 783 |   bottom: "inception_4a/5x5"
 784 |   bottom: "inception_4a/pool_proj"
 785 |   top: "inception_4a/output"
 786 | }
 787 | layer {
 788 |   name: "loss1/ave_pool"
 789 |   type: "Pooling"
 790 |   bottom: "inception_4a/output"
 791 |   top: "loss1/ave_pool"
 792 |   pooling_param {
 793 |     pool: AVE
 794 |     kernel_size: 5
 795 |     stride: 3
 796 |   }
 797 | }
 798 | layer {
 799 |   name: "loss1/conv"
 800 |   type: "Convolution"
 801 |   bottom: "loss1/ave_pool"
 802 |   top: "loss1/conv"
 803 |   param {
 804 |     lr_mult: 1
 805 |     decay_mult: 1
 806 |   }
 807 |   param {
 808 |     lr_mult: 2
 809 |     decay_mult: 0
 810 |   }
 811 |   convolution_param {
 812 |     num_output: 128
 813 |     kernel_size: 1
 814 |     weight_filler {
 815 |       type: "xavier"
 816 |     }
 817 |     bias_filler {
 818 |       type: "constant"
 819 |       value: 0.2
 820 |     }
 821 |   }
 822 | }
 823 | layer {
 824 |   name: "loss1/relu_conv"
 825 |   type: "ReLU"
 826 |   bottom: "loss1/conv"
 827 |   top: "loss1/conv"
 828 | }
 829 | layer {
 830 |   name: "loss1/fc"
 831 |   type: "InnerProduct"
 832 |   bottom: "loss1/conv"
 833 |   top: "loss1/fc"
 834 |   param {
 835 |     lr_mult: 1
 836 |     decay_mult: 1
 837 |   }
 838 |   param {
 839 |     lr_mult: 2
 840 |     decay_mult: 0
 841 |   }
 842 |   inner_product_param {
 843 |     num_output: 1024
 844 |     weight_filler {
 845 |       type: "xavier"
 846 |     }
 847 |     bias_filler {
 848 |       type: "constant"
 849 |       value: 0.2
 850 |     }
 851 |   }
 852 | }
 853 | layer {
 854 |   name: "loss1/relu_fc"
 855 |   type: "ReLU"
 856 |   bottom: "loss1/fc"
 857 |   top: "loss1/fc"
 858 | }
 859 | layer {
 860 |   name: "loss1/drop_fc"
 861 |   type: "Dropout"
 862 |   bottom: "loss1/fc"
 863 |   top: "loss1/fc"
 864 |   dropout_param {
 865 |     dropout_ratio: 0.7
 866 |   }
 867 | }
 868 | layer {
 869 |   name: "inception_4b/1x1"
 870 |   type: "Convolution"
 871 |   bottom: "inception_4a/output"
 872 |   top: "inception_4b/1x1"
 873 |   param {
 874 |     lr_mult: 1
 875 |     decay_mult: 1
 876 |   }
 877 |   param {
 878 |     lr_mult: 2
 879 |     decay_mult: 0
 880 |   }
 881 |   convolution_param {
 882 |     num_output: 160
 883 |     kernel_size: 1
 884 |     weight_filler {
 885 |       type: "xavier"
 886 |     }
 887 |     bias_filler {
 888 |       type: "constant"
 889 |       value: 0.2
 890 |     }
 891 |   }
 892 | }
 893 | layer {
 894 |   name: "inception_4b/relu_1x1"
 895 |   type: "ReLU"
 896 |   bottom: "inception_4b/1x1"
 897 |   top: "inception_4b/1x1"
 898 | }
 899 | layer {
 900 |   name: "inception_4b/3x3_reduce"
 901 |   type: "Convolution"
 902 |   bottom: "inception_4a/output"
 903 |   top: "inception_4b/3x3_reduce"
 904 |   param {
 905 |     lr_mult: 1
 906 |     decay_mult: 1
 907 |   }
 908 |   param {
 909 |     lr_mult: 2
 910 |     decay_mult: 0
 911 |   }
 912 |   convolution_param {
 913 |     num_output: 112
 914 |     kernel_size: 1
 915 |     weight_filler {
 916 |       type: "xavier"
 917 |     }
 918 |     bias_filler {
 919 |       type: "constant"
 920 |       value: 0.2
 921 |     }
 922 |   }
 923 | }
 924 | layer {
 925 |   name: "inception_4b/relu_3x3_reduce"
 926 |   type: "ReLU"
 927 |   bottom: "inception_4b/3x3_reduce"
 928 |   top: "inception_4b/3x3_reduce"
 929 | }
 930 | layer {
 931 |   name: "inception_4b/3x3"
 932 |   type: "Convolution"
 933 |   bottom: "inception_4b/3x3_reduce"
 934 |   top: "inception_4b/3x3"
 935 |   param {
 936 |     lr_mult: 1
 937 |     decay_mult: 1
 938 |   }
 939 |   param {
 940 |     lr_mult: 2
 941 |     decay_mult: 0
 942 |   }
 943 |   convolution_param {
 944 |     num_output: 224
 945 |     pad: 1
 946 |     kernel_size: 3
 947 |     weight_filler {
 948 |       type: "xavier"
 949 |     }
 950 |     bias_filler {
 951 |       type: "constant"
 952 |       value: 0.2
 953 |     }
 954 |   }
 955 | }
 956 | layer {
 957 |   name: "inception_4b/relu_3x3"
 958 |   type: "ReLU"
 959 |   bottom: "inception_4b/3x3"
 960 |   top: "inception_4b/3x3"
 961 | }
 962 | layer {
 963 |   name: "inception_4b/5x5_reduce"
 964 |   type: "Convolution"
 965 |   bottom: "inception_4a/output"
 966 |   top: "inception_4b/5x5_reduce"
 967 |   param {
 968 |     lr_mult: 1
 969 |     decay_mult: 1
 970 |   }
 971 |   param {
 972 |     lr_mult: 2
 973 |     decay_mult: 0
 974 |   }
 975 |   convolution_param {
 976 |     num_output: 24
 977 |     kernel_size: 1
 978 |     weight_filler {
 979 |       type: "xavier"
 980 |     }
 981 |     bias_filler {
 982 |       type: "constant"
 983 |       value: 0.2
 984 |     }
 985 |   }
 986 | }
 987 | layer {
 988 |   name: "inception_4b/relu_5x5_reduce"
 989 |   type: "ReLU"
 990 |   bottom: "inception_4b/5x5_reduce"
 991 |   top: "inception_4b/5x5_reduce"
 992 | }
 993 | layer {
 994 |   name: "inception_4b/5x5"
 995 |   type: "Convolution"
 996 |   bottom: "inception_4b/5x5_reduce"
 997 |   top: "inception_4b/5x5"
 998 |   param {
 999 |     lr_mult: 1
1000 |     decay_mult: 1
1001 |   }
1002 |   param {
1003 |     lr_mult: 2
1004 |     decay_mult: 0
1005 |   }
1006 |   convolution_param {
1007 |     num_output: 64
1008 |     pad: 2
1009 |     kernel_size: 5
1010 |     weight_filler {
1011 |       type: "xavier"
1012 |     }
1013 |     bias_filler {
1014 |       type: "constant"
1015 |       value: 0.2
1016 |     }
1017 |   }
1018 | }
1019 | layer {
1020 |   name: "inception_4b/relu_5x5"
1021 |   type: "ReLU"
1022 |   bottom: "inception_4b/5x5"
1023 |   top: "inception_4b/5x5"
1024 | }
1025 | layer {
1026 |   name: "inception_4b/pool"
1027 |   type: "Pooling"
1028 |   bottom: "inception_4a/output"
1029 |   top: "inception_4b/pool"
1030 |   pooling_param {
1031 |     pool: MAX
1032 |     kernel_size: 3
1033 |     stride: 1
1034 |     pad: 1
1035 |   }
1036 | }
1037 | layer {
1038 |   name: "inception_4b/pool_proj"
1039 |   type: "Convolution"
1040 |   bottom: "inception_4b/pool"
1041 |   top: "inception_4b/pool_proj"
1042 |   param {
1043 |     lr_mult: 1
1044 |     decay_mult: 1
1045 |   }
1046 |   param {
1047 |     lr_mult: 2
1048 |     decay_mult: 0
1049 |   }
1050 |   convolution_param {
1051 |     num_output: 64
1052 |     kernel_size: 1
1053 |     weight_filler {
1054 |       type: "xavier"
1055 |     }
1056 |     bias_filler {
1057 |       type: "constant"
1058 |       value: 0.2
1059 |     }
1060 |   }
1061 | }
1062 | layer {
1063 |   name: "inception_4b/relu_pool_proj"
1064 |   type: "ReLU"
1065 |   bottom: "inception_4b/pool_proj"
1066 |   top: "inception_4b/pool_proj"
1067 | }
1068 | layer {
1069 |   name: "inception_4b/output"
1070 |   type: "Concat"
1071 |   bottom: "inception_4b/1x1"
1072 |   bottom: "inception_4b/3x3"
1073 |   bottom: "inception_4b/5x5"
1074 |   bottom: "inception_4b/pool_proj"
1075 |   top: "inception_4b/output"
1076 | }
1077 | layer {
1078 |   name: "inception_4c/1x1"
1079 |   type: "Convolution"
1080 |   bottom: "inception_4b/output"
1081 |   top: "inception_4c/1x1"
1082 |   param {
1083 |     lr_mult: 1
1084 |     decay_mult: 1
1085 |   }
1086 |   param {
1087 |     lr_mult: 2
1088 |     decay_mult: 0
1089 |   }
1090 |   convolution_param {
1091 |     num_output: 128
1092 |     kernel_size: 1
1093 |     weight_filler {
1094 |       type: "xavier"
1095 |     }
1096 |     bias_filler {
1097 |       type: "constant"
1098 |       value: 0.2
1099 |     }
1100 |   }
1101 | }
1102 | layer {
1103 |   name: "inception_4c/relu_1x1"
1104 |   type: "ReLU"
1105 |   bottom: "inception_4c/1x1"
1106 |   top: "inception_4c/1x1"
1107 | }
1108 | layer {
1109 |   name: "inception_4c/3x3_reduce"
1110 |   type: "Convolution"
1111 |   bottom: "inception_4b/output"
1112 |   top: "inception_4c/3x3_reduce"
1113 |   param {
1114 |     lr_mult: 1
1115 |     decay_mult: 1
1116 |   }
1117 |   param {
1118 |     lr_mult: 2
1119 |     decay_mult: 0
1120 |   }
1121 |   convolution_param {
1122 |     num_output: 128
1123 |     kernel_size: 1
1124 |     weight_filler {
1125 |       type: "xavier"
1126 |     }
1127 |     bias_filler {
1128 |       type: "constant"
1129 |       value: 0.2
1130 |     }
1131 |   }
1132 | }
1133 | layer {
1134 |   name: "inception_4c/relu_3x3_reduce"
1135 |   type: "ReLU"
1136 |   bottom: "inception_4c/3x3_reduce"
1137 |   top: "inception_4c/3x3_reduce"
1138 | }
1139 | layer {
1140 |   name: "inception_4c/3x3"
1141 |   type: "Convolution"
1142 |   bottom: "inception_4c/3x3_reduce"
1143 |   top: "inception_4c/3x3"
1144 |   param {
1145 |     lr_mult: 1
1146 |     decay_mult: 1
1147 |   }
1148 |   param {
1149 |     lr_mult: 2
1150 |     decay_mult: 0
1151 |   }
1152 |   convolution_param {
1153 |     num_output: 256
1154 |     pad: 1
1155 |     kernel_size: 3
1156 |     weight_filler {
1157 |       type: "xavier"
1158 |     }
1159 |     bias_filler {
1160 |       type: "constant"
1161 |       value: 0.2
1162 |     }
1163 |   }
1164 | }
1165 | layer {
1166 |   name: "inception_4c/relu_3x3"
1167 |   type: "ReLU"
1168 |   bottom: "inception_4c/3x3"
1169 |   top: "inception_4c/3x3"
1170 | }
1171 | layer {
1172 |   name: "inception_4c/5x5_reduce"
1173 |   type: "Convolution"
1174 |   bottom: "inception_4b/output"
1175 |   top: "inception_4c/5x5_reduce"
1176 |   param {
1177 |     lr_mult: 1
1178 |     decay_mult: 1
1179 |   }
1180 |   param {
1181 |     lr_mult: 2
1182 |     decay_mult: 0
1183 |   }
1184 |   convolution_param {
1185 |     num_output: 24
1186 |     kernel_size: 1
1187 |     weight_filler {
1188 |       type: "xavier"
1189 |     }
1190 |     bias_filler {
1191 |       type: "constant"
1192 |       value: 0.2
1193 |     }
1194 |   }
1195 | }
1196 | layer {
1197 |   name: "inception_4c/relu_5x5_reduce"
1198 |   type: "ReLU"
1199 |   bottom: "inception_4c/5x5_reduce"
1200 |   top: "inception_4c/5x5_reduce"
1201 | }
1202 | layer {
1203 |   name: "inception_4c/5x5"
1204 |   type: "Convolution"
1205 |   bottom: "inception_4c/5x5_reduce"
1206 |   top: "inception_4c/5x5"
1207 |   param {
1208 |     lr_mult: 1
1209 |     decay_mult: 1
1210 |   }
1211 |   param {
1212 |     lr_mult: 2
1213 |     decay_mult: 0
1214 |   }
1215 |   convolution_param {
1216 |     num_output: 64
1217 |     pad: 2
1218 |     kernel_size: 5
1219 |     weight_filler {
1220 |       type: "xavier"
1221 |     }
1222 |     bias_filler {
1223 |       type: "constant"
1224 |       value: 0.2
1225 |     }
1226 |   }
1227 | }
1228 | layer {
1229 |   name: "inception_4c/relu_5x5"
1230 |   type: "ReLU"
1231 |   bottom: "inception_4c/5x5"
1232 |   top: "inception_4c/5x5"
1233 | }
1234 | layer {
1235 |   name: "inception_4c/pool"
1236 |   type: "Pooling"
1237 |   bottom: "inception_4b/output"
1238 |   top: "inception_4c/pool"
1239 |   pooling_param {
1240 |     pool: MAX
1241 |     kernel_size: 3
1242 |     stride: 1
1243 |     pad: 1
1244 |   }
1245 | }
1246 | layer {
1247 |   name: "inception_4c/pool_proj"
1248 |   type: "Convolution"
1249 |   bottom: "inception_4c/pool"
1250 |   top: "inception_4c/pool_proj"
1251 |   param {
1252 |     lr_mult: 1
1253 |     decay_mult: 1
1254 |   }
1255 |   param {
1256 |     lr_mult: 2
1257 |     decay_mult: 0
1258 |   }
1259 |   convolution_param {
1260 |     num_output: 64
1261 |     kernel_size: 1
1262 |     weight_filler {
1263 |       type: "xavier"
1264 |     }
1265 |     bias_filler {
1266 |       type: "constant"
1267 |       value: 0.2
1268 |     }
1269 |   }
1270 | }
1271 | layer {
1272 |   name: "inception_4c/relu_pool_proj"
1273 |   type: "ReLU"
1274 |   bottom: "inception_4c/pool_proj"
1275 |   top: "inception_4c/pool_proj"
1276 | }
1277 | layer {
1278 |   name: "inception_4c/output"
1279 |   type: "Concat"
1280 |   bottom: "inception_4c/1x1"
1281 |   bottom: "inception_4c/3x3"
1282 |   bottom: "inception_4c/5x5"
1283 |   bottom: "inception_4c/pool_proj"
1284 |   top: "inception_4c/output"
1285 | }
1286 | layer {
1287 |   name: "inception_4d/1x1"
1288 |   type: "Convolution"
1289 |   bottom: "inception_4c/output"
1290 |   top: "inception_4d/1x1"
1291 |   param {
1292 |     lr_mult: 1
1293 |     decay_mult: 1
1294 |   }
1295 |   param {
1296 |     lr_mult: 2
1297 |     decay_mult: 0
1298 |   }
1299 |   convolution_param {
1300 |     num_output: 112
1301 |     kernel_size: 1
1302 |     weight_filler {
1303 |       type: "xavier"
1304 |     }
1305 |     bias_filler {
1306 |       type: "constant"
1307 |       value: 0.2
1308 |     }
1309 |   }
1310 | }
1311 | layer {
1312 |   name: "inception_4d/relu_1x1"
1313 |   type: "ReLU"
1314 |   bottom: "inception_4d/1x1"
1315 |   top: "inception_4d/1x1"
1316 | }
1317 | layer {
1318 |   name: "inception_4d/3x3_reduce"
1319 |   type: "Convolution"
1320 |   bottom: "inception_4c/output"
1321 |   top: "inception_4d/3x3_reduce"
1322 |   param {
1323 |     lr_mult: 1
1324 |     decay_mult: 1
1325 |   }
1326 |   param {
1327 |     lr_mult: 2
1328 |     decay_mult: 0
1329 |   }
1330 |   convolution_param {
1331 |     num_output: 144
1332 |     kernel_size: 1
1333 |     weight_filler {
1334 |       type: "xavier"
1335 |     }
1336 |     bias_filler {
1337 |       type: "constant"
1338 |       value: 0.2
1339 |     }
1340 |   }
1341 | }
1342 | layer {
1343 |   name: "inception_4d/relu_3x3_reduce"
1344 |   type: "ReLU"
1345 |   bottom: "inception_4d/3x3_reduce"
1346 |   top: "inception_4d/3x3_reduce"
1347 | }
1348 | layer {
1349 |   name: "inception_4d/3x3"
1350 |   type: "Convolution"
1351 |   bottom: "inception_4d/3x3_reduce"
1352 |   top: "inception_4d/3x3"
1353 |   param {
1354 |     lr_mult: 1
1355 |     decay_mult: 1
1356 |   }
1357 |   param {
1358 |     lr_mult: 2
1359 |     decay_mult: 0
1360 |   }
1361 |   convolution_param {
1362 |     num_output: 288
1363 |     pad: 1
1364 |     kernel_size: 3
1365 |     weight_filler {
1366 |       type: "xavier"
1367 |     }
1368 |     bias_filler {
1369 |       type: "constant"
1370 |       value: 0.2
1371 |     }
1372 |   }
1373 | }
1374 | layer {
1375 |   name: "inception_4d/relu_3x3"
1376 |   type: "ReLU"
1377 |   bottom: "inception_4d/3x3"
1378 |   top: "inception_4d/3x3"
1379 | }
1380 | layer {
1381 |   name: "inception_4d/5x5_reduce"
1382 |   type: "Convolution"
1383 |   bottom: "inception_4c/output"
1384 |   top: "inception_4d/5x5_reduce"
1385 |   param {
1386 |     lr_mult: 1
1387 |     decay_mult: 1
1388 |   }
1389 |   param {
1390 |     lr_mult: 2
1391 |     decay_mult: 0
1392 |   }
1393 |   convolution_param {
1394 |     num_output: 32
1395 |     kernel_size: 1
1396 |     weight_filler {
1397 |       type: "xavier"
1398 |     }
1399 |     bias_filler {
1400 |       type: "constant"
1401 |       value: 0.2
1402 |     }
1403 |   }
1404 | }
1405 | layer {
1406 |   name: "inception_4d/relu_5x5_reduce"
1407 |   type: "ReLU"
1408 |   bottom: "inception_4d/5x5_reduce"
1409 |   top: "inception_4d/5x5_reduce"
1410 | }
1411 | layer {
1412 |   name: "inception_4d/5x5"
1413 |   type: "Convolution"
1414 |   bottom: "inception_4d/5x5_reduce"
1415 |   top: "inception_4d/5x5"
1416 |   param {
1417 |     lr_mult: 1
1418 |     decay_mult: 1
1419 |   }
1420 |   param {
1421 |     lr_mult: 2
1422 |     decay_mult: 0
1423 |   }
1424 |   convolution_param {
1425 |     num_output: 64
1426 |     pad: 2
1427 |     kernel_size: 5
1428 |     weight_filler {
1429 |       type: "xavier"
1430 |     }
1431 |     bias_filler {
1432 |       type: "constant"
1433 |       value: 0.2
1434 |     }
1435 |   }
1436 | }
1437 | layer {
1438 |   name: "inception_4d/relu_5x5"
1439 |   type: "ReLU"
1440 |   bottom: "inception_4d/5x5"
1441 |   top: "inception_4d/5x5"
1442 | }
1443 | layer {
1444 |   name: "inception_4d/pool"
1445 |   type: "Pooling"
1446 |   bottom: "inception_4c/output"
1447 |   top: "inception_4d/pool"
1448 |   pooling_param {
1449 |     pool: MAX
1450 |     kernel_size: 3
1451 |     stride: 1
1452 |     pad: 1
1453 |   }
1454 | }
1455 | layer {
1456 |   name: "inception_4d/pool_proj"
1457 |   type: "Convolution"
1458 |   bottom: "inception_4d/pool"
1459 |   top: "inception_4d/pool_proj"
1460 |   param {
1461 |     lr_mult: 1
1462 |     decay_mult: 1
1463 |   }
1464 |   param {
1465 |     lr_mult: 2
1466 |     decay_mult: 0
1467 |   }
1468 |   convolution_param {
1469 |     num_output: 64
1470 |     kernel_size: 1
1471 |     weight_filler {
1472 |       type: "xavier"
1473 |     }
1474 |     bias_filler {
1475 |       type: "constant"
1476 |       value: 0.2
1477 |     }
1478 |   }
1479 | }
1480 | layer {
1481 |   name: "inception_4d/relu_pool_proj"
1482 |   type: "ReLU"
1483 |   bottom: "inception_4d/pool_proj"
1484 |   top: "inception_4d/pool_proj"
1485 | }
1486 | layer {
1487 |   name: "inception_4d/output"
1488 |   type: "Concat"
1489 |   bottom: "inception_4d/1x1"
1490 |   bottom: "inception_4d/3x3"
1491 |   bottom: "inception_4d/5x5"
1492 |   bottom: "inception_4d/pool_proj"
1493 |   top: "inception_4d/output"
1494 | }
1495 | layer {
1496 |   name: "loss2/ave_pool"
1497 |   type: "Pooling"
1498 |   bottom: "inception_4d/output"
1499 |   top: "loss2/ave_pool"
1500 |   pooling_param {
1501 |     pool: AVE
1502 |     kernel_size: 5
1503 |     stride: 3
1504 |   }
1505 | }
1506 | layer {
1507 |   name: "loss2/conv"
1508 |   type: "Convolution"
1509 |   bottom: "loss2/ave_pool"
1510 |   top: "loss2/conv"
1511 |   param {
1512 |     lr_mult: 1
1513 |     decay_mult: 1
1514 |   }
1515 |   param {
1516 |     lr_mult: 2
1517 |     decay_mult: 0
1518 |   }
1519 |   convolution_param {
1520 |     num_output: 128
1521 |     kernel_size: 1
1522 |     weight_filler {
1523 |       type: "xavier"
1524 |     }
1525 |     bias_filler {
1526 |       type: "constant"
1527 |       value: 0.2
1528 |     }
1529 |   }
1530 | }
1531 | layer {
1532 |   name: "loss2/relu_conv"
1533 |   type: "ReLU"
1534 |   bottom: "loss2/conv"
1535 |   top: "loss2/conv"
1536 | }
1537 | layer {
1538 |   name: "loss2/fc"
1539 |   type: "InnerProduct"
1540 |   bottom: "loss2/conv"
1541 |   top: "loss2/fc"
1542 |   param {
1543 |     lr_mult: 1
1544 |     decay_mult: 1
1545 |   }
1546 |   param {
1547 |     lr_mult: 2
1548 |     decay_mult: 0
1549 |   }
1550 |   inner_product_param {
1551 |     num_output: 1024
1552 |     weight_filler {
1553 |       type: "xavier"
1554 |     }
1555 |     bias_filler {
1556 |       type: "constant"
1557 |       value: 0.2
1558 |     }
1559 |   }
1560 | }
1561 | layer {
1562 |   name: "loss2/relu_fc"
1563 |   type: "ReLU"
1564 |   bottom: "loss2/fc"
1565 |   top: "loss2/fc"
1566 | }
1567 | layer {
1568 |   name: "loss2/drop_fc"
1569 |   type: "Dropout"
1570 |   bottom: "loss2/fc"
1571 |   top: "loss2/fc"
1572 |   dropout_param {
1573 |     dropout_ratio: 0.7
1574 |   }
1575 | }
1576 | layer {
1577 |   name: "inception_4e/1x1"
1578 |   type: "Convolution"
1579 |   bottom: "inception_4d/output"
1580 |   top: "inception_4e/1x1"
1581 |   param {
1582 |     lr_mult: 1
1583 |     decay_mult: 1
1584 |   }
1585 |   param {
1586 |     lr_mult: 2
1587 |     decay_mult: 0
1588 |   }
1589 |   convolution_param {
1590 |     num_output: 256
1591 |     kernel_size: 1
1592 |     weight_filler {
1593 |       type: "xavier"
1594 |     }
1595 |     bias_filler {
1596 |       type: "constant"
1597 |       value: 0.2
1598 |     }
1599 |   }
1600 | }
1601 | layer {
1602 |   name: "inception_4e/relu_1x1"
1603 |   type: "ReLU"
1604 |   bottom: "inception_4e/1x1"
1605 |   top: "inception_4e/1x1"
1606 | }
1607 | layer {
1608 |   name: "inception_4e/3x3_reduce"
1609 |   type: "Convolution"
1610 |   bottom: "inception_4d/output"
1611 |   top: "inception_4e/3x3_reduce"
1612 |   param {
1613 |     lr_mult: 1
1614 |     decay_mult: 1
1615 |   }
1616 |   param {
1617 |     lr_mult: 2
1618 |     decay_mult: 0
1619 |   }
1620 |   convolution_param {
1621 |     num_output: 160
1622 |     kernel_size: 1
1623 |     weight_filler {
1624 |       type: "xavier"
1625 |     }
1626 |     bias_filler {
1627 |       type: "constant"
1628 |       value: 0.2
1629 |     }
1630 |   }
1631 | }
1632 | layer {
1633 |   name: "inception_4e/relu_3x3_reduce"
1634 |   type: "ReLU"
1635 |   bottom: "inception_4e/3x3_reduce"
1636 |   top: "inception_4e/3x3_reduce"
1637 | }
1638 | layer {
1639 |   name: "inception_4e/3x3"
1640 |   type: "Convolution"
1641 |   bottom: "inception_4e/3x3_reduce"
1642 |   top: "inception_4e/3x3"
1643 |   param {
1644 |     lr_mult: 1
1645 |     decay_mult: 1
1646 |   }
1647 |   param {
1648 |     lr_mult: 2
1649 |     decay_mult: 0
1650 |   }
1651 |   convolution_param {
1652 |     num_output: 320
1653 |     pad: 1
1654 |     kernel_size: 3
1655 |     weight_filler {
1656 |       type: "xavier"
1657 |     }
1658 |     bias_filler {
1659 |       type: "constant"
1660 |       value: 0.2
1661 |     }
1662 |   }
1663 | }
1664 | layer {
1665 |   name: "inception_4e/relu_3x3"
1666 |   type: "ReLU"
1667 |   bottom: "inception_4e/3x3"
1668 |   top: "inception_4e/3x3"
1669 | }
1670 | layer {
1671 |   name: "inception_4e/5x5_reduce"
1672 |   type: "Convolution"
1673 |   bottom: "inception_4d/output"
1674 |   top: "inception_4e/5x5_reduce"
1675 |   param {
1676 |     lr_mult: 1
1677 |     decay_mult: 1
1678 |   }
1679 |   param {
1680 |     lr_mult: 2
1681 |     decay_mult: 0
1682 |   }
1683 |   convolution_param {
1684 |     num_output: 32
1685 |     kernel_size: 1
1686 |     weight_filler {
1687 |       type: "xavier"
1688 |     }
1689 |     bias_filler {
1690 |       type: "constant"
1691 |       value: 0.2
1692 |     }
1693 |   }
1694 | }
1695 | layer {
1696 |   name: "inception_4e/relu_5x5_reduce"
1697 |   type: "ReLU"
1698 |   bottom: "inception_4e/5x5_reduce"
1699 |   top: "inception_4e/5x5_reduce"
1700 | }
1701 | layer {
1702 |   name: "inception_4e/5x5"
1703 |   type: "Convolution"
1704 |   bottom: "inception_4e/5x5_reduce"
1705 |   top: "inception_4e/5x5"
1706 |   param {
1707 |     lr_mult: 1
1708 |     decay_mult: 1
1709 |   }
1710 |   param {
1711 |     lr_mult: 2
1712 |     decay_mult: 0
1713 |   }
1714 |   convolution_param {
1715 |     num_output: 128
1716 |     pad: 2
1717 |     kernel_size: 5
1718 |     weight_filler {
1719 |       type: "xavier"
1720 |     }
1721 |     bias_filler {
1722 |       type: "constant"
1723 |       value: 0.2
1724 |     }
1725 |   }
1726 | }
1727 | layer {
1728 |   name: "inception_4e/relu_5x5"
1729 |   type: "ReLU"
1730 |   bottom: "inception_4e/5x5"
1731 |   top: "inception_4e/5x5"
1732 | }
1733 | layer {
1734 |   name: "inception_4e/pool"
1735 |   type: "Pooling"
1736 |   bottom: "inception_4d/output"
1737 |   top: "inception_4e/pool"
1738 |   pooling_param {
1739 |     pool: MAX
1740 |     kernel_size: 3
1741 |     stride: 1
1742 |     pad: 1
1743 |   }
1744 | }
1745 | layer {
1746 |   name: "inception_4e/pool_proj"
1747 |   type: "Convolution"
1748 |   bottom: "inception_4e/pool"
1749 |   top: "inception_4e/pool_proj"
1750 |   param {
1751 |     lr_mult: 1
1752 |     decay_mult: 1
1753 |   }
1754 |   param {
1755 |     lr_mult: 2
1756 |     decay_mult: 0
1757 |   }
1758 |   convolution_param {
1759 |     num_output: 128
1760 |     kernel_size: 1
1761 |     weight_filler {
1762 |       type: "xavier"
1763 |     }
1764 |     bias_filler {
1765 |       type: "constant"
1766 |       value: 0.2
1767 |     }
1768 |   }
1769 | }
1770 | layer {
1771 |   name: "inception_4e/relu_pool_proj"
1772 |   type: "ReLU"
1773 |   bottom: "inception_4e/pool_proj"
1774 |   top: "inception_4e/pool_proj"
1775 | }
1776 | layer {
1777 |   name: "inception_4e/output"
1778 |   type: "Concat"
1779 |   bottom: "inception_4e/1x1"
1780 |   bottom: "inception_4e/3x3"
1781 |   bottom: "inception_4e/5x5"
1782 |   bottom: "inception_4e/pool_proj"
1783 |   top: "inception_4e/output"
1784 | }
1785 | layer {
1786 |   name: "pool4/3x3_s2"
1787 |   type: "Pooling"
1788 |   bottom: "inception_4e/output"
1789 |   top: "pool4/3x3_s2"
1790 |   pooling_param {
1791 |     pool: MAX
1792 |     kernel_size: 3
1793 |     stride: 2
1794 |   }
1795 | }
1796 | layer {
1797 |   name: "inception_5a/1x1"
1798 |   type: "Convolution"
1799 |   bottom: "pool4/3x3_s2"
1800 |   top: "inception_5a/1x1"
1801 |   param {
1802 |     lr_mult: 1
1803 |     decay_mult: 1
1804 |   }
1805 |   param {
1806 |     lr_mult: 2
1807 |     decay_mult: 0
1808 |   }
1809 |   convolution_param {
1810 |     num_output: 256
1811 |     kernel_size: 1
1812 |     weight_filler {
1813 |       type: "xavier"
1814 |     }
1815 |     bias_filler {
1816 |       type: "constant"
1817 |       value: 0.2
1818 |     }
1819 |   }
1820 | }
1821 | layer {
1822 |   name: "inception_5a/relu_1x1"
1823 |   type: "ReLU"
1824 |   bottom: "inception_5a/1x1"
1825 |   top: "inception_5a/1x1"
1826 | }
1827 | layer {
1828 |   name: "inception_5a/3x3_reduce"
1829 |   type: "Convolution"
1830 |   bottom: "pool4/3x3_s2"
1831 |   top: "inception_5a/3x3_reduce"
1832 |   param {
1833 |     lr_mult: 1
1834 |     decay_mult: 1
1835 |   }
1836 |   param {
1837 |     lr_mult: 2
1838 |     decay_mult: 0
1839 |   }
1840 |   convolution_param {
1841 |     num_output: 160
1842 |     kernel_size: 1
1843 |     weight_filler {
1844 |       type: "xavier"
1845 |     }
1846 |     bias_filler {
1847 |       type: "constant"
1848 |       value: 0.2
1849 |     }
1850 |   }
1851 | }
1852 | layer {
1853 |   name: "inception_5a/relu_3x3_reduce"
1854 |   type: "ReLU"
1855 |   bottom: "inception_5a/3x3_reduce"
1856 |   top: "inception_5a/3x3_reduce"
1857 | }
1858 | layer {
1859 |   name: "inception_5a/3x3"
1860 |   type: "Convolution"
1861 |   bottom: "inception_5a/3x3_reduce"
1862 |   top: "inception_5a/3x3"
1863 |   param {
1864 |     lr_mult: 1
1865 |     decay_mult: 1
1866 |   }
1867 |   param {
1868 |     lr_mult: 2
1869 |     decay_mult: 0
1870 |   }
1871 |   convolution_param {
1872 |     num_output: 320
1873 |     pad: 1
1874 |     kernel_size: 3
1875 |     weight_filler {
1876 |       type: "xavier"
1877 |     }
1878 |     bias_filler {
1879 |       type: "constant"
1880 |       value: 0.2
1881 |     }
1882 |   }
1883 | }
1884 | layer {
1885 |   name: "inception_5a/relu_3x3"
1886 |   type: "ReLU"
1887 |   bottom: "inception_5a/3x3"
1888 |   top: "inception_5a/3x3"
1889 | }
1890 | layer {
1891 |   name: "inception_5a/5x5_reduce"
1892 |   type: "Convolution"
1893 |   bottom: "pool4/3x3_s2"
1894 |   top: "inception_5a/5x5_reduce"
1895 |   param {
1896 |     lr_mult: 1
1897 |     decay_mult: 1
1898 |   }
1899 |   param {
1900 |     lr_mult: 2
1901 |     decay_mult: 0
1902 |   }
1903 |   convolution_param {
1904 |     num_output: 32
1905 |     kernel_size: 1
1906 |     weight_filler {
1907 |       type: "xavier"
1908 |     }
1909 |     bias_filler {
1910 |       type: "constant"
1911 |       value: 0.2
1912 |     }
1913 |   }
1914 | }
1915 | layer {
1916 |   name: "inception_5a/relu_5x5_reduce"
1917 |   type: "ReLU"
1918 |   bottom: "inception_5a/5x5_reduce"
1919 |   top: "inception_5a/5x5_reduce"
1920 | }
1921 | layer {
1922 |   name: "inception_5a/5x5"
1923 |   type: "Convolution"
1924 |   bottom: "inception_5a/5x5_reduce"
1925 |   top: "inception_5a/5x5"
1926 |   param {
1927 |     lr_mult: 1
1928 |     decay_mult: 1
1929 |   }
1930 |   param {
1931 |     lr_mult: 2
1932 |     decay_mult: 0
1933 |   }
1934 |   convolution_param {
1935 |     num_output: 128
1936 |     pad: 2
1937 |     kernel_size: 5
1938 |     weight_filler {
1939 |       type: "xavier"
1940 |     }
1941 |     bias_filler {
1942 |       type: "constant"
1943 |       value: 0.2
1944 |     }
1945 |   }
1946 | }
1947 | layer {
1948 |   name: "inception_5a/relu_5x5"
1949 |   type: "ReLU"
1950 |   bottom: "inception_5a/5x5"
1951 |   top: "inception_5a/5x5"
1952 | }
1953 | layer {
1954 |   name: "inception_5a/pool"
1955 |   type: "Pooling"
1956 |   bottom: "pool4/3x3_s2"
1957 |   top: "inception_5a/pool"
1958 |   pooling_param {
1959 |     pool: MAX
1960 |     kernel_size: 3
1961 |     stride: 1
1962 |     pad: 1
1963 |   }
1964 | }
1965 | layer {
1966 |   name: "inception_5a/pool_proj"
1967 |   type: "Convolution"
1968 |   bottom: "inception_5a/pool"
1969 |   top: "inception_5a/pool_proj"
1970 |   param {
1971 |     lr_mult: 1
1972 |     decay_mult: 1
1973 |   }
1974 |   param {
1975 |     lr_mult: 2
1976 |     decay_mult: 0
1977 |   }
1978 |   convolution_param {
1979 |     num_output: 128
1980 |     kernel_size: 1
1981 |     weight_filler {
1982 |       type: "xavier"
1983 |     }
1984 |     bias_filler {
1985 |       type: "constant"
1986 |       value: 0.2
1987 |     }
1988 |   }
1989 | }
1990 | layer {
1991 |   name: "inception_5a/relu_pool_proj"
1992 |   type: "ReLU"
1993 |   bottom: "inception_5a/pool_proj"
1994 |   top: "inception_5a/pool_proj"
1995 | }
1996 | layer {
1997 |   name: "inception_5a/output"
1998 |   type: "Concat"
1999 |   bottom: "inception_5a/1x1"
2000 |   bottom: "inception_5a/3x3"
2001 |   bottom: "inception_5a/5x5"
2002 |   bottom: "inception_5a/pool_proj"
2003 |   top: "inception_5a/output"
2004 | }
2005 | layer {
2006 |   name: "inception_5b/1x1"
2007 |   type: "Convolution"
2008 |   bottom: "inception_5a/output"
2009 |   top: "inception_5b/1x1"
2010 |   param {
2011 |     lr_mult: 1
2012 |     decay_mult: 1
2013 |   }
2014 |   param {
2015 |     lr_mult: 2
2016 |     decay_mult: 0
2017 |   }
2018 |   convolution_param {
2019 |     num_output: 384
2020 |     kernel_size: 1
2021 |     weight_filler {
2022 |       type: "xavier"
2023 |     }
2024 |     bias_filler {
2025 |       type: "constant"
2026 |       value: 0.2
2027 |     }
2028 |   }
2029 | }
2030 | layer {
2031 |   name: "inception_5b/relu_1x1"
2032 |   type: "ReLU"
2033 |   bottom: "inception_5b/1x1"
2034 |   top: "inception_5b/1x1"
2035 | }
2036 | layer {
2037 |   name: "inception_5b/3x3_reduce"
2038 |   type: "Convolution"
2039 |   bottom: "inception_5a/output"
2040 |   top: "inception_5b/3x3_reduce"
2041 |   param {
2042 |     lr_mult: 1
2043 |     decay_mult: 1
2044 |   }
2045 |   param {
2046 |     lr_mult: 2
2047 |     decay_mult: 0
2048 |   }
2049 |   convolution_param {
2050 |     num_output: 192
2051 |     kernel_size: 1
2052 |     weight_filler {
2053 |       type: "xavier"
2054 |     }
2055 |     bias_filler {
2056 |       type: "constant"
2057 |       value: 0.2
2058 |     }
2059 |   }
2060 | }
2061 | layer {
2062 |   name: "inception_5b/relu_3x3_reduce"
2063 |   type: "ReLU"
2064 |   bottom: "inception_5b/3x3_reduce"
2065 |   top: "inception_5b/3x3_reduce"
2066 | }
2067 | layer {
2068 |   name: "inception_5b/3x3"
2069 |   type: "Convolution"
2070 |   bottom: "inception_5b/3x3_reduce"
2071 |   top: "inception_5b/3x3"
2072 |   param {
2073 |     lr_mult: 1
2074 |     decay_mult: 1
2075 |   }
2076 |   param {
2077 |     lr_mult: 2
2078 |     decay_mult: 0
2079 |   }
2080 |   convolution_param {
2081 |     num_output: 384
2082 |     pad: 1
2083 |     kernel_size: 3
2084 |     weight_filler {
2085 |       type: "xavier"
2086 |     }
2087 |     bias_filler {
2088 |       type: "constant"
2089 |       value: 0.2
2090 |     }
2091 |   }
2092 | }
2093 | layer {
2094 |   name: "inception_5b/relu_3x3"
2095 |   type: "ReLU"
2096 |   bottom: "inception_5b/3x3"
2097 |   top: "inception_5b/3x3"
2098 | }
2099 | layer {
2100 |   name: "inception_5b/5x5_reduce"
2101 |   type: "Convolution"
2102 |   bottom: "inception_5a/output"
2103 |   top: "inception_5b/5x5_reduce"
2104 |   param {
2105 |     lr_mult: 1
2106 |     decay_mult: 1
2107 |   }
2108 |   param {
2109 |     lr_mult: 2
2110 |     decay_mult: 0
2111 |   }
2112 |   convolution_param {
2113 |     num_output: 48
2114 |     kernel_size: 1
2115 |     weight_filler {
2116 |       type: "xavier"
2117 |     }
2118 |     bias_filler {
2119 |       type: "constant"
2120 |       value: 0.2
2121 |     }
2122 |   }
2123 | }
2124 | layer {
2125 |   name: "inception_5b/relu_5x5_reduce"
2126 |   type: "ReLU"
2127 |   bottom: "inception_5b/5x5_reduce"
2128 |   top: "inception_5b/5x5_reduce"
2129 | }
2130 | layer {
2131 |   name: "inception_5b/5x5"
2132 |   type: "Convolution"
2133 |   bottom: "inception_5b/5x5_reduce"
2134 |   top: "inception_5b/5x5"
2135 |   param {
2136 |     lr_mult: 1
2137 |     decay_mult: 1
2138 |   }
2139 |   param {
2140 |     lr_mult: 2
2141 |     decay_mult: 0
2142 |   }
2143 |   convolution_param {
2144 |     num_output: 128
2145 |     pad: 2
2146 |     kernel_size: 5
2147 |     weight_filler {
2148 |       type: "xavier"
2149 |     }
2150 |     bias_filler {
2151 |       type: "constant"
2152 |       value: 0.2
2153 |     }
2154 |   }
2155 | }
2156 | layer {
2157 |   name: "inception_5b/relu_5x5"
2158 |   type: "ReLU"
2159 |   bottom: "inception_5b/5x5"
2160 |   top: "inception_5b/5x5"
2161 | }
2162 | layer {
2163 |   name: "inception_5b/pool"
2164 |   type: "Pooling"
2165 |   bottom: "inception_5a/output"
2166 |   top: "inception_5b/pool"
2167 |   pooling_param {
2168 |     pool: MAX
2169 |     kernel_size: 3
2170 |     stride: 1
2171 |     pad: 1
2172 |   }
2173 | }
2174 | layer {
2175 |   name: "inception_5b/pool_proj"
2176 |   type: "Convolution"
2177 |   bottom: "inception_5b/pool"
2178 |   top: "inception_5b/pool_proj"
2179 |   param {
2180 |     lr_mult: 1
2181 |     decay_mult: 1
2182 |   }
2183 |   param {
2184 |     lr_mult: 2
2185 |     decay_mult: 0
2186 |   }
2187 |   convolution_param {
2188 |     num_output: 128
2189 |     kernel_size: 1
2190 |     weight_filler {
2191 |       type: "xavier"
2192 |     }
2193 |     bias_filler {
2194 |       type: "constant"
2195 |       value: 0.2
2196 |     }
2197 |   }
2198 | }
2199 | layer {
2200 |   name: "inception_5b/relu_pool_proj"
2201 |   type: "ReLU"
2202 |   bottom: "inception_5b/pool_proj"
2203 |   top: "inception_5b/pool_proj"
2204 | }
2205 | layer {
2206 |   name: "inception_5b/output"
2207 |   type: "Concat"
2208 |   bottom: "inception_5b/1x1"
2209 |   bottom: "inception_5b/3x3"
2210 |   bottom: "inception_5b/5x5"
2211 |   bottom: "inception_5b/pool_proj"
2212 |   top: "inception_5b/output"
2213 | }
2214 | layer {
2215 |   name: "pool5/7x7_s1"
2216 |   type: "Pooling"
2217 |   bottom: "inception_5b/output"
2218 |   top: "pool5/7x7_s1"
2219 |   pooling_param {
2220 |     pool: AVE
2221 |     kernel_size: 7
2222 |     stride: 1
2223 |   }
2224 | }
2225 | layer {
2226 |   name: "pool5/drop_7x7_s1"
2227 |   type: "Dropout"
2228 |   bottom: "pool5/7x7_s1"
2229 |   top: "pool5/7x7_s1"
2230 |   dropout_param {
2231 |     dropout_ratio: 0.4
2232 |   }
2233 | }
2234 | layer {
2235 |   name: "my-classifier"
2236 |   type: "InnerProduct"
2237 |   bottom: "pool5/7x7_s1"
2238 |   top: "my-classifier"
2239 |   param {
2240 |     lr_mult: 1
2241 |     decay_mult: 1
2242 |   }
2243 |   param {
2244 |     lr_mult: 2
2245 |     decay_mult: 0
2246 |   }
2247 |   inner_product_param {
2248 |     num_output: 3
2249 |     weight_filler {
2250 |       type: "xavier"
2251 |     }
2252 |     bias_filler {
2253 |       type: "constant"
2254 |       value: 0
2255 |     }
2256 |   }
2257 | }
2258 | layer {
2259 |   name: "prob"
2260 |   type: "Softmax"
2261 |   bottom: "my-classifier"
2262 |   top: "prob"
2263 | }
2264 | 


--------------------------------------------------------------------------------
/GoogLeNet/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 10001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 10
26 | # save path
27 | snapshot_prefix: "inception-v1-cervix"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/Inception-ResNet-v2/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | clip_gradients: 40 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 50001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 10
26 | # save path
27 | snapshot_prefix: "inception-resnet-v2-cervix"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/Inception-v3/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 100
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 50001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 100
26 | # save path
27 | snapshot_prefix: "inception-v3-cervix-all"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/Inception-v4/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 10001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 10
26 | # save path
27 | snapshot_prefix: "inception-v4-cervix"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # caffe-model-zoo
2 | ![screenshot](https://user-images.githubusercontent.com/21311442/33640664-cbcbeff2-da6c-11e7-97c8-1ad8d7fdf4c0.png)
3 | 
4 | Caffe Pretrained Models can download from  
5 | Link:[http://pan.baidu.com/s/1pL0ZIYJ](http://pan.baidu.com/s/1pL0ZIYJ)  
6 | Password: [xp68]()
7 | 


--------------------------------------------------------------------------------
/ResNet-101/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 10001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 20
26 | # save path
27 | snapshot_prefix: "resnet-101-cervix"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/ResNet-152/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 10001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 20
26 | # save path
27 | snapshot_prefix: "resnet-152-cervix"
28 | solver_mode: CPU
29 | 


--------------------------------------------------------------------------------
/ResNet-152/train_val.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 10001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 20
26 | # save path
27 | snapshot_prefix: "resnet-152-cervix"
28 | solver_mode: CPU
29 | 


--------------------------------------------------------------------------------
/ResNet-50/deploy.prototxt:
--------------------------------------------------------------------------------
   1 | name: "ResNet-50"
   2 | input: "data"
   3 | input_dim: 1
   4 | input_dim: 3
   5 | input_dim: 224
   6 | input_dim: 224
   7 | layer {
   8 | 	bottom: "data"
   9 | 	top: "conv1"
  10 | 	name: "conv1"
  11 | 	type: "Convolution"
  12 | 	convolution_param {
  13 | 		num_output: 64
  14 | 		kernel_size: 7
  15 | 		pad: 3
  16 | 		stride: 2
  17 | 	}
  18 | }
  19 | 
  20 | layer {
  21 | 	bottom: "conv1"
  22 | 	top: "conv1"
  23 | 	name: "bn_conv1"
  24 | 	type: "BatchNorm"
  25 | 	batch_norm_param {
  26 | 		use_global_stats: true
  27 | 	}
  28 | }
  29 | 
  30 | layer {
  31 | 	bottom: "conv1"
  32 | 	top: "conv1"
  33 | 	name: "scale_conv1"
  34 | 	type: "Scale"
  35 | 	scale_param {
  36 | 		bias_term: true
  37 | 	}
  38 | }
  39 | 
  40 | layer {
  41 | 	bottom: "conv1"
  42 | 	top: "conv1"
  43 | 	name: "conv1_relu"
  44 | 	type: "ReLU"
  45 | }
  46 | 
  47 | layer {
  48 | 	bottom: "conv1"
  49 | 	top: "pool1"
  50 | 	name: "pool1"
  51 | 	type: "Pooling"
  52 | 	pooling_param {
  53 | 		kernel_size: 3
  54 | 		stride: 2
  55 | 		pool: MAX
  56 | 	}
  57 | }
  58 | 
  59 | layer {
  60 | 	bottom: "pool1"
  61 | 	top: "res2a_branch1"
  62 | 	name: "res2a_branch1"
  63 | 	type: "Convolution"
  64 | 	convolution_param {
  65 | 		num_output: 256
  66 | 		kernel_size: 1
  67 | 		pad: 0
  68 | 		stride: 1
  69 | 		bias_term: false
  70 | 	}
  71 | }
  72 | 
  73 | layer {
  74 | 	bottom: "res2a_branch1"
  75 | 	top: "res2a_branch1"
  76 | 	name: "bn2a_branch1"
  77 | 	type: "BatchNorm"
  78 | 	batch_norm_param {
  79 | 		use_global_stats: true
  80 | 	}
  81 | }
  82 | 
  83 | layer {
  84 | 	bottom: "res2a_branch1"
  85 | 	top: "res2a_branch1"
  86 | 	name: "scale2a_branch1"
  87 | 	type: "Scale"
  88 | 	scale_param {
  89 | 		bias_term: true
  90 | 	}
  91 | }
  92 | 
  93 | layer {
  94 | 	bottom: "pool1"
  95 | 	top: "res2a_branch2a"
  96 | 	name: "res2a_branch2a"
  97 | 	type: "Convolution"
  98 | 	convolution_param {
  99 | 		num_output: 64
 100 | 		kernel_size: 1
 101 | 		pad: 0
 102 | 		stride: 1
 103 | 		bias_term: false
 104 | 	}
 105 | }
 106 | 
 107 | layer {
 108 | 	bottom: "res2a_branch2a"
 109 | 	top: "res2a_branch2a"
 110 | 	name: "bn2a_branch2a"
 111 | 	type: "BatchNorm"
 112 | 	batch_norm_param {
 113 | 		use_global_stats: true
 114 | 	}
 115 | }
 116 | 
 117 | layer {
 118 | 	bottom: "res2a_branch2a"
 119 | 	top: "res2a_branch2a"
 120 | 	name: "scale2a_branch2a"
 121 | 	type: "Scale"
 122 | 	scale_param {
 123 | 		bias_term: true
 124 | 	}
 125 | }
 126 | 
 127 | layer {
 128 | 	bottom: "res2a_branch2a"
 129 | 	top: "res2a_branch2a"
 130 | 	name: "res2a_branch2a_relu"
 131 | 	type: "ReLU"
 132 | }
 133 | 
 134 | layer {
 135 | 	bottom: "res2a_branch2a"
 136 | 	top: "res2a_branch2b"
 137 | 	name: "res2a_branch2b"
 138 | 	type: "Convolution"
 139 | 	convolution_param {
 140 | 		num_output: 64
 141 | 		kernel_size: 3
 142 | 		pad: 1
 143 | 		stride: 1
 144 | 		bias_term: false
 145 | 	}
 146 | }
 147 | 
 148 | layer {
 149 | 	bottom: "res2a_branch2b"
 150 | 	top: "res2a_branch2b"
 151 | 	name: "bn2a_branch2b"
 152 | 	type: "BatchNorm"
 153 | 	batch_norm_param {
 154 | 		use_global_stats: true
 155 | 	}
 156 | }
 157 | 
 158 | layer {
 159 | 	bottom: "res2a_branch2b"
 160 | 	top: "res2a_branch2b"
 161 | 	name: "scale2a_branch2b"
 162 | 	type: "Scale"
 163 | 	scale_param {
 164 | 		bias_term: true
 165 | 	}
 166 | }
 167 | 
 168 | layer {
 169 | 	bottom: "res2a_branch2b"
 170 | 	top: "res2a_branch2b"
 171 | 	name: "res2a_branch2b_relu"
 172 | 	type: "ReLU"
 173 | }
 174 | 
 175 | layer {
 176 | 	bottom: "res2a_branch2b"
 177 | 	top: "res2a_branch2c"
 178 | 	name: "res2a_branch2c"
 179 | 	type: "Convolution"
 180 | 	convolution_param {
 181 | 		num_output: 256
 182 | 		kernel_size: 1
 183 | 		pad: 0
 184 | 		stride: 1
 185 | 		bias_term: false
 186 | 	}
 187 | }
 188 | 
 189 | layer {
 190 | 	bottom: "res2a_branch2c"
 191 | 	top: "res2a_branch2c"
 192 | 	name: "bn2a_branch2c"
 193 | 	type: "BatchNorm"
 194 | 	batch_norm_param {
 195 | 		use_global_stats: true
 196 | 	}
 197 | }
 198 | 
 199 | layer {
 200 | 	bottom: "res2a_branch2c"
 201 | 	top: "res2a_branch2c"
 202 | 	name: "scale2a_branch2c"
 203 | 	type: "Scale"
 204 | 	scale_param {
 205 | 		bias_term: true
 206 | 	}
 207 | }
 208 | 
 209 | layer {
 210 | 	bottom: "res2a_branch1"
 211 | 	bottom: "res2a_branch2c"
 212 | 	top: "res2a"
 213 | 	name: "res2a"
 214 | 	type: "Eltwise"
 215 | }
 216 | 
 217 | layer {
 218 | 	bottom: "res2a"
 219 | 	top: "res2a"
 220 | 	name: "res2a_relu"
 221 | 	type: "ReLU"
 222 | }
 223 | 
 224 | layer {
 225 | 	bottom: "res2a"
 226 | 	top: "res2b_branch2a"
 227 | 	name: "res2b_branch2a"
 228 | 	type: "Convolution"
 229 | 	convolution_param {
 230 | 		num_output: 64
 231 | 		kernel_size: 1
 232 | 		pad: 0
 233 | 		stride: 1
 234 | 		bias_term: false
 235 | 	}
 236 | }
 237 | 
 238 | layer {
 239 | 	bottom: "res2b_branch2a"
 240 | 	top: "res2b_branch2a"
 241 | 	name: "bn2b_branch2a"
 242 | 	type: "BatchNorm"
 243 | 	batch_norm_param {
 244 | 		use_global_stats: true
 245 | 	}
 246 | }
 247 | 
 248 | layer {
 249 | 	bottom: "res2b_branch2a"
 250 | 	top: "res2b_branch2a"
 251 | 	name: "scale2b_branch2a"
 252 | 	type: "Scale"
 253 | 	scale_param {
 254 | 		bias_term: true
 255 | 	}
 256 | }
 257 | 
 258 | layer {
 259 | 	bottom: "res2b_branch2a"
 260 | 	top: "res2b_branch2a"
 261 | 	name: "res2b_branch2a_relu"
 262 | 	type: "ReLU"
 263 | }
 264 | 
 265 | layer {
 266 | 	bottom: "res2b_branch2a"
 267 | 	top: "res2b_branch2b"
 268 | 	name: "res2b_branch2b"
 269 | 	type: "Convolution"
 270 | 	convolution_param {
 271 | 		num_output: 64
 272 | 		kernel_size: 3
 273 | 		pad: 1
 274 | 		stride: 1
 275 | 		bias_term: false
 276 | 	}
 277 | }
 278 | 
 279 | layer {
 280 | 	bottom: "res2b_branch2b"
 281 | 	top: "res2b_branch2b"
 282 | 	name: "bn2b_branch2b"
 283 | 	type: "BatchNorm"
 284 | 	batch_norm_param {
 285 | 		use_global_stats: true
 286 | 	}
 287 | }
 288 | 
 289 | layer {
 290 | 	bottom: "res2b_branch2b"
 291 | 	top: "res2b_branch2b"
 292 | 	name: "scale2b_branch2b"
 293 | 	type: "Scale"
 294 | 	scale_param {
 295 | 		bias_term: true
 296 | 	}
 297 | }
 298 | 
 299 | layer {
 300 | 	bottom: "res2b_branch2b"
 301 | 	top: "res2b_branch2b"
 302 | 	name: "res2b_branch2b_relu"
 303 | 	type: "ReLU"
 304 | }
 305 | 
 306 | layer {
 307 | 	bottom: "res2b_branch2b"
 308 | 	top: "res2b_branch2c"
 309 | 	name: "res2b_branch2c"
 310 | 	type: "Convolution"
 311 | 	convolution_param {
 312 | 		num_output: 256
 313 | 		kernel_size: 1
 314 | 		pad: 0
 315 | 		stride: 1
 316 | 		bias_term: false
 317 | 	}
 318 | }
 319 | 
 320 | layer {
 321 | 	bottom: "res2b_branch2c"
 322 | 	top: "res2b_branch2c"
 323 | 	name: "bn2b_branch2c"
 324 | 	type: "BatchNorm"
 325 | 	batch_norm_param {
 326 | 		use_global_stats: true
 327 | 	}
 328 | }
 329 | 
 330 | layer {
 331 | 	bottom: "res2b_branch2c"
 332 | 	top: "res2b_branch2c"
 333 | 	name: "scale2b_branch2c"
 334 | 	type: "Scale"
 335 | 	scale_param {
 336 | 		bias_term: true
 337 | 	}
 338 | }
 339 | 
 340 | layer {
 341 | 	bottom: "res2a"
 342 | 	bottom: "res2b_branch2c"
 343 | 	top: "res2b"
 344 | 	name: "res2b"
 345 | 	type: "Eltwise"
 346 | }
 347 | 
 348 | layer {
 349 | 	bottom: "res2b"
 350 | 	top: "res2b"
 351 | 	name: "res2b_relu"
 352 | 	type: "ReLU"
 353 | }
 354 | 
 355 | layer {
 356 | 	bottom: "res2b"
 357 | 	top: "res2c_branch2a"
 358 | 	name: "res2c_branch2a"
 359 | 	type: "Convolution"
 360 | 	convolution_param {
 361 | 		num_output: 64
 362 | 		kernel_size: 1
 363 | 		pad: 0
 364 | 		stride: 1
 365 | 		bias_term: false
 366 | 	}
 367 | }
 368 | 
 369 | layer {
 370 | 	bottom: "res2c_branch2a"
 371 | 	top: "res2c_branch2a"
 372 | 	name: "bn2c_branch2a"
 373 | 	type: "BatchNorm"
 374 | 	batch_norm_param {
 375 | 		use_global_stats: true
 376 | 	}
 377 | }
 378 | 
 379 | layer {
 380 | 	bottom: "res2c_branch2a"
 381 | 	top: "res2c_branch2a"
 382 | 	name: "scale2c_branch2a"
 383 | 	type: "Scale"
 384 | 	scale_param {
 385 | 		bias_term: true
 386 | 	}
 387 | }
 388 | 
 389 | layer {
 390 | 	bottom: "res2c_branch2a"
 391 | 	top: "res2c_branch2a"
 392 | 	name: "res2c_branch2a_relu"
 393 | 	type: "ReLU"
 394 | }
 395 | 
 396 | layer {
 397 | 	bottom: "res2c_branch2a"
 398 | 	top: "res2c_branch2b"
 399 | 	name: "res2c_branch2b"
 400 | 	type: "Convolution"
 401 | 	convolution_param {
 402 | 		num_output: 64
 403 | 		kernel_size: 3
 404 | 		pad: 1
 405 | 		stride: 1
 406 | 		bias_term: false
 407 | 	}
 408 | }
 409 | 
 410 | layer {
 411 | 	bottom: "res2c_branch2b"
 412 | 	top: "res2c_branch2b"
 413 | 	name: "bn2c_branch2b"
 414 | 	type: "BatchNorm"
 415 | 	batch_norm_param {
 416 | 		use_global_stats: true
 417 | 	}
 418 | }
 419 | 
 420 | layer {
 421 | 	bottom: "res2c_branch2b"
 422 | 	top: "res2c_branch2b"
 423 | 	name: "scale2c_branch2b"
 424 | 	type: "Scale"
 425 | 	scale_param {
 426 | 		bias_term: true
 427 | 	}
 428 | }
 429 | 
 430 | layer {
 431 | 	bottom: "res2c_branch2b"
 432 | 	top: "res2c_branch2b"
 433 | 	name: "res2c_branch2b_relu"
 434 | 	type: "ReLU"
 435 | }
 436 | 
 437 | layer {
 438 | 	bottom: "res2c_branch2b"
 439 | 	top: "res2c_branch2c"
 440 | 	name: "res2c_branch2c"
 441 | 	type: "Convolution"
 442 | 	convolution_param {
 443 | 		num_output: 256
 444 | 		kernel_size: 1
 445 | 		pad: 0
 446 | 		stride: 1
 447 | 		bias_term: false
 448 | 	}
 449 | }
 450 | 
 451 | layer {
 452 | 	bottom: "res2c_branch2c"
 453 | 	top: "res2c_branch2c"
 454 | 	name: "bn2c_branch2c"
 455 | 	type: "BatchNorm"
 456 | 	batch_norm_param {
 457 | 		use_global_stats: true
 458 | 	}
 459 | }
 460 | 
 461 | layer {
 462 | 	bottom: "res2c_branch2c"
 463 | 	top: "res2c_branch2c"
 464 | 	name: "scale2c_branch2c"
 465 | 	type: "Scale"
 466 | 	scale_param {
 467 | 		bias_term: true
 468 | 	}
 469 | }
 470 | 
 471 | layer {
 472 | 	bottom: "res2b"
 473 | 	bottom: "res2c_branch2c"
 474 | 	top: "res2c"
 475 | 	name: "res2c"
 476 | 	type: "Eltwise"
 477 | }
 478 | 
 479 | layer {
 480 | 	bottom: "res2c"
 481 | 	top: "res2c"
 482 | 	name: "res2c_relu"
 483 | 	type: "ReLU"
 484 | }
 485 | 
 486 | layer {
 487 | 	bottom: "res2c"
 488 | 	top: "res3a_branch1"
 489 | 	name: "res3a_branch1"
 490 | 	type: "Convolution"
 491 | 	convolution_param {
 492 | 		num_output: 512
 493 | 		kernel_size: 1
 494 | 		pad: 0
 495 | 		stride: 2
 496 | 		bias_term: false
 497 | 	}
 498 | }
 499 | 
 500 | layer {
 501 | 	bottom: "res3a_branch1"
 502 | 	top: "res3a_branch1"
 503 | 	name: "bn3a_branch1"
 504 | 	type: "BatchNorm"
 505 | 	batch_norm_param {
 506 | 		use_global_stats: true
 507 | 	}
 508 | }
 509 | 
 510 | layer {
 511 | 	bottom: "res3a_branch1"
 512 | 	top: "res3a_branch1"
 513 | 	name: "scale3a_branch1"
 514 | 	type: "Scale"
 515 | 	scale_param {
 516 | 		bias_term: true
 517 | 	}
 518 | }
 519 | 
 520 | layer {
 521 | 	bottom: "res2c"
 522 | 	top: "res3a_branch2a"
 523 | 	name: "res3a_branch2a"
 524 | 	type: "Convolution"
 525 | 	convolution_param {
 526 | 		num_output: 128
 527 | 		kernel_size: 1
 528 | 		pad: 0
 529 | 		stride: 2
 530 | 		bias_term: false
 531 | 	}
 532 | }
 533 | 
 534 | layer {
 535 | 	bottom: "res3a_branch2a"
 536 | 	top: "res3a_branch2a"
 537 | 	name: "bn3a_branch2a"
 538 | 	type: "BatchNorm"
 539 | 	batch_norm_param {
 540 | 		use_global_stats: true
 541 | 	}
 542 | }
 543 | 
 544 | layer {
 545 | 	bottom: "res3a_branch2a"
 546 | 	top: "res3a_branch2a"
 547 | 	name: "scale3a_branch2a"
 548 | 	type: "Scale"
 549 | 	scale_param {
 550 | 		bias_term: true
 551 | 	}
 552 | }
 553 | 
 554 | layer {
 555 | 	bottom: "res3a_branch2a"
 556 | 	top: "res3a_branch2a"
 557 | 	name: "res3a_branch2a_relu"
 558 | 	type: "ReLU"
 559 | }
 560 | 
 561 | layer {
 562 | 	bottom: "res3a_branch2a"
 563 | 	top: "res3a_branch2b"
 564 | 	name: "res3a_branch2b"
 565 | 	type: "Convolution"
 566 | 	convolution_param {
 567 | 		num_output: 128
 568 | 		kernel_size: 3
 569 | 		pad: 1
 570 | 		stride: 1
 571 | 		bias_term: false
 572 | 	}
 573 | }
 574 | 
 575 | layer {
 576 | 	bottom: "res3a_branch2b"
 577 | 	top: "res3a_branch2b"
 578 | 	name: "bn3a_branch2b"
 579 | 	type: "BatchNorm"
 580 | 	batch_norm_param {
 581 | 		use_global_stats: true
 582 | 	}
 583 | }
 584 | 
 585 | layer {
 586 | 	bottom: "res3a_branch2b"
 587 | 	top: "res3a_branch2b"
 588 | 	name: "scale3a_branch2b"
 589 | 	type: "Scale"
 590 | 	scale_param {
 591 | 		bias_term: true
 592 | 	}
 593 | }
 594 | 
 595 | layer {
 596 | 	bottom: "res3a_branch2b"
 597 | 	top: "res3a_branch2b"
 598 | 	name: "res3a_branch2b_relu"
 599 | 	type: "ReLU"
 600 | }
 601 | 
 602 | layer {
 603 | 	bottom: "res3a_branch2b"
 604 | 	top: "res3a_branch2c"
 605 | 	name: "res3a_branch2c"
 606 | 	type: "Convolution"
 607 | 	convolution_param {
 608 | 		num_output: 512
 609 | 		kernel_size: 1
 610 | 		pad: 0
 611 | 		stride: 1
 612 | 		bias_term: false
 613 | 	}
 614 | }
 615 | 
 616 | layer {
 617 | 	bottom: "res3a_branch2c"
 618 | 	top: "res3a_branch2c"
 619 | 	name: "bn3a_branch2c"
 620 | 	type: "BatchNorm"
 621 | 	batch_norm_param {
 622 | 		use_global_stats: true
 623 | 	}
 624 | }
 625 | 
 626 | layer {
 627 | 	bottom: "res3a_branch2c"
 628 | 	top: "res3a_branch2c"
 629 | 	name: "scale3a_branch2c"
 630 | 	type: "Scale"
 631 | 	scale_param {
 632 | 		bias_term: true
 633 | 	}
 634 | }
 635 | 
 636 | layer {
 637 | 	bottom: "res3a_branch1"
 638 | 	bottom: "res3a_branch2c"
 639 | 	top: "res3a"
 640 | 	name: "res3a"
 641 | 	type: "Eltwise"
 642 | }
 643 | 
 644 | layer {
 645 | 	bottom: "res3a"
 646 | 	top: "res3a"
 647 | 	name: "res3a_relu"
 648 | 	type: "ReLU"
 649 | }
 650 | 
 651 | layer {
 652 | 	bottom: "res3a"
 653 | 	top: "res3b_branch2a"
 654 | 	name: "res3b_branch2a"
 655 | 	type: "Convolution"
 656 | 	convolution_param {
 657 | 		num_output: 128
 658 | 		kernel_size: 1
 659 | 		pad: 0
 660 | 		stride: 1
 661 | 		bias_term: false
 662 | 	}
 663 | }
 664 | 
 665 | layer {
 666 | 	bottom: "res3b_branch2a"
 667 | 	top: "res3b_branch2a"
 668 | 	name: "bn3b_branch2a"
 669 | 	type: "BatchNorm"
 670 | 	batch_norm_param {
 671 | 		use_global_stats: true
 672 | 	}
 673 | }
 674 | 
 675 | layer {
 676 | 	bottom: "res3b_branch2a"
 677 | 	top: "res3b_branch2a"
 678 | 	name: "scale3b_branch2a"
 679 | 	type: "Scale"
 680 | 	scale_param {
 681 | 		bias_term: true
 682 | 	}
 683 | }
 684 | 
 685 | layer {
 686 | 	bottom: "res3b_branch2a"
 687 | 	top: "res3b_branch2a"
 688 | 	name: "res3b_branch2a_relu"
 689 | 	type: "ReLU"
 690 | }
 691 | 
 692 | layer {
 693 | 	bottom: "res3b_branch2a"
 694 | 	top: "res3b_branch2b"
 695 | 	name: "res3b_branch2b"
 696 | 	type: "Convolution"
 697 | 	convolution_param {
 698 | 		num_output: 128
 699 | 		kernel_size: 3
 700 | 		pad: 1
 701 | 		stride: 1
 702 | 		bias_term: false
 703 | 	}
 704 | }
 705 | 
 706 | layer {
 707 | 	bottom: "res3b_branch2b"
 708 | 	top: "res3b_branch2b"
 709 | 	name: "bn3b_branch2b"
 710 | 	type: "BatchNorm"
 711 | 	batch_norm_param {
 712 | 		use_global_stats: true
 713 | 	}
 714 | }
 715 | 
 716 | layer {
 717 | 	bottom: "res3b_branch2b"
 718 | 	top: "res3b_branch2b"
 719 | 	name: "scale3b_branch2b"
 720 | 	type: "Scale"
 721 | 	scale_param {
 722 | 		bias_term: true
 723 | 	}
 724 | }
 725 | 
 726 | layer {
 727 | 	bottom: "res3b_branch2b"
 728 | 	top: "res3b_branch2b"
 729 | 	name: "res3b_branch2b_relu"
 730 | 	type: "ReLU"
 731 | }
 732 | 
 733 | layer {
 734 | 	bottom: "res3b_branch2b"
 735 | 	top: "res3b_branch2c"
 736 | 	name: "res3b_branch2c"
 737 | 	type: "Convolution"
 738 | 	convolution_param {
 739 | 		num_output: 512
 740 | 		kernel_size: 1
 741 | 		pad: 0
 742 | 		stride: 1
 743 | 		bias_term: false
 744 | 	}
 745 | }
 746 | 
 747 | layer {
 748 | 	bottom: "res3b_branch2c"
 749 | 	top: "res3b_branch2c"
 750 | 	name: "bn3b_branch2c"
 751 | 	type: "BatchNorm"
 752 | 	batch_norm_param {
 753 | 		use_global_stats: true
 754 | 	}
 755 | }
 756 | 
 757 | layer {
 758 | 	bottom: "res3b_branch2c"
 759 | 	top: "res3b_branch2c"
 760 | 	name: "scale3b_branch2c"
 761 | 	type: "Scale"
 762 | 	scale_param {
 763 | 		bias_term: true
 764 | 	}
 765 | }
 766 | 
 767 | layer {
 768 | 	bottom: "res3a"
 769 | 	bottom: "res3b_branch2c"
 770 | 	top: "res3b"
 771 | 	name: "res3b"
 772 | 	type: "Eltwise"
 773 | }
 774 | 
 775 | layer {
 776 | 	bottom: "res3b"
 777 | 	top: "res3b"
 778 | 	name: "res3b_relu"
 779 | 	type: "ReLU"
 780 | }
 781 | 
 782 | layer {
 783 | 	bottom: "res3b"
 784 | 	top: "res3c_branch2a"
 785 | 	name: "res3c_branch2a"
 786 | 	type: "Convolution"
 787 | 	convolution_param {
 788 | 		num_output: 128
 789 | 		kernel_size: 1
 790 | 		pad: 0
 791 | 		stride: 1
 792 | 		bias_term: false
 793 | 	}
 794 | }
 795 | 
 796 | layer {
 797 | 	bottom: "res3c_branch2a"
 798 | 	top: "res3c_branch2a"
 799 | 	name: "bn3c_branch2a"
 800 | 	type: "BatchNorm"
 801 | 	batch_norm_param {
 802 | 		use_global_stats: true
 803 | 	}
 804 | }
 805 | 
 806 | layer {
 807 | 	bottom: "res3c_branch2a"
 808 | 	top: "res3c_branch2a"
 809 | 	name: "scale3c_branch2a"
 810 | 	type: "Scale"
 811 | 	scale_param {
 812 | 		bias_term: true
 813 | 	}
 814 | }
 815 | 
 816 | layer {
 817 | 	bottom: "res3c_branch2a"
 818 | 	top: "res3c_branch2a"
 819 | 	name: "res3c_branch2a_relu"
 820 | 	type: "ReLU"
 821 | }
 822 | 
 823 | layer {
 824 | 	bottom: "res3c_branch2a"
 825 | 	top: "res3c_branch2b"
 826 | 	name: "res3c_branch2b"
 827 | 	type: "Convolution"
 828 | 	convolution_param {
 829 | 		num_output: 128
 830 | 		kernel_size: 3
 831 | 		pad: 1
 832 | 		stride: 1
 833 | 		bias_term: false
 834 | 	}
 835 | }
 836 | 
 837 | layer {
 838 | 	bottom: "res3c_branch2b"
 839 | 	top: "res3c_branch2b"
 840 | 	name: "bn3c_branch2b"
 841 | 	type: "BatchNorm"
 842 | 	batch_norm_param {
 843 | 		use_global_stats: true
 844 | 	}
 845 | }
 846 | 
 847 | layer {
 848 | 	bottom: "res3c_branch2b"
 849 | 	top: "res3c_branch2b"
 850 | 	name: "scale3c_branch2b"
 851 | 	type: "Scale"
 852 | 	scale_param {
 853 | 		bias_term: true
 854 | 	}
 855 | }
 856 | 
 857 | layer {
 858 | 	bottom: "res3c_branch2b"
 859 | 	top: "res3c_branch2b"
 860 | 	name: "res3c_branch2b_relu"
 861 | 	type: "ReLU"
 862 | }
 863 | 
 864 | layer {
 865 | 	bottom: "res3c_branch2b"
 866 | 	top: "res3c_branch2c"
 867 | 	name: "res3c_branch2c"
 868 | 	type: "Convolution"
 869 | 	convolution_param {
 870 | 		num_output: 512
 871 | 		kernel_size: 1
 872 | 		pad: 0
 873 | 		stride: 1
 874 | 		bias_term: false
 875 | 	}
 876 | }
 877 | 
 878 | layer {
 879 | 	bottom: "res3c_branch2c"
 880 | 	top: "res3c_branch2c"
 881 | 	name: "bn3c_branch2c"
 882 | 	type: "BatchNorm"
 883 | 	batch_norm_param {
 884 | 		use_global_stats: true
 885 | 	}
 886 | }
 887 | 
 888 | layer {
 889 | 	bottom: "res3c_branch2c"
 890 | 	top: "res3c_branch2c"
 891 | 	name: "scale3c_branch2c"
 892 | 	type: "Scale"
 893 | 	scale_param {
 894 | 		bias_term: true
 895 | 	}
 896 | }
 897 | 
 898 | layer {
 899 | 	bottom: "res3b"
 900 | 	bottom: "res3c_branch2c"
 901 | 	top: "res3c"
 902 | 	name: "res3c"
 903 | 	type: "Eltwise"
 904 | }
 905 | 
 906 | layer {
 907 | 	bottom: "res3c"
 908 | 	top: "res3c"
 909 | 	name: "res3c_relu"
 910 | 	type: "ReLU"
 911 | }
 912 | 
 913 | layer {
 914 | 	bottom: "res3c"
 915 | 	top: "res3d_branch2a"
 916 | 	name: "res3d_branch2a"
 917 | 	type: "Convolution"
 918 | 	convolution_param {
 919 | 		num_output: 128
 920 | 		kernel_size: 1
 921 | 		pad: 0
 922 | 		stride: 1
 923 | 		bias_term: false
 924 | 	}
 925 | }
 926 | 
 927 | layer {
 928 | 	bottom: "res3d_branch2a"
 929 | 	top: "res3d_branch2a"
 930 | 	name: "bn3d_branch2a"
 931 | 	type: "BatchNorm"
 932 | 	batch_norm_param {
 933 | 		use_global_stats: true
 934 | 	}
 935 | }
 936 | 
 937 | layer {
 938 | 	bottom: "res3d_branch2a"
 939 | 	top: "res3d_branch2a"
 940 | 	name: "scale3d_branch2a"
 941 | 	type: "Scale"
 942 | 	scale_param {
 943 | 		bias_term: true
 944 | 	}
 945 | }
 946 | 
 947 | layer {
 948 | 	bottom: "res3d_branch2a"
 949 | 	top: "res3d_branch2a"
 950 | 	name: "res3d_branch2a_relu"
 951 | 	type: "ReLU"
 952 | }
 953 | 
 954 | layer {
 955 | 	bottom: "res3d_branch2a"
 956 | 	top: "res3d_branch2b"
 957 | 	name: "res3d_branch2b"
 958 | 	type: "Convolution"
 959 | 	convolution_param {
 960 | 		num_output: 128
 961 | 		kernel_size: 3
 962 | 		pad: 1
 963 | 		stride: 1
 964 | 		bias_term: false
 965 | 	}
 966 | }
 967 | 
 968 | layer {
 969 | 	bottom: "res3d_branch2b"
 970 | 	top: "res3d_branch2b"
 971 | 	name: "bn3d_branch2b"
 972 | 	type: "BatchNorm"
 973 | 	batch_norm_param {
 974 | 		use_global_stats: true
 975 | 	}
 976 | }
 977 | 
 978 | layer {
 979 | 	bottom: "res3d_branch2b"
 980 | 	top: "res3d_branch2b"
 981 | 	name: "scale3d_branch2b"
 982 | 	type: "Scale"
 983 | 	scale_param {
 984 | 		bias_term: true
 985 | 	}
 986 | }
 987 | 
 988 | layer {
 989 | 	bottom: "res3d_branch2b"
 990 | 	top: "res3d_branch2b"
 991 | 	name: "res3d_branch2b_relu"
 992 | 	type: "ReLU"
 993 | }
 994 | 
 995 | layer {
 996 | 	bottom: "res3d_branch2b"
 997 | 	top: "res3d_branch2c"
 998 | 	name: "res3d_branch2c"
 999 | 	type: "Convolution"
1000 | 	convolution_param {
1001 | 		num_output: 512
1002 | 		kernel_size: 1
1003 | 		pad: 0
1004 | 		stride: 1
1005 | 		bias_term: false
1006 | 	}
1007 | }
1008 | 
1009 | layer {
1010 | 	bottom: "res3d_branch2c"
1011 | 	top: "res3d_branch2c"
1012 | 	name: "bn3d_branch2c"
1013 | 	type: "BatchNorm"
1014 | 	batch_norm_param {
1015 | 		use_global_stats: true
1016 | 	}
1017 | }
1018 | 
1019 | layer {
1020 | 	bottom: "res3d_branch2c"
1021 | 	top: "res3d_branch2c"
1022 | 	name: "scale3d_branch2c"
1023 | 	type: "Scale"
1024 | 	scale_param {
1025 | 		bias_term: true
1026 | 	}
1027 | }
1028 | 
1029 | layer {
1030 | 	bottom: "res3c"
1031 | 	bottom: "res3d_branch2c"
1032 | 	top: "res3d"
1033 | 	name: "res3d"
1034 | 	type: "Eltwise"
1035 | }
1036 | 
1037 | layer {
1038 | 	bottom: "res3d"
1039 | 	top: "res3d"
1040 | 	name: "res3d_relu"
1041 | 	type: "ReLU"
1042 | }
1043 | 
1044 | layer {
1045 | 	bottom: "res3d"
1046 | 	top: "res4a_branch1"
1047 | 	name: "res4a_branch1"
1048 | 	type: "Convolution"
1049 | 	convolution_param {
1050 | 		num_output: 1024
1051 | 		kernel_size: 1
1052 | 		pad: 0
1053 | 		stride: 2
1054 | 		bias_term: false
1055 | 	}
1056 | }
1057 | 
1058 | layer {
1059 | 	bottom: "res4a_branch1"
1060 | 	top: "res4a_branch1"
1061 | 	name: "bn4a_branch1"
1062 | 	type: "BatchNorm"
1063 | 	batch_norm_param {
1064 | 		use_global_stats: true
1065 | 	}
1066 | }
1067 | 
1068 | layer {
1069 | 	bottom: "res4a_branch1"
1070 | 	top: "res4a_branch1"
1071 | 	name: "scale4a_branch1"
1072 | 	type: "Scale"
1073 | 	scale_param {
1074 | 		bias_term: true
1075 | 	}
1076 | }
1077 | 
1078 | layer {
1079 | 	bottom: "res3d"
1080 | 	top: "res4a_branch2a"
1081 | 	name: "res4a_branch2a"
1082 | 	type: "Convolution"
1083 | 	convolution_param {
1084 | 		num_output: 256
1085 | 		kernel_size: 1
1086 | 		pad: 0
1087 | 		stride: 2
1088 | 		bias_term: false
1089 | 	}
1090 | }
1091 | 
1092 | layer {
1093 | 	bottom: "res4a_branch2a"
1094 | 	top: "res4a_branch2a"
1095 | 	name: "bn4a_branch2a"
1096 | 	type: "BatchNorm"
1097 | 	batch_norm_param {
1098 | 		use_global_stats: true
1099 | 	}
1100 | }
1101 | 
1102 | layer {
1103 | 	bottom: "res4a_branch2a"
1104 | 	top: "res4a_branch2a"
1105 | 	name: "scale4a_branch2a"
1106 | 	type: "Scale"
1107 | 	scale_param {
1108 | 		bias_term: true
1109 | 	}
1110 | }
1111 | 
1112 | layer {
1113 | 	bottom: "res4a_branch2a"
1114 | 	top: "res4a_branch2a"
1115 | 	name: "res4a_branch2a_relu"
1116 | 	type: "ReLU"
1117 | }
1118 | 
1119 | layer {
1120 | 	bottom: "res4a_branch2a"
1121 | 	top: "res4a_branch2b"
1122 | 	name: "res4a_branch2b"
1123 | 	type: "Convolution"
1124 | 	convolution_param {
1125 | 		num_output: 256
1126 | 		kernel_size: 3
1127 | 		pad: 1
1128 | 		stride: 1
1129 | 		bias_term: false
1130 | 	}
1131 | }
1132 | 
1133 | layer {
1134 | 	bottom: "res4a_branch2b"
1135 | 	top: "res4a_branch2b"
1136 | 	name: "bn4a_branch2b"
1137 | 	type: "BatchNorm"
1138 | 	batch_norm_param {
1139 | 		use_global_stats: true
1140 | 	}
1141 | }
1142 | 
1143 | layer {
1144 | 	bottom: "res4a_branch2b"
1145 | 	top: "res4a_branch2b"
1146 | 	name: "scale4a_branch2b"
1147 | 	type: "Scale"
1148 | 	scale_param {
1149 | 		bias_term: true
1150 | 	}
1151 | }
1152 | 
1153 | layer {
1154 | 	bottom: "res4a_branch2b"
1155 | 	top: "res4a_branch2b"
1156 | 	name: "res4a_branch2b_relu"
1157 | 	type: "ReLU"
1158 | }
1159 | 
1160 | layer {
1161 | 	bottom: "res4a_branch2b"
1162 | 	top: "res4a_branch2c"
1163 | 	name: "res4a_branch2c"
1164 | 	type: "Convolution"
1165 | 	convolution_param {
1166 | 		num_output: 1024
1167 | 		kernel_size: 1
1168 | 		pad: 0
1169 | 		stride: 1
1170 | 		bias_term: false
1171 | 	}
1172 | }
1173 | 
1174 | layer {
1175 | 	bottom: "res4a_branch2c"
1176 | 	top: "res4a_branch2c"
1177 | 	name: "bn4a_branch2c"
1178 | 	type: "BatchNorm"
1179 | 	batch_norm_param {
1180 | 		use_global_stats: true
1181 | 	}
1182 | }
1183 | 
1184 | layer {
1185 | 	bottom: "res4a_branch2c"
1186 | 	top: "res4a_branch2c"
1187 | 	name: "scale4a_branch2c"
1188 | 	type: "Scale"
1189 | 	scale_param {
1190 | 		bias_term: true
1191 | 	}
1192 | }
1193 | 
1194 | layer {
1195 | 	bottom: "res4a_branch1"
1196 | 	bottom: "res4a_branch2c"
1197 | 	top: "res4a"
1198 | 	name: "res4a"
1199 | 	type: "Eltwise"
1200 | }
1201 | 
1202 | layer {
1203 | 	bottom: "res4a"
1204 | 	top: "res4a"
1205 | 	name: "res4a_relu"
1206 | 	type: "ReLU"
1207 | }
1208 | 
1209 | layer {
1210 | 	bottom: "res4a"
1211 | 	top: "res4b_branch2a"
1212 | 	name: "res4b_branch2a"
1213 | 	type: "Convolution"
1214 | 	convolution_param {
1215 | 		num_output: 256
1216 | 		kernel_size: 1
1217 | 		pad: 0
1218 | 		stride: 1
1219 | 		bias_term: false
1220 | 	}
1221 | }
1222 | 
1223 | layer {
1224 | 	bottom: "res4b_branch2a"
1225 | 	top: "res4b_branch2a"
1226 | 	name: "bn4b_branch2a"
1227 | 	type: "BatchNorm"
1228 | 	batch_norm_param {
1229 | 		use_global_stats: true
1230 | 	}
1231 | }
1232 | 
1233 | layer {
1234 | 	bottom: "res4b_branch2a"
1235 | 	top: "res4b_branch2a"
1236 | 	name: "scale4b_branch2a"
1237 | 	type: "Scale"
1238 | 	scale_param {
1239 | 		bias_term: true
1240 | 	}
1241 | }
1242 | 
1243 | layer {
1244 | 	bottom: "res4b_branch2a"
1245 | 	top: "res4b_branch2a"
1246 | 	name: "res4b_branch2a_relu"
1247 | 	type: "ReLU"
1248 | }
1249 | 
1250 | layer {
1251 | 	bottom: "res4b_branch2a"
1252 | 	top: "res4b_branch2b"
1253 | 	name: "res4b_branch2b"
1254 | 	type: "Convolution"
1255 | 	convolution_param {
1256 | 		num_output: 256
1257 | 		kernel_size: 3
1258 | 		pad: 1
1259 | 		stride: 1
1260 | 		bias_term: false
1261 | 	}
1262 | }
1263 | 
1264 | layer {
1265 | 	bottom: "res4b_branch2b"
1266 | 	top: "res4b_branch2b"
1267 | 	name: "bn4b_branch2b"
1268 | 	type: "BatchNorm"
1269 | 	batch_norm_param {
1270 | 		use_global_stats: true
1271 | 	}
1272 | }
1273 | 
1274 | layer {
1275 | 	bottom: "res4b_branch2b"
1276 | 	top: "res4b_branch2b"
1277 | 	name: "scale4b_branch2b"
1278 | 	type: "Scale"
1279 | 	scale_param {
1280 | 		bias_term: true
1281 | 	}
1282 | }
1283 | 
1284 | layer {
1285 | 	bottom: "res4b_branch2b"
1286 | 	top: "res4b_branch2b"
1287 | 	name: "res4b_branch2b_relu"
1288 | 	type: "ReLU"
1289 | }
1290 | 
1291 | layer {
1292 | 	bottom: "res4b_branch2b"
1293 | 	top: "res4b_branch2c"
1294 | 	name: "res4b_branch2c"
1295 | 	type: "Convolution"
1296 | 	convolution_param {
1297 | 		num_output: 1024
1298 | 		kernel_size: 1
1299 | 		pad: 0
1300 | 		stride: 1
1301 | 		bias_term: false
1302 | 	}
1303 | }
1304 | 
1305 | layer {
1306 | 	bottom: "res4b_branch2c"
1307 | 	top: "res4b_branch2c"
1308 | 	name: "bn4b_branch2c"
1309 | 	type: "BatchNorm"
1310 | 	batch_norm_param {
1311 | 		use_global_stats: true
1312 | 	}
1313 | }
1314 | 
1315 | layer {
1316 | 	bottom: "res4b_branch2c"
1317 | 	top: "res4b_branch2c"
1318 | 	name: "scale4b_branch2c"
1319 | 	type: "Scale"
1320 | 	scale_param {
1321 | 		bias_term: true
1322 | 	}
1323 | }
1324 | 
1325 | layer {
1326 | 	bottom: "res4a"
1327 | 	bottom: "res4b_branch2c"
1328 | 	top: "res4b"
1329 | 	name: "res4b"
1330 | 	type: "Eltwise"
1331 | }
1332 | 
1333 | layer {
1334 | 	bottom: "res4b"
1335 | 	top: "res4b"
1336 | 	name: "res4b_relu"
1337 | 	type: "ReLU"
1338 | }
1339 | 
1340 | layer {
1341 | 	bottom: "res4b"
1342 | 	top: "res4c_branch2a"
1343 | 	name: "res4c_branch2a"
1344 | 	type: "Convolution"
1345 | 	convolution_param {
1346 | 		num_output: 256
1347 | 		kernel_size: 1
1348 | 		pad: 0
1349 | 		stride: 1
1350 | 		bias_term: false
1351 | 	}
1352 | }
1353 | 
1354 | layer {
1355 | 	bottom: "res4c_branch2a"
1356 | 	top: "res4c_branch2a"
1357 | 	name: "bn4c_branch2a"
1358 | 	type: "BatchNorm"
1359 | 	batch_norm_param {
1360 | 		use_global_stats: true
1361 | 	}
1362 | }
1363 | 
1364 | layer {
1365 | 	bottom: "res4c_branch2a"
1366 | 	top: "res4c_branch2a"
1367 | 	name: "scale4c_branch2a"
1368 | 	type: "Scale"
1369 | 	scale_param {
1370 | 		bias_term: true
1371 | 	}
1372 | }
1373 | 
1374 | layer {
1375 | 	bottom: "res4c_branch2a"
1376 | 	top: "res4c_branch2a"
1377 | 	name: "res4c_branch2a_relu"
1378 | 	type: "ReLU"
1379 | }
1380 | 
1381 | layer {
1382 | 	bottom: "res4c_branch2a"
1383 | 	top: "res4c_branch2b"
1384 | 	name: "res4c_branch2b"
1385 | 	type: "Convolution"
1386 | 	convolution_param {
1387 | 		num_output: 256
1388 | 		kernel_size: 3
1389 | 		pad: 1
1390 | 		stride: 1
1391 | 		bias_term: false
1392 | 	}
1393 | }
1394 | 
1395 | layer {
1396 | 	bottom: "res4c_branch2b"
1397 | 	top: "res4c_branch2b"
1398 | 	name: "bn4c_branch2b"
1399 | 	type: "BatchNorm"
1400 | 	batch_norm_param {
1401 | 		use_global_stats: true
1402 | 	}
1403 | }
1404 | 
1405 | layer {
1406 | 	bottom: "res4c_branch2b"
1407 | 	top: "res4c_branch2b"
1408 | 	name: "scale4c_branch2b"
1409 | 	type: "Scale"
1410 | 	scale_param {
1411 | 		bias_term: true
1412 | 	}
1413 | }
1414 | 
1415 | layer {
1416 | 	bottom: "res4c_branch2b"
1417 | 	top: "res4c_branch2b"
1418 | 	name: "res4c_branch2b_relu"
1419 | 	type: "ReLU"
1420 | }
1421 | 
1422 | layer {
1423 | 	bottom: "res4c_branch2b"
1424 | 	top: "res4c_branch2c"
1425 | 	name: "res4c_branch2c"
1426 | 	type: "Convolution"
1427 | 	convolution_param {
1428 | 		num_output: 1024
1429 | 		kernel_size: 1
1430 | 		pad: 0
1431 | 		stride: 1
1432 | 		bias_term: false
1433 | 	}
1434 | }
1435 | 
1436 | layer {
1437 | 	bottom: "res4c_branch2c"
1438 | 	top: "res4c_branch2c"
1439 | 	name: "bn4c_branch2c"
1440 | 	type: "BatchNorm"
1441 | 	batch_norm_param {
1442 | 		use_global_stats: true
1443 | 	}
1444 | }
1445 | 
1446 | layer {
1447 | 	bottom: "res4c_branch2c"
1448 | 	top: "res4c_branch2c"
1449 | 	name: "scale4c_branch2c"
1450 | 	type: "Scale"
1451 | 	scale_param {
1452 | 		bias_term: true
1453 | 	}
1454 | }
1455 | 
1456 | layer {
1457 | 	bottom: "res4b"
1458 | 	bottom: "res4c_branch2c"
1459 | 	top: "res4c"
1460 | 	name: "res4c"
1461 | 	type: "Eltwise"
1462 | }
1463 | 
1464 | layer {
1465 | 	bottom: "res4c"
1466 | 	top: "res4c"
1467 | 	name: "res4c_relu"
1468 | 	type: "ReLU"
1469 | }
1470 | 
1471 | layer {
1472 | 	bottom: "res4c"
1473 | 	top: "res4d_branch2a"
1474 | 	name: "res4d_branch2a"
1475 | 	type: "Convolution"
1476 | 	convolution_param {
1477 | 		num_output: 256
1478 | 		kernel_size: 1
1479 | 		pad: 0
1480 | 		stride: 1
1481 | 		bias_term: false
1482 | 	}
1483 | }
1484 | 
1485 | layer {
1486 | 	bottom: "res4d_branch2a"
1487 | 	top: "res4d_branch2a"
1488 | 	name: "bn4d_branch2a"
1489 | 	type: "BatchNorm"
1490 | 	batch_norm_param {
1491 | 		use_global_stats: true
1492 | 	}
1493 | }
1494 | 
1495 | layer {
1496 | 	bottom: "res4d_branch2a"
1497 | 	top: "res4d_branch2a"
1498 | 	name: "scale4d_branch2a"
1499 | 	type: "Scale"
1500 | 	scale_param {
1501 | 		bias_term: true
1502 | 	}
1503 | }
1504 | 
1505 | layer {
1506 | 	bottom: "res4d_branch2a"
1507 | 	top: "res4d_branch2a"
1508 | 	name: "res4d_branch2a_relu"
1509 | 	type: "ReLU"
1510 | }
1511 | 
1512 | layer {
1513 | 	bottom: "res4d_branch2a"
1514 | 	top: "res4d_branch2b"
1515 | 	name: "res4d_branch2b"
1516 | 	type: "Convolution"
1517 | 	convolution_param {
1518 | 		num_output: 256
1519 | 		kernel_size: 3
1520 | 		pad: 1
1521 | 		stride: 1
1522 | 		bias_term: false
1523 | 	}
1524 | }
1525 | 
1526 | layer {
1527 | 	bottom: "res4d_branch2b"
1528 | 	top: "res4d_branch2b"
1529 | 	name: "bn4d_branch2b"
1530 | 	type: "BatchNorm"
1531 | 	batch_norm_param {
1532 | 		use_global_stats: true
1533 | 	}
1534 | }
1535 | 
1536 | layer {
1537 | 	bottom: "res4d_branch2b"
1538 | 	top: "res4d_branch2b"
1539 | 	name: "scale4d_branch2b"
1540 | 	type: "Scale"
1541 | 	scale_param {
1542 | 		bias_term: true
1543 | 	}
1544 | }
1545 | 
1546 | layer {
1547 | 	bottom: "res4d_branch2b"
1548 | 	top: "res4d_branch2b"
1549 | 	name: "res4d_branch2b_relu"
1550 | 	type: "ReLU"
1551 | }
1552 | 
1553 | layer {
1554 | 	bottom: "res4d_branch2b"
1555 | 	top: "res4d_branch2c"
1556 | 	name: "res4d_branch2c"
1557 | 	type: "Convolution"
1558 | 	convolution_param {
1559 | 		num_output: 1024
1560 | 		kernel_size: 1
1561 | 		pad: 0
1562 | 		stride: 1
1563 | 		bias_term: false
1564 | 	}
1565 | }
1566 | 
1567 | layer {
1568 | 	bottom: "res4d_branch2c"
1569 | 	top: "res4d_branch2c"
1570 | 	name: "bn4d_branch2c"
1571 | 	type: "BatchNorm"
1572 | 	batch_norm_param {
1573 | 		use_global_stats: true
1574 | 	}
1575 | }
1576 | 
1577 | layer {
1578 | 	bottom: "res4d_branch2c"
1579 | 	top: "res4d_branch2c"
1580 | 	name: "scale4d_branch2c"
1581 | 	type: "Scale"
1582 | 	scale_param {
1583 | 		bias_term: true
1584 | 	}
1585 | }
1586 | 
1587 | layer {
1588 | 	bottom: "res4c"
1589 | 	bottom: "res4d_branch2c"
1590 | 	top: "res4d"
1591 | 	name: "res4d"
1592 | 	type: "Eltwise"
1593 | }
1594 | 
1595 | layer {
1596 | 	bottom: "res4d"
1597 | 	top: "res4d"
1598 | 	name: "res4d_relu"
1599 | 	type: "ReLU"
1600 | }
1601 | 
1602 | layer {
1603 | 	bottom: "res4d"
1604 | 	top: "res4e_branch2a"
1605 | 	name: "res4e_branch2a"
1606 | 	type: "Convolution"
1607 | 	convolution_param {
1608 | 		num_output: 256
1609 | 		kernel_size: 1
1610 | 		pad: 0
1611 | 		stride: 1
1612 | 		bias_term: false
1613 | 	}
1614 | }
1615 | 
1616 | layer {
1617 | 	bottom: "res4e_branch2a"
1618 | 	top: "res4e_branch2a"
1619 | 	name: "bn4e_branch2a"
1620 | 	type: "BatchNorm"
1621 | 	batch_norm_param {
1622 | 		use_global_stats: true
1623 | 	}
1624 | }
1625 | 
1626 | layer {
1627 | 	bottom: "res4e_branch2a"
1628 | 	top: "res4e_branch2a"
1629 | 	name: "scale4e_branch2a"
1630 | 	type: "Scale"
1631 | 	scale_param {
1632 | 		bias_term: true
1633 | 	}
1634 | }
1635 | 
1636 | layer {
1637 | 	bottom: "res4e_branch2a"
1638 | 	top: "res4e_branch2a"
1639 | 	name: "res4e_branch2a_relu"
1640 | 	type: "ReLU"
1641 | }
1642 | 
1643 | layer {
1644 | 	bottom: "res4e_branch2a"
1645 | 	top: "res4e_branch2b"
1646 | 	name: "res4e_branch2b"
1647 | 	type: "Convolution"
1648 | 	convolution_param {
1649 | 		num_output: 256
1650 | 		kernel_size: 3
1651 | 		pad: 1
1652 | 		stride: 1
1653 | 		bias_term: false
1654 | 	}
1655 | }
1656 | 
1657 | layer {
1658 | 	bottom: "res4e_branch2b"
1659 | 	top: "res4e_branch2b"
1660 | 	name: "bn4e_branch2b"
1661 | 	type: "BatchNorm"
1662 | 	batch_norm_param {
1663 | 		use_global_stats: true
1664 | 	}
1665 | }
1666 | 
1667 | layer {
1668 | 	bottom: "res4e_branch2b"
1669 | 	top: "res4e_branch2b"
1670 | 	name: "scale4e_branch2b"
1671 | 	type: "Scale"
1672 | 	scale_param {
1673 | 		bias_term: true
1674 | 	}
1675 | }
1676 | 
1677 | layer {
1678 | 	bottom: "res4e_branch2b"
1679 | 	top: "res4e_branch2b"
1680 | 	name: "res4e_branch2b_relu"
1681 | 	type: "ReLU"
1682 | }
1683 | 
1684 | layer {
1685 | 	bottom: "res4e_branch2b"
1686 | 	top: "res4e_branch2c"
1687 | 	name: "res4e_branch2c"
1688 | 	type: "Convolution"
1689 | 	convolution_param {
1690 | 		num_output: 1024
1691 | 		kernel_size: 1
1692 | 		pad: 0
1693 | 		stride: 1
1694 | 		bias_term: false
1695 | 	}
1696 | }
1697 | 
1698 | layer {
1699 | 	bottom: "res4e_branch2c"
1700 | 	top: "res4e_branch2c"
1701 | 	name: "bn4e_branch2c"
1702 | 	type: "BatchNorm"
1703 | 	batch_norm_param {
1704 | 		use_global_stats: true
1705 | 	}
1706 | }
1707 | 
1708 | layer {
1709 | 	bottom: "res4e_branch2c"
1710 | 	top: "res4e_branch2c"
1711 | 	name: "scale4e_branch2c"
1712 | 	type: "Scale"
1713 | 	scale_param {
1714 | 		bias_term: true
1715 | 	}
1716 | }
1717 | 
1718 | layer {
1719 | 	bottom: "res4d"
1720 | 	bottom: "res4e_branch2c"
1721 | 	top: "res4e"
1722 | 	name: "res4e"
1723 | 	type: "Eltwise"
1724 | }
1725 | 
1726 | layer {
1727 | 	bottom: "res4e"
1728 | 	top: "res4e"
1729 | 	name: "res4e_relu"
1730 | 	type: "ReLU"
1731 | }
1732 | 
1733 | layer {
1734 | 	bottom: "res4e"
1735 | 	top: "res4f_branch2a"
1736 | 	name: "res4f_branch2a"
1737 | 	type: "Convolution"
1738 | 	convolution_param {
1739 | 		num_output: 256
1740 | 		kernel_size: 1
1741 | 		pad: 0
1742 | 		stride: 1
1743 | 		bias_term: false
1744 | 	}
1745 | }
1746 | 
1747 | layer {
1748 | 	bottom: "res4f_branch2a"
1749 | 	top: "res4f_branch2a"
1750 | 	name: "bn4f_branch2a"
1751 | 	type: "BatchNorm"
1752 | 	batch_norm_param {
1753 | 		use_global_stats: true
1754 | 	}
1755 | }
1756 | 
1757 | layer {
1758 | 	bottom: "res4f_branch2a"
1759 | 	top: "res4f_branch2a"
1760 | 	name: "scale4f_branch2a"
1761 | 	type: "Scale"
1762 | 	scale_param {
1763 | 		bias_term: true
1764 | 	}
1765 | }
1766 | 
1767 | layer {
1768 | 	bottom: "res4f_branch2a"
1769 | 	top: "res4f_branch2a"
1770 | 	name: "res4f_branch2a_relu"
1771 | 	type: "ReLU"
1772 | }
1773 | 
1774 | layer {
1775 | 	bottom: "res4f_branch2a"
1776 | 	top: "res4f_branch2b"
1777 | 	name: "res4f_branch2b"
1778 | 	type: "Convolution"
1779 | 	convolution_param {
1780 | 		num_output: 256
1781 | 		kernel_size: 3
1782 | 		pad: 1
1783 | 		stride: 1
1784 | 		bias_term: false
1785 | 	}
1786 | }
1787 | 
1788 | layer {
1789 | 	bottom: "res4f_branch2b"
1790 | 	top: "res4f_branch2b"
1791 | 	name: "bn4f_branch2b"
1792 | 	type: "BatchNorm"
1793 | 	batch_norm_param {
1794 | 		use_global_stats: true
1795 | 	}
1796 | }
1797 | 
1798 | layer {
1799 | 	bottom: "res4f_branch2b"
1800 | 	top: "res4f_branch2b"
1801 | 	name: "scale4f_branch2b"
1802 | 	type: "Scale"
1803 | 	scale_param {
1804 | 		bias_term: true
1805 | 	}
1806 | }
1807 | 
1808 | layer {
1809 | 	bottom: "res4f_branch2b"
1810 | 	top: "res4f_branch2b"
1811 | 	name: "res4f_branch2b_relu"
1812 | 	type: "ReLU"
1813 | }
1814 | 
1815 | layer {
1816 | 	bottom: "res4f_branch2b"
1817 | 	top: "res4f_branch2c"
1818 | 	name: "res4f_branch2c"
1819 | 	type: "Convolution"
1820 | 	convolution_param {
1821 | 		num_output: 1024
1822 | 		kernel_size: 1
1823 | 		pad: 0
1824 | 		stride: 1
1825 | 		bias_term: false
1826 | 	}
1827 | }
1828 | 
1829 | layer {
1830 | 	bottom: "res4f_branch2c"
1831 | 	top: "res4f_branch2c"
1832 | 	name: "bn4f_branch2c"
1833 | 	type: "BatchNorm"
1834 | 	batch_norm_param {
1835 | 		use_global_stats: true
1836 | 	}
1837 | }
1838 | 
1839 | layer {
1840 | 	bottom: "res4f_branch2c"
1841 | 	top: "res4f_branch2c"
1842 | 	name: "scale4f_branch2c"
1843 | 	type: "Scale"
1844 | 	scale_param {
1845 | 		bias_term: true
1846 | 	}
1847 | }
1848 | 
1849 | layer {
1850 | 	bottom: "res4e"
1851 | 	bottom: "res4f_branch2c"
1852 | 	top: "res4f"
1853 | 	name: "res4f"
1854 | 	type: "Eltwise"
1855 | }
1856 | 
1857 | layer {
1858 | 	bottom: "res4f"
1859 | 	top: "res4f"
1860 | 	name: "res4f_relu"
1861 | 	type: "ReLU"
1862 | }
1863 | 
1864 | layer {
1865 | 	bottom: "res4f"
1866 | 	top: "res5a_branch1"
1867 | 	name: "res5a_branch1"
1868 | 	type: "Convolution"
1869 | 	convolution_param {
1870 | 		num_output: 2048
1871 | 		kernel_size: 1
1872 | 		pad: 0
1873 | 		stride: 2
1874 | 		bias_term: false
1875 | 	}
1876 | }
1877 | 
1878 | layer {
1879 | 	bottom: "res5a_branch1"
1880 | 	top: "res5a_branch1"
1881 | 	name: "bn5a_branch1"
1882 | 	type: "BatchNorm"
1883 | 	batch_norm_param {
1884 | 		use_global_stats: true
1885 | 	}
1886 | }
1887 | 
1888 | layer {
1889 | 	bottom: "res5a_branch1"
1890 | 	top: "res5a_branch1"
1891 | 	name: "scale5a_branch1"
1892 | 	type: "Scale"
1893 | 	scale_param {
1894 | 		bias_term: true
1895 | 	}
1896 | }
1897 | 
1898 | layer {
1899 | 	bottom: "res4f"
1900 | 	top: "res5a_branch2a"
1901 | 	name: "res5a_branch2a"
1902 | 	type: "Convolution"
1903 | 	convolution_param {
1904 | 		num_output: 512
1905 | 		kernel_size: 1
1906 | 		pad: 0
1907 | 		stride: 2
1908 | 		bias_term: false
1909 | 	}
1910 | }
1911 | 
1912 | layer {
1913 | 	bottom: "res5a_branch2a"
1914 | 	top: "res5a_branch2a"
1915 | 	name: "bn5a_branch2a"
1916 | 	type: "BatchNorm"
1917 | 	batch_norm_param {
1918 | 		use_global_stats: true
1919 | 	}
1920 | }
1921 | 
1922 | layer {
1923 | 	bottom: "res5a_branch2a"
1924 | 	top: "res5a_branch2a"
1925 | 	name: "scale5a_branch2a"
1926 | 	type: "Scale"
1927 | 	scale_param {
1928 | 		bias_term: true
1929 | 	}
1930 | }
1931 | 
1932 | layer {
1933 | 	bottom: "res5a_branch2a"
1934 | 	top: "res5a_branch2a"
1935 | 	name: "res5a_branch2a_relu"
1936 | 	type: "ReLU"
1937 | }
1938 | 
1939 | layer {
1940 | 	bottom: "res5a_branch2a"
1941 | 	top: "res5a_branch2b"
1942 | 	name: "res5a_branch2b"
1943 | 	type: "Convolution"
1944 | 	convolution_param {
1945 | 		num_output: 512
1946 | 		kernel_size: 3
1947 | 		pad: 1
1948 | 		stride: 1
1949 | 		bias_term: false
1950 | 	}
1951 | }
1952 | 
1953 | layer {
1954 | 	bottom: "res5a_branch2b"
1955 | 	top: "res5a_branch2b"
1956 | 	name: "bn5a_branch2b"
1957 | 	type: "BatchNorm"
1958 | 	batch_norm_param {
1959 | 		use_global_stats: true
1960 | 	}
1961 | }
1962 | 
1963 | layer {
1964 | 	bottom: "res5a_branch2b"
1965 | 	top: "res5a_branch2b"
1966 | 	name: "scale5a_branch2b"
1967 | 	type: "Scale"
1968 | 	scale_param {
1969 | 		bias_term: true
1970 | 	}
1971 | }
1972 | 
1973 | layer {
1974 | 	bottom: "res5a_branch2b"
1975 | 	top: "res5a_branch2b"
1976 | 	name: "res5a_branch2b_relu"
1977 | 	type: "ReLU"
1978 | }
1979 | 
1980 | layer {
1981 | 	bottom: "res5a_branch2b"
1982 | 	top: "res5a_branch2c"
1983 | 	name: "res5a_branch2c"
1984 | 	type: "Convolution"
1985 | 	convolution_param {
1986 | 		num_output: 2048
1987 | 		kernel_size: 1
1988 | 		pad: 0
1989 | 		stride: 1
1990 | 		bias_term: false
1991 | 	}
1992 | }
1993 | 
1994 | layer {
1995 | 	bottom: "res5a_branch2c"
1996 | 	top: "res5a_branch2c"
1997 | 	name: "bn5a_branch2c"
1998 | 	type: "BatchNorm"
1999 | 	batch_norm_param {
2000 | 		use_global_stats: true
2001 | 	}
2002 | }
2003 | 
2004 | layer {
2005 | 	bottom: "res5a_branch2c"
2006 | 	top: "res5a_branch2c"
2007 | 	name: "scale5a_branch2c"
2008 | 	type: "Scale"
2009 | 	scale_param {
2010 | 		bias_term: true
2011 | 	}
2012 | }
2013 | 
2014 | layer {
2015 | 	bottom: "res5a_branch1"
2016 | 	bottom: "res5a_branch2c"
2017 | 	top: "res5a"
2018 | 	name: "res5a"
2019 | 	type: "Eltwise"
2020 | }
2021 | 
2022 | layer {
2023 | 	bottom: "res5a"
2024 | 	top: "res5a"
2025 | 	name: "res5a_relu"
2026 | 	type: "ReLU"
2027 | }
2028 | 
2029 | layer {
2030 | 	bottom: "res5a"
2031 | 	top: "res5b_branch2a"
2032 | 	name: "res5b_branch2a"
2033 | 	type: "Convolution"
2034 | 	convolution_param {
2035 | 		num_output: 512
2036 | 		kernel_size: 1
2037 | 		pad: 0
2038 | 		stride: 1
2039 | 		bias_term: false
2040 | 	}
2041 | }
2042 | 
2043 | layer {
2044 | 	bottom: "res5b_branch2a"
2045 | 	top: "res5b_branch2a"
2046 | 	name: "bn5b_branch2a"
2047 | 	type: "BatchNorm"
2048 | 	batch_norm_param {
2049 | 		use_global_stats: true
2050 | 	}
2051 | }
2052 | 
2053 | layer {
2054 | 	bottom: "res5b_branch2a"
2055 | 	top: "res5b_branch2a"
2056 | 	name: "scale5b_branch2a"
2057 | 	type: "Scale"
2058 | 	scale_param {
2059 | 		bias_term: true
2060 | 	}
2061 | }
2062 | 
2063 | layer {
2064 | 	bottom: "res5b_branch2a"
2065 | 	top: "res5b_branch2a"
2066 | 	name: "res5b_branch2a_relu"
2067 | 	type: "ReLU"
2068 | }
2069 | 
2070 | layer {
2071 | 	bottom: "res5b_branch2a"
2072 | 	top: "res5b_branch2b"
2073 | 	name: "res5b_branch2b"
2074 | 	type: "Convolution"
2075 | 	convolution_param {
2076 | 		num_output: 512
2077 | 		kernel_size: 3
2078 | 		pad: 1
2079 | 		stride: 1
2080 | 		bias_term: false
2081 | 	}
2082 | }
2083 | 
2084 | layer {
2085 | 	bottom: "res5b_branch2b"
2086 | 	top: "res5b_branch2b"
2087 | 	name: "bn5b_branch2b"
2088 | 	type: "BatchNorm"
2089 | 	batch_norm_param {
2090 | 		use_global_stats: true
2091 | 	}
2092 | }
2093 | 
2094 | layer {
2095 | 	bottom: "res5b_branch2b"
2096 | 	top: "res5b_branch2b"
2097 | 	name: "scale5b_branch2b"
2098 | 	type: "Scale"
2099 | 	scale_param {
2100 | 		bias_term: true
2101 | 	}
2102 | }
2103 | 
2104 | layer {
2105 | 	bottom: "res5b_branch2b"
2106 | 	top: "res5b_branch2b"
2107 | 	name: "res5b_branch2b_relu"
2108 | 	type: "ReLU"
2109 | }
2110 | 
2111 | layer {
2112 | 	bottom: "res5b_branch2b"
2113 | 	top: "res5b_branch2c"
2114 | 	name: "res5b_branch2c"
2115 | 	type: "Convolution"
2116 | 	convolution_param {
2117 | 		num_output: 2048
2118 | 		kernel_size: 1
2119 | 		pad: 0
2120 | 		stride: 1
2121 | 		bias_term: false
2122 | 	}
2123 | }
2124 | 
2125 | layer {
2126 | 	bottom: "res5b_branch2c"
2127 | 	top: "res5b_branch2c"
2128 | 	name: "bn5b_branch2c"
2129 | 	type: "BatchNorm"
2130 | 	batch_norm_param {
2131 | 		use_global_stats: true
2132 | 	}
2133 | }
2134 | 
2135 | layer {
2136 | 	bottom: "res5b_branch2c"
2137 | 	top: "res5b_branch2c"
2138 | 	name: "scale5b_branch2c"
2139 | 	type: "Scale"
2140 | 	scale_param {
2141 | 		bias_term: true
2142 | 	}
2143 | }
2144 | 
2145 | layer {
2146 | 	bottom: "res5a"
2147 | 	bottom: "res5b_branch2c"
2148 | 	top: "res5b"
2149 | 	name: "res5b"
2150 | 	type: "Eltwise"
2151 | }
2152 | 
2153 | layer {
2154 | 	bottom: "res5b"
2155 | 	top: "res5b"
2156 | 	name: "res5b_relu"
2157 | 	type: "ReLU"
2158 | }
2159 | 
2160 | layer {
2161 | 	bottom: "res5b"
2162 | 	top: "res5c_branch2a"
2163 | 	name: "res5c_branch2a"
2164 | 	type: "Convolution"
2165 | 	convolution_param {
2166 | 		num_output: 512
2167 | 		kernel_size: 1
2168 | 		pad: 0
2169 | 		stride: 1
2170 | 		bias_term: false
2171 | 	}
2172 | }
2173 | 
2174 | layer {
2175 | 	bottom: "res5c_branch2a"
2176 | 	top: "res5c_branch2a"
2177 | 	name: "bn5c_branch2a"
2178 | 	type: "BatchNorm"
2179 | 	batch_norm_param {
2180 | 		use_global_stats: true
2181 | 	}
2182 | }
2183 | 
2184 | layer {
2185 | 	bottom: "res5c_branch2a"
2186 | 	top: "res5c_branch2a"
2187 | 	name: "scale5c_branch2a"
2188 | 	type: "Scale"
2189 | 	scale_param {
2190 | 		bias_term: true
2191 | 	}
2192 | }
2193 | 
2194 | layer {
2195 | 	bottom: "res5c_branch2a"
2196 | 	top: "res5c_branch2a"
2197 | 	name: "res5c_branch2a_relu"
2198 | 	type: "ReLU"
2199 | }
2200 | 
2201 | layer {
2202 | 	bottom: "res5c_branch2a"
2203 | 	top: "res5c_branch2b"
2204 | 	name: "res5c_branch2b"
2205 | 	type: "Convolution"
2206 | 	convolution_param {
2207 | 		num_output: 512
2208 | 		kernel_size: 3
2209 | 		pad: 1
2210 | 		stride: 1
2211 | 		bias_term: false
2212 | 	}
2213 | }
2214 | 
2215 | layer {
2216 | 	bottom: "res5c_branch2b"
2217 | 	top: "res5c_branch2b"
2218 | 	name: "bn5c_branch2b"
2219 | 	type: "BatchNorm"
2220 | 	batch_norm_param {
2221 | 		use_global_stats: true
2222 | 	}
2223 | }
2224 | 
2225 | layer {
2226 | 	bottom: "res5c_branch2b"
2227 | 	top: "res5c_branch2b"
2228 | 	name: "scale5c_branch2b"
2229 | 	type: "Scale"
2230 | 	scale_param {
2231 | 		bias_term: true
2232 | 	}
2233 | }
2234 | 
2235 | layer {
2236 | 	bottom: "res5c_branch2b"
2237 | 	top: "res5c_branch2b"
2238 | 	name: "res5c_branch2b_relu"
2239 | 	type: "ReLU"
2240 | }
2241 | 
2242 | layer {
2243 | 	bottom: "res5c_branch2b"
2244 | 	top: "res5c_branch2c"
2245 | 	name: "res5c_branch2c"
2246 | 	type: "Convolution"
2247 | 	convolution_param {
2248 | 		num_output: 2048
2249 | 		kernel_size: 1
2250 | 		pad: 0
2251 | 		stride: 1
2252 | 		bias_term: false
2253 | 	}
2254 | }
2255 | 
2256 | layer {
2257 | 	bottom: "res5c_branch2c"
2258 | 	top: "res5c_branch2c"
2259 | 	name: "bn5c_branch2c"
2260 | 	type: "BatchNorm"
2261 | 	batch_norm_param {
2262 | 		use_global_stats: true
2263 | 	}
2264 | }
2265 | 
2266 | layer {
2267 | 	bottom: "res5c_branch2c"
2268 | 	top: "res5c_branch2c"
2269 | 	name: "scale5c_branch2c"
2270 | 	type: "Scale"
2271 | 	scale_param {
2272 | 		bias_term: true
2273 | 	}
2274 | }
2275 | 
2276 | layer {
2277 | 	bottom: "res5b"
2278 | 	bottom: "res5c_branch2c"
2279 | 	top: "res5c"
2280 | 	name: "res5c"
2281 | 	type: "Eltwise"
2282 | }
2283 | 
2284 | layer {
2285 | 	bottom: "res5c"
2286 | 	top: "res5c"
2287 | 	name: "res5c_relu"
2288 | 	type: "ReLU"
2289 | }
2290 | 
2291 | layer {
2292 | 	bottom: "res5c"
2293 | 	top: "pool5"
2294 | 	name: "pool5"
2295 | 	type: "Pooling"
2296 | 	pooling_param {
2297 | 		kernel_size: 7
2298 | 		stride: 1
2299 | 		pool: AVE
2300 | 	}
2301 | }
2302 | layer {
2303 |         bottom: "pool5"
2304 |         top: "my-classifier"
2305 |         name: "my-classifier"
2306 |         type: "InnerProduct"
2307 |         inner_product_param {
2308 |                 num_output: 3
2309 |         }
2310 | }
2311 | layer {
2312 | 	bottom: "my-classifier"
2313 | 	top: "prob"
2314 | 	name: "prob"
2315 | 	type: "Softmax"
2316 | }
2317 | 
2318 | 


--------------------------------------------------------------------------------
/ResNet-50/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 512
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 501
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 10
26 | # save path
27 | snapshot_prefix: "resnet-50-cervix"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/ResNet-50/train_val.prototxt:
--------------------------------------------------------------------------------
   1 | name: "ResNet-50"
   2 | layer {
   3 |   name: "data"
   4 |   type: "Data"
   5 |   top: "data"
   6 |   top: "label"
   7 |   include {
   8 |     phase: TRAIN
   9 |   }
  10 |   transform_param {
  11 |     mirror: true
  12 |     crop_size: 224
  13 |     mean_file: "mean.binaryproto"
  14 |   }
  15 |   data_param {
  16 |     source: "train_lmdb"
  17 |     batch_size: 24
  18 |     backend: LMDB
  19 |   }
  20 | }
  21 | layer {
  22 |   name: "data"
  23 |   type: "Data"
  24 |   top: "data"
  25 |   top: "label"
  26 |   include {
  27 |     phase: TEST
  28 |   }
  29 |   transform_param {
  30 |     mirror: false
  31 |     crop_size: 224
  32 |     mean_file: "mean.binaryproto"
  33 |   }
  34 |   data_param {
  35 |     source: "val_lmdb"
  36 |     batch_size: 1
  37 |     backend: LMDB
  38 |   }
  39 | }
  40 | layer {
  41 | 	bottom: "data"
  42 | 	top: "conv1"
  43 | 	name: "conv1"
  44 | 	type: "Convolution"
  45 | 	convolution_param {
  46 | 		num_output: 64
  47 | 		kernel_size: 7
  48 | 		pad: 3
  49 | 		stride: 2
  50 | 	}
  51 | }
  52 | 
  53 | layer {
  54 | 	bottom: "conv1"
  55 | 	top: "conv1"
  56 | 	name: "bn_conv1"
  57 | 	type: "BatchNorm"
  58 | 	batch_norm_param {
  59 | 		use_global_stats: true
  60 | 	}
  61 | }
  62 | 
  63 | layer {
  64 | 	bottom: "conv1"
  65 | 	top: "conv1"
  66 | 	name: "scale_conv1"
  67 | 	type: "Scale"
  68 | 	scale_param {
  69 | 		bias_term: true
  70 | 	}
  71 | }
  72 | 
  73 | layer {
  74 | 	bottom: "conv1"
  75 | 	top: "conv1"
  76 | 	name: "conv1_relu"
  77 | 	type: "ReLU"
  78 | }
  79 | 
  80 | layer {
  81 | 	bottom: "conv1"
  82 | 	top: "pool1"
  83 | 	name: "pool1"
  84 | 	type: "Pooling"
  85 | 	pooling_param {
  86 | 		kernel_size: 3
  87 | 		stride: 2
  88 | 		pool: MAX
  89 | 	}
  90 | }
  91 | 
  92 | layer {
  93 | 	bottom: "pool1"
  94 | 	top: "res2a_branch1"
  95 | 	name: "res2a_branch1"
  96 | 	type: "Convolution"
  97 | 	convolution_param {
  98 | 		num_output: 256
  99 | 		kernel_size: 1
 100 | 		pad: 0
 101 | 		stride: 1
 102 | 		bias_term: false
 103 | 	}
 104 | }
 105 | 
 106 | layer {
 107 | 	bottom: "res2a_branch1"
 108 | 	top: "res2a_branch1"
 109 | 	name: "bn2a_branch1"
 110 | 	type: "BatchNorm"
 111 | 	batch_norm_param {
 112 | 		use_global_stats: true
 113 | 	}
 114 | }
 115 | 
 116 | layer {
 117 | 	bottom: "res2a_branch1"
 118 | 	top: "res2a_branch1"
 119 | 	name: "scale2a_branch1"
 120 | 	type: "Scale"
 121 | 	scale_param {
 122 | 		bias_term: true
 123 | 	}
 124 | }
 125 | 
 126 | layer {
 127 | 	bottom: "pool1"
 128 | 	top: "res2a_branch2a"
 129 | 	name: "res2a_branch2a"
 130 | 	type: "Convolution"
 131 | 	convolution_param {
 132 | 		num_output: 64
 133 | 		kernel_size: 1
 134 | 		pad: 0
 135 | 		stride: 1
 136 | 		bias_term: false
 137 | 	}
 138 | }
 139 | 
 140 | layer {
 141 | 	bottom: "res2a_branch2a"
 142 | 	top: "res2a_branch2a"
 143 | 	name: "bn2a_branch2a"
 144 | 	type: "BatchNorm"
 145 | 	batch_norm_param {
 146 | 		use_global_stats: true
 147 | 	}
 148 | }
 149 | 
 150 | layer {
 151 | 	bottom: "res2a_branch2a"
 152 | 	top: "res2a_branch2a"
 153 | 	name: "scale2a_branch2a"
 154 | 	type: "Scale"
 155 | 	scale_param {
 156 | 		bias_term: true
 157 | 	}
 158 | }
 159 | 
 160 | layer {
 161 | 	bottom: "res2a_branch2a"
 162 | 	top: "res2a_branch2a"
 163 | 	name: "res2a_branch2a_relu"
 164 | 	type: "ReLU"
 165 | }
 166 | 
 167 | layer {
 168 | 	bottom: "res2a_branch2a"
 169 | 	top: "res2a_branch2b"
 170 | 	name: "res2a_branch2b"
 171 | 	type: "Convolution"
 172 | 	convolution_param {
 173 | 		num_output: 64
 174 | 		kernel_size: 3
 175 | 		pad: 1
 176 | 		stride: 1
 177 | 		bias_term: false
 178 | 	}
 179 | }
 180 | 
 181 | layer {
 182 | 	bottom: "res2a_branch2b"
 183 | 	top: "res2a_branch2b"
 184 | 	name: "bn2a_branch2b"
 185 | 	type: "BatchNorm"
 186 | 	batch_norm_param {
 187 | 		use_global_stats: true
 188 | 	}
 189 | }
 190 | 
 191 | layer {
 192 | 	bottom: "res2a_branch2b"
 193 | 	top: "res2a_branch2b"
 194 | 	name: "scale2a_branch2b"
 195 | 	type: "Scale"
 196 | 	scale_param {
 197 | 		bias_term: true
 198 | 	}
 199 | }
 200 | 
 201 | layer {
 202 | 	bottom: "res2a_branch2b"
 203 | 	top: "res2a_branch2b"
 204 | 	name: "res2a_branch2b_relu"
 205 | 	type: "ReLU"
 206 | }
 207 | 
 208 | layer {
 209 | 	bottom: "res2a_branch2b"
 210 | 	top: "res2a_branch2c"
 211 | 	name: "res2a_branch2c"
 212 | 	type: "Convolution"
 213 | 	convolution_param {
 214 | 		num_output: 256
 215 | 		kernel_size: 1
 216 | 		pad: 0
 217 | 		stride: 1
 218 | 		bias_term: false
 219 | 	}
 220 | }
 221 | 
 222 | layer {
 223 | 	bottom: "res2a_branch2c"
 224 | 	top: "res2a_branch2c"
 225 | 	name: "bn2a_branch2c"
 226 | 	type: "BatchNorm"
 227 | 	batch_norm_param {
 228 | 		use_global_stats: true
 229 | 	}
 230 | }
 231 | 
 232 | layer {
 233 | 	bottom: "res2a_branch2c"
 234 | 	top: "res2a_branch2c"
 235 | 	name: "scale2a_branch2c"
 236 | 	type: "Scale"
 237 | 	scale_param {
 238 | 		bias_term: true
 239 | 	}
 240 | }
 241 | 
 242 | layer {
 243 | 	bottom: "res2a_branch1"
 244 | 	bottom: "res2a_branch2c"
 245 | 	top: "res2a"
 246 | 	name: "res2a"
 247 | 	type: "Eltwise"
 248 | }
 249 | 
 250 | layer {
 251 | 	bottom: "res2a"
 252 | 	top: "res2a"
 253 | 	name: "res2a_relu"
 254 | 	type: "ReLU"
 255 | }
 256 | 
 257 | layer {
 258 | 	bottom: "res2a"
 259 | 	top: "res2b_branch2a"
 260 | 	name: "res2b_branch2a"
 261 | 	type: "Convolution"
 262 | 	convolution_param {
 263 | 		num_output: 64
 264 | 		kernel_size: 1
 265 | 		pad: 0
 266 | 		stride: 1
 267 | 		bias_term: false
 268 | 	}
 269 | }
 270 | 
 271 | layer {
 272 | 	bottom: "res2b_branch2a"
 273 | 	top: "res2b_branch2a"
 274 | 	name: "bn2b_branch2a"
 275 | 	type: "BatchNorm"
 276 | 	batch_norm_param {
 277 | 		use_global_stats: true
 278 | 	}
 279 | }
 280 | 
 281 | layer {
 282 | 	bottom: "res2b_branch2a"
 283 | 	top: "res2b_branch2a"
 284 | 	name: "scale2b_branch2a"
 285 | 	type: "Scale"
 286 | 	scale_param {
 287 | 		bias_term: true
 288 | 	}
 289 | }
 290 | 
 291 | layer {
 292 | 	bottom: "res2b_branch2a"
 293 | 	top: "res2b_branch2a"
 294 | 	name: "res2b_branch2a_relu"
 295 | 	type: "ReLU"
 296 | }
 297 | 
 298 | layer {
 299 | 	bottom: "res2b_branch2a"
 300 | 	top: "res2b_branch2b"
 301 | 	name: "res2b_branch2b"
 302 | 	type: "Convolution"
 303 | 	convolution_param {
 304 | 		num_output: 64
 305 | 		kernel_size: 3
 306 | 		pad: 1
 307 | 		stride: 1
 308 | 		bias_term: false
 309 | 	}
 310 | }
 311 | 
 312 | layer {
 313 | 	bottom: "res2b_branch2b"
 314 | 	top: "res2b_branch2b"
 315 | 	name: "bn2b_branch2b"
 316 | 	type: "BatchNorm"
 317 | 	batch_norm_param {
 318 | 		use_global_stats: true
 319 | 	}
 320 | }
 321 | 
 322 | layer {
 323 | 	bottom: "res2b_branch2b"
 324 | 	top: "res2b_branch2b"
 325 | 	name: "scale2b_branch2b"
 326 | 	type: "Scale"
 327 | 	scale_param {
 328 | 		bias_term: true
 329 | 	}
 330 | }
 331 | 
 332 | layer {
 333 | 	bottom: "res2b_branch2b"
 334 | 	top: "res2b_branch2b"
 335 | 	name: "res2b_branch2b_relu"
 336 | 	type: "ReLU"
 337 | }
 338 | 
 339 | layer {
 340 | 	bottom: "res2b_branch2b"
 341 | 	top: "res2b_branch2c"
 342 | 	name: "res2b_branch2c"
 343 | 	type: "Convolution"
 344 | 	convolution_param {
 345 | 		num_output: 256
 346 | 		kernel_size: 1
 347 | 		pad: 0
 348 | 		stride: 1
 349 | 		bias_term: false
 350 | 	}
 351 | }
 352 | 
 353 | layer {
 354 | 	bottom: "res2b_branch2c"
 355 | 	top: "res2b_branch2c"
 356 | 	name: "bn2b_branch2c"
 357 | 	type: "BatchNorm"
 358 | 	batch_norm_param {
 359 | 		use_global_stats: true
 360 | 	}
 361 | }
 362 | 
 363 | layer {
 364 | 	bottom: "res2b_branch2c"
 365 | 	top: "res2b_branch2c"
 366 | 	name: "scale2b_branch2c"
 367 | 	type: "Scale"
 368 | 	scale_param {
 369 | 		bias_term: true
 370 | 	}
 371 | }
 372 | 
 373 | layer {
 374 | 	bottom: "res2a"
 375 | 	bottom: "res2b_branch2c"
 376 | 	top: "res2b"
 377 | 	name: "res2b"
 378 | 	type: "Eltwise"
 379 | }
 380 | 
 381 | layer {
 382 | 	bottom: "res2b"
 383 | 	top: "res2b"
 384 | 	name: "res2b_relu"
 385 | 	type: "ReLU"
 386 | }
 387 | 
 388 | layer {
 389 | 	bottom: "res2b"
 390 | 	top: "res2c_branch2a"
 391 | 	name: "res2c_branch2a"
 392 | 	type: "Convolution"
 393 | 	convolution_param {
 394 | 		num_output: 64
 395 | 		kernel_size: 1
 396 | 		pad: 0
 397 | 		stride: 1
 398 | 		bias_term: false
 399 | 	}
 400 | }
 401 | 
 402 | layer {
 403 | 	bottom: "res2c_branch2a"
 404 | 	top: "res2c_branch2a"
 405 | 	name: "bn2c_branch2a"
 406 | 	type: "BatchNorm"
 407 | 	batch_norm_param {
 408 | 		use_global_stats: true
 409 | 	}
 410 | }
 411 | 
 412 | layer {
 413 | 	bottom: "res2c_branch2a"
 414 | 	top: "res2c_branch2a"
 415 | 	name: "scale2c_branch2a"
 416 | 	type: "Scale"
 417 | 	scale_param {
 418 | 		bias_term: true
 419 | 	}
 420 | }
 421 | 
 422 | layer {
 423 | 	bottom: "res2c_branch2a"
 424 | 	top: "res2c_branch2a"
 425 | 	name: "res2c_branch2a_relu"
 426 | 	type: "ReLU"
 427 | }
 428 | 
 429 | layer {
 430 | 	bottom: "res2c_branch2a"
 431 | 	top: "res2c_branch2b"
 432 | 	name: "res2c_branch2b"
 433 | 	type: "Convolution"
 434 | 	convolution_param {
 435 | 		num_output: 64
 436 | 		kernel_size: 3
 437 | 		pad: 1
 438 | 		stride: 1
 439 | 		bias_term: false
 440 | 	}
 441 | }
 442 | 
 443 | layer {
 444 | 	bottom: "res2c_branch2b"
 445 | 	top: "res2c_branch2b"
 446 | 	name: "bn2c_branch2b"
 447 | 	type: "BatchNorm"
 448 | 	batch_norm_param {
 449 | 		use_global_stats: true
 450 | 	}
 451 | }
 452 | 
 453 | layer {
 454 | 	bottom: "res2c_branch2b"
 455 | 	top: "res2c_branch2b"
 456 | 	name: "scale2c_branch2b"
 457 | 	type: "Scale"
 458 | 	scale_param {
 459 | 		bias_term: true
 460 | 	}
 461 | }
 462 | 
 463 | layer {
 464 | 	bottom: "res2c_branch2b"
 465 | 	top: "res2c_branch2b"
 466 | 	name: "res2c_branch2b_relu"
 467 | 	type: "ReLU"
 468 | }
 469 | 
 470 | layer {
 471 | 	bottom: "res2c_branch2b"
 472 | 	top: "res2c_branch2c"
 473 | 	name: "res2c_branch2c"
 474 | 	type: "Convolution"
 475 | 	convolution_param {
 476 | 		num_output: 256
 477 | 		kernel_size: 1
 478 | 		pad: 0
 479 | 		stride: 1
 480 | 		bias_term: false
 481 | 	}
 482 | }
 483 | 
 484 | layer {
 485 | 	bottom: "res2c_branch2c"
 486 | 	top: "res2c_branch2c"
 487 | 	name: "bn2c_branch2c"
 488 | 	type: "BatchNorm"
 489 | 	batch_norm_param {
 490 | 		use_global_stats: true
 491 | 	}
 492 | }
 493 | 
 494 | layer {
 495 | 	bottom: "res2c_branch2c"
 496 | 	top: "res2c_branch2c"
 497 | 	name: "scale2c_branch2c"
 498 | 	type: "Scale"
 499 | 	scale_param {
 500 | 		bias_term: true
 501 | 	}
 502 | }
 503 | 
 504 | layer {
 505 | 	bottom: "res2b"
 506 | 	bottom: "res2c_branch2c"
 507 | 	top: "res2c"
 508 | 	name: "res2c"
 509 | 	type: "Eltwise"
 510 | }
 511 | 
 512 | layer {
 513 | 	bottom: "res2c"
 514 | 	top: "res2c"
 515 | 	name: "res2c_relu"
 516 | 	type: "ReLU"
 517 | }
 518 | 
 519 | layer {
 520 | 	bottom: "res2c"
 521 | 	top: "res3a_branch1"
 522 | 	name: "res3a_branch1"
 523 | 	type: "Convolution"
 524 | 	convolution_param {
 525 | 		num_output: 512
 526 | 		kernel_size: 1
 527 | 		pad: 0
 528 | 		stride: 2
 529 | 		bias_term: false
 530 | 	}
 531 | }
 532 | 
 533 | layer {
 534 | 	bottom: "res3a_branch1"
 535 | 	top: "res3a_branch1"
 536 | 	name: "bn3a_branch1"
 537 | 	type: "BatchNorm"
 538 | 	batch_norm_param {
 539 | 		use_global_stats: true
 540 | 	}
 541 | }
 542 | 
 543 | layer {
 544 | 	bottom: "res3a_branch1"
 545 | 	top: "res3a_branch1"
 546 | 	name: "scale3a_branch1"
 547 | 	type: "Scale"
 548 | 	scale_param {
 549 | 		bias_term: true
 550 | 	}
 551 | }
 552 | 
 553 | layer {
 554 | 	bottom: "res2c"
 555 | 	top: "res3a_branch2a"
 556 | 	name: "res3a_branch2a"
 557 | 	type: "Convolution"
 558 | 	convolution_param {
 559 | 		num_output: 128
 560 | 		kernel_size: 1
 561 | 		pad: 0
 562 | 		stride: 2
 563 | 		bias_term: false
 564 | 	}
 565 | }
 566 | 
 567 | layer {
 568 | 	bottom: "res3a_branch2a"
 569 | 	top: "res3a_branch2a"
 570 | 	name: "bn3a_branch2a"
 571 | 	type: "BatchNorm"
 572 | 	batch_norm_param {
 573 | 		use_global_stats: true
 574 | 	}
 575 | }
 576 | 
 577 | layer {
 578 | 	bottom: "res3a_branch2a"
 579 | 	top: "res3a_branch2a"
 580 | 	name: "scale3a_branch2a"
 581 | 	type: "Scale"
 582 | 	scale_param {
 583 | 		bias_term: true
 584 | 	}
 585 | }
 586 | 
 587 | layer {
 588 | 	bottom: "res3a_branch2a"
 589 | 	top: "res3a_branch2a"
 590 | 	name: "res3a_branch2a_relu"
 591 | 	type: "ReLU"
 592 | }
 593 | 
 594 | layer {
 595 | 	bottom: "res3a_branch2a"
 596 | 	top: "res3a_branch2b"
 597 | 	name: "res3a_branch2b"
 598 | 	type: "Convolution"
 599 | 	convolution_param {
 600 | 		num_output: 128
 601 | 		kernel_size: 3
 602 | 		pad: 1
 603 | 		stride: 1
 604 | 		bias_term: false
 605 | 	}
 606 | }
 607 | 
 608 | layer {
 609 | 	bottom: "res3a_branch2b"
 610 | 	top: "res3a_branch2b"
 611 | 	name: "bn3a_branch2b"
 612 | 	type: "BatchNorm"
 613 | 	batch_norm_param {
 614 | 		use_global_stats: true
 615 | 	}
 616 | }
 617 | 
 618 | layer {
 619 | 	bottom: "res3a_branch2b"
 620 | 	top: "res3a_branch2b"
 621 | 	name: "scale3a_branch2b"
 622 | 	type: "Scale"
 623 | 	scale_param {
 624 | 		bias_term: true
 625 | 	}
 626 | }
 627 | 
 628 | layer {
 629 | 	bottom: "res3a_branch2b"
 630 | 	top: "res3a_branch2b"
 631 | 	name: "res3a_branch2b_relu"
 632 | 	type: "ReLU"
 633 | }
 634 | 
 635 | layer {
 636 | 	bottom: "res3a_branch2b"
 637 | 	top: "res3a_branch2c"
 638 | 	name: "res3a_branch2c"
 639 | 	type: "Convolution"
 640 | 	convolution_param {
 641 | 		num_output: 512
 642 | 		kernel_size: 1
 643 | 		pad: 0
 644 | 		stride: 1
 645 | 		bias_term: false
 646 | 	}
 647 | }
 648 | 
 649 | layer {
 650 | 	bottom: "res3a_branch2c"
 651 | 	top: "res3a_branch2c"
 652 | 	name: "bn3a_branch2c"
 653 | 	type: "BatchNorm"
 654 | 	batch_norm_param {
 655 | 		use_global_stats: true
 656 | 	}
 657 | }
 658 | 
 659 | layer {
 660 | 	bottom: "res3a_branch2c"
 661 | 	top: "res3a_branch2c"
 662 | 	name: "scale3a_branch2c"
 663 | 	type: "Scale"
 664 | 	scale_param {
 665 | 		bias_term: true
 666 | 	}
 667 | }
 668 | 
 669 | layer {
 670 | 	bottom: "res3a_branch1"
 671 | 	bottom: "res3a_branch2c"
 672 | 	top: "res3a"
 673 | 	name: "res3a"
 674 | 	type: "Eltwise"
 675 | }
 676 | 
 677 | layer {
 678 | 	bottom: "res3a"
 679 | 	top: "res3a"
 680 | 	name: "res3a_relu"
 681 | 	type: "ReLU"
 682 | }
 683 | 
 684 | layer {
 685 | 	bottom: "res3a"
 686 | 	top: "res3b_branch2a"
 687 | 	name: "res3b_branch2a"
 688 | 	type: "Convolution"
 689 | 	convolution_param {
 690 | 		num_output: 128
 691 | 		kernel_size: 1
 692 | 		pad: 0
 693 | 		stride: 1
 694 | 		bias_term: false
 695 | 	}
 696 | }
 697 | 
 698 | layer {
 699 | 	bottom: "res3b_branch2a"
 700 | 	top: "res3b_branch2a"
 701 | 	name: "bn3b_branch2a"
 702 | 	type: "BatchNorm"
 703 | 	batch_norm_param {
 704 | 		use_global_stats: true
 705 | 	}
 706 | }
 707 | 
 708 | layer {
 709 | 	bottom: "res3b_branch2a"
 710 | 	top: "res3b_branch2a"
 711 | 	name: "scale3b_branch2a"
 712 | 	type: "Scale"
 713 | 	scale_param {
 714 | 		bias_term: true
 715 | 	}
 716 | }
 717 | 
 718 | layer {
 719 | 	bottom: "res3b_branch2a"
 720 | 	top: "res3b_branch2a"
 721 | 	name: "res3b_branch2a_relu"
 722 | 	type: "ReLU"
 723 | }
 724 | 
 725 | layer {
 726 | 	bottom: "res3b_branch2a"
 727 | 	top: "res3b_branch2b"
 728 | 	name: "res3b_branch2b"
 729 | 	type: "Convolution"
 730 | 	convolution_param {
 731 | 		num_output: 128
 732 | 		kernel_size: 3
 733 | 		pad: 1
 734 | 		stride: 1
 735 | 		bias_term: false
 736 | 	}
 737 | }
 738 | 
 739 | layer {
 740 | 	bottom: "res3b_branch2b"
 741 | 	top: "res3b_branch2b"
 742 | 	name: "bn3b_branch2b"
 743 | 	type: "BatchNorm"
 744 | 	batch_norm_param {
 745 | 		use_global_stats: true
 746 | 	}
 747 | }
 748 | 
 749 | layer {
 750 | 	bottom: "res3b_branch2b"
 751 | 	top: "res3b_branch2b"
 752 | 	name: "scale3b_branch2b"
 753 | 	type: "Scale"
 754 | 	scale_param {
 755 | 		bias_term: true
 756 | 	}
 757 | }
 758 | 
 759 | layer {
 760 | 	bottom: "res3b_branch2b"
 761 | 	top: "res3b_branch2b"
 762 | 	name: "res3b_branch2b_relu"
 763 | 	type: "ReLU"
 764 | }
 765 | 
 766 | layer {
 767 | 	bottom: "res3b_branch2b"
 768 | 	top: "res3b_branch2c"
 769 | 	name: "res3b_branch2c"
 770 | 	type: "Convolution"
 771 | 	convolution_param {
 772 | 		num_output: 512
 773 | 		kernel_size: 1
 774 | 		pad: 0
 775 | 		stride: 1
 776 | 		bias_term: false
 777 | 	}
 778 | }
 779 | 
 780 | layer {
 781 | 	bottom: "res3b_branch2c"
 782 | 	top: "res3b_branch2c"
 783 | 	name: "bn3b_branch2c"
 784 | 	type: "BatchNorm"
 785 | 	batch_norm_param {
 786 | 		use_global_stats: true
 787 | 	}
 788 | }
 789 | 
 790 | layer {
 791 | 	bottom: "res3b_branch2c"
 792 | 	top: "res3b_branch2c"
 793 | 	name: "scale3b_branch2c"
 794 | 	type: "Scale"
 795 | 	scale_param {
 796 | 		bias_term: true
 797 | 	}
 798 | }
 799 | 
 800 | layer {
 801 | 	bottom: "res3a"
 802 | 	bottom: "res3b_branch2c"
 803 | 	top: "res3b"
 804 | 	name: "res3b"
 805 | 	type: "Eltwise"
 806 | }
 807 | 
 808 | layer {
 809 | 	bottom: "res3b"
 810 | 	top: "res3b"
 811 | 	name: "res3b_relu"
 812 | 	type: "ReLU"
 813 | }
 814 | 
 815 | layer {
 816 | 	bottom: "res3b"
 817 | 	top: "res3c_branch2a"
 818 | 	name: "res3c_branch2a"
 819 | 	type: "Convolution"
 820 | 	convolution_param {
 821 | 		num_output: 128
 822 | 		kernel_size: 1
 823 | 		pad: 0
 824 | 		stride: 1
 825 | 		bias_term: false
 826 | 	}
 827 | }
 828 | 
 829 | layer {
 830 | 	bottom: "res3c_branch2a"
 831 | 	top: "res3c_branch2a"
 832 | 	name: "bn3c_branch2a"
 833 | 	type: "BatchNorm"
 834 | 	batch_norm_param {
 835 | 		use_global_stats: true
 836 | 	}
 837 | }
 838 | 
 839 | layer {
 840 | 	bottom: "res3c_branch2a"
 841 | 	top: "res3c_branch2a"
 842 | 	name: "scale3c_branch2a"
 843 | 	type: "Scale"
 844 | 	scale_param {
 845 | 		bias_term: true
 846 | 	}
 847 | }
 848 | 
 849 | layer {
 850 | 	bottom: "res3c_branch2a"
 851 | 	top: "res3c_branch2a"
 852 | 	name: "res3c_branch2a_relu"
 853 | 	type: "ReLU"
 854 | }
 855 | 
 856 | layer {
 857 | 	bottom: "res3c_branch2a"
 858 | 	top: "res3c_branch2b"
 859 | 	name: "res3c_branch2b"
 860 | 	type: "Convolution"
 861 | 	convolution_param {
 862 | 		num_output: 128
 863 | 		kernel_size: 3
 864 | 		pad: 1
 865 | 		stride: 1
 866 | 		bias_term: false
 867 | 	}
 868 | }
 869 | 
 870 | layer {
 871 | 	bottom: "res3c_branch2b"
 872 | 	top: "res3c_branch2b"
 873 | 	name: "bn3c_branch2b"
 874 | 	type: "BatchNorm"
 875 | 	batch_norm_param {
 876 | 		use_global_stats: true
 877 | 	}
 878 | }
 879 | 
 880 | layer {
 881 | 	bottom: "res3c_branch2b"
 882 | 	top: "res3c_branch2b"
 883 | 	name: "scale3c_branch2b"
 884 | 	type: "Scale"
 885 | 	scale_param {
 886 | 		bias_term: true
 887 | 	}
 888 | }
 889 | 
 890 | layer {
 891 | 	bottom: "res3c_branch2b"
 892 | 	top: "res3c_branch2b"
 893 | 	name: "res3c_branch2b_relu"
 894 | 	type: "ReLU"
 895 | }
 896 | 
 897 | layer {
 898 | 	bottom: "res3c_branch2b"
 899 | 	top: "res3c_branch2c"
 900 | 	name: "res3c_branch2c"
 901 | 	type: "Convolution"
 902 | 	convolution_param {
 903 | 		num_output: 512
 904 | 		kernel_size: 1
 905 | 		pad: 0
 906 | 		stride: 1
 907 | 		bias_term: false
 908 | 	}
 909 | }
 910 | 
 911 | layer {
 912 | 	bottom: "res3c_branch2c"
 913 | 	top: "res3c_branch2c"
 914 | 	name: "bn3c_branch2c"
 915 | 	type: "BatchNorm"
 916 | 	batch_norm_param {
 917 | 		use_global_stats: true
 918 | 	}
 919 | }
 920 | 
 921 | layer {
 922 | 	bottom: "res3c_branch2c"
 923 | 	top: "res3c_branch2c"
 924 | 	name: "scale3c_branch2c"
 925 | 	type: "Scale"
 926 | 	scale_param {
 927 | 		bias_term: true
 928 | 	}
 929 | }
 930 | 
 931 | layer {
 932 | 	bottom: "res3b"
 933 | 	bottom: "res3c_branch2c"
 934 | 	top: "res3c"
 935 | 	name: "res3c"
 936 | 	type: "Eltwise"
 937 | }
 938 | 
 939 | layer {
 940 | 	bottom: "res3c"
 941 | 	top: "res3c"
 942 | 	name: "res3c_relu"
 943 | 	type: "ReLU"
 944 | }
 945 | 
 946 | layer {
 947 | 	bottom: "res3c"
 948 | 	top: "res3d_branch2a"
 949 | 	name: "res3d_branch2a"
 950 | 	type: "Convolution"
 951 | 	convolution_param {
 952 | 		num_output: 128
 953 | 		kernel_size: 1
 954 | 		pad: 0
 955 | 		stride: 1
 956 | 		bias_term: false
 957 | 	}
 958 | }
 959 | 
 960 | layer {
 961 | 	bottom: "res3d_branch2a"
 962 | 	top: "res3d_branch2a"
 963 | 	name: "bn3d_branch2a"
 964 | 	type: "BatchNorm"
 965 | 	batch_norm_param {
 966 | 		use_global_stats: true
 967 | 	}
 968 | }
 969 | 
 970 | layer {
 971 | 	bottom: "res3d_branch2a"
 972 | 	top: "res3d_branch2a"
 973 | 	name: "scale3d_branch2a"
 974 | 	type: "Scale"
 975 | 	scale_param {
 976 | 		bias_term: true
 977 | 	}
 978 | }
 979 | 
 980 | layer {
 981 | 	bottom: "res3d_branch2a"
 982 | 	top: "res3d_branch2a"
 983 | 	name: "res3d_branch2a_relu"
 984 | 	type: "ReLU"
 985 | }
 986 | 
 987 | layer {
 988 | 	bottom: "res3d_branch2a"
 989 | 	top: "res3d_branch2b"
 990 | 	name: "res3d_branch2b"
 991 | 	type: "Convolution"
 992 | 	convolution_param {
 993 | 		num_output: 128
 994 | 		kernel_size: 3
 995 | 		pad: 1
 996 | 		stride: 1
 997 | 		bias_term: false
 998 | 	}
 999 | }
1000 | 
1001 | layer {
1002 | 	bottom: "res3d_branch2b"
1003 | 	top: "res3d_branch2b"
1004 | 	name: "bn3d_branch2b"
1005 | 	type: "BatchNorm"
1006 | 	batch_norm_param {
1007 | 		use_global_stats: true
1008 | 	}
1009 | }
1010 | 
1011 | layer {
1012 | 	bottom: "res3d_branch2b"
1013 | 	top: "res3d_branch2b"
1014 | 	name: "scale3d_branch2b"
1015 | 	type: "Scale"
1016 | 	scale_param {
1017 | 		bias_term: true
1018 | 	}
1019 | }
1020 | 
1021 | layer {
1022 | 	bottom: "res3d_branch2b"
1023 | 	top: "res3d_branch2b"
1024 | 	name: "res3d_branch2b_relu"
1025 | 	type: "ReLU"
1026 | }
1027 | 
1028 | layer {
1029 | 	bottom: "res3d_branch2b"
1030 | 	top: "res3d_branch2c"
1031 | 	name: "res3d_branch2c"
1032 | 	type: "Convolution"
1033 | 	convolution_param {
1034 | 		num_output: 512
1035 | 		kernel_size: 1
1036 | 		pad: 0
1037 | 		stride: 1
1038 | 		bias_term: false
1039 | 	}
1040 | }
1041 | 
1042 | layer {
1043 | 	bottom: "res3d_branch2c"
1044 | 	top: "res3d_branch2c"
1045 | 	name: "bn3d_branch2c"
1046 | 	type: "BatchNorm"
1047 | 	batch_norm_param {
1048 | 		use_global_stats: true
1049 | 	}
1050 | }
1051 | 
1052 | layer {
1053 | 	bottom: "res3d_branch2c"
1054 | 	top: "res3d_branch2c"
1055 | 	name: "scale3d_branch2c"
1056 | 	type: "Scale"
1057 | 	scale_param {
1058 | 		bias_term: true
1059 | 	}
1060 | }
1061 | 
1062 | layer {
1063 | 	bottom: "res3c"
1064 | 	bottom: "res3d_branch2c"
1065 | 	top: "res3d"
1066 | 	name: "res3d"
1067 | 	type: "Eltwise"
1068 | }
1069 | 
1070 | layer {
1071 | 	bottom: "res3d"
1072 | 	top: "res3d"
1073 | 	name: "res3d_relu"
1074 | 	type: "ReLU"
1075 | }
1076 | 
1077 | layer {
1078 | 	bottom: "res3d"
1079 | 	top: "res4a_branch1"
1080 | 	name: "res4a_branch1"
1081 | 	type: "Convolution"
1082 | 	convolution_param {
1083 | 		num_output: 1024
1084 | 		kernel_size: 1
1085 | 		pad: 0
1086 | 		stride: 2
1087 | 		bias_term: false
1088 | 	}
1089 | }
1090 | 
1091 | layer {
1092 | 	bottom: "res4a_branch1"
1093 | 	top: "res4a_branch1"
1094 | 	name: "bn4a_branch1"
1095 | 	type: "BatchNorm"
1096 | 	batch_norm_param {
1097 | 		use_global_stats: true
1098 | 	}
1099 | }
1100 | 
1101 | layer {
1102 | 	bottom: "res4a_branch1"
1103 | 	top: "res4a_branch1"
1104 | 	name: "scale4a_branch1"
1105 | 	type: "Scale"
1106 | 	scale_param {
1107 | 		bias_term: true
1108 | 	}
1109 | }
1110 | 
1111 | layer {
1112 | 	bottom: "res3d"
1113 | 	top: "res4a_branch2a"
1114 | 	name: "res4a_branch2a"
1115 | 	type: "Convolution"
1116 | 	convolution_param {
1117 | 		num_output: 256
1118 | 		kernel_size: 1
1119 | 		pad: 0
1120 | 		stride: 2
1121 | 		bias_term: false
1122 | 	}
1123 | }
1124 | 
1125 | layer {
1126 | 	bottom: "res4a_branch2a"
1127 | 	top: "res4a_branch2a"
1128 | 	name: "bn4a_branch2a"
1129 | 	type: "BatchNorm"
1130 | 	batch_norm_param {
1131 | 		use_global_stats: true
1132 | 	}
1133 | }
1134 | 
1135 | layer {
1136 | 	bottom: "res4a_branch2a"
1137 | 	top: "res4a_branch2a"
1138 | 	name: "scale4a_branch2a"
1139 | 	type: "Scale"
1140 | 	scale_param {
1141 | 		bias_term: true
1142 | 	}
1143 | }
1144 | 
1145 | layer {
1146 | 	bottom: "res4a_branch2a"
1147 | 	top: "res4a_branch2a"
1148 | 	name: "res4a_branch2a_relu"
1149 | 	type: "ReLU"
1150 | }
1151 | 
1152 | layer {
1153 | 	bottom: "res4a_branch2a"
1154 | 	top: "res4a_branch2b"
1155 | 	name: "res4a_branch2b"
1156 | 	type: "Convolution"
1157 | 	convolution_param {
1158 | 		num_output: 256
1159 | 		kernel_size: 3
1160 | 		pad: 1
1161 | 		stride: 1
1162 | 		bias_term: false
1163 | 	}
1164 | }
1165 | 
1166 | layer {
1167 | 	bottom: "res4a_branch2b"
1168 | 	top: "res4a_branch2b"
1169 | 	name: "bn4a_branch2b"
1170 | 	type: "BatchNorm"
1171 | 	batch_norm_param {
1172 | 		use_global_stats: true
1173 | 	}
1174 | }
1175 | 
1176 | layer {
1177 | 	bottom: "res4a_branch2b"
1178 | 	top: "res4a_branch2b"
1179 | 	name: "scale4a_branch2b"
1180 | 	type: "Scale"
1181 | 	scale_param {
1182 | 		bias_term: true
1183 | 	}
1184 | }
1185 | 
1186 | layer {
1187 | 	bottom: "res4a_branch2b"
1188 | 	top: "res4a_branch2b"
1189 | 	name: "res4a_branch2b_relu"
1190 | 	type: "ReLU"
1191 | }
1192 | 
1193 | layer {
1194 | 	bottom: "res4a_branch2b"
1195 | 	top: "res4a_branch2c"
1196 | 	name: "res4a_branch2c"
1197 | 	type: "Convolution"
1198 | 	convolution_param {
1199 | 		num_output: 1024
1200 | 		kernel_size: 1
1201 | 		pad: 0
1202 | 		stride: 1
1203 | 		bias_term: false
1204 | 	}
1205 | }
1206 | 
1207 | layer {
1208 | 	bottom: "res4a_branch2c"
1209 | 	top: "res4a_branch2c"
1210 | 	name: "bn4a_branch2c"
1211 | 	type: "BatchNorm"
1212 | 	batch_norm_param {
1213 | 		use_global_stats: true
1214 | 	}
1215 | }
1216 | 
1217 | layer {
1218 | 	bottom: "res4a_branch2c"
1219 | 	top: "res4a_branch2c"
1220 | 	name: "scale4a_branch2c"
1221 | 	type: "Scale"
1222 | 	scale_param {
1223 | 		bias_term: true
1224 | 	}
1225 | }
1226 | 
1227 | layer {
1228 | 	bottom: "res4a_branch1"
1229 | 	bottom: "res4a_branch2c"
1230 | 	top: "res4a"
1231 | 	name: "res4a"
1232 | 	type: "Eltwise"
1233 | }
1234 | 
1235 | layer {
1236 | 	bottom: "res4a"
1237 | 	top: "res4a"
1238 | 	name: "res4a_relu"
1239 | 	type: "ReLU"
1240 | }
1241 | 
1242 | layer {
1243 | 	bottom: "res4a"
1244 | 	top: "res4b_branch2a"
1245 | 	name: "res4b_branch2a"
1246 | 	type: "Convolution"
1247 | 	convolution_param {
1248 | 		num_output: 256
1249 | 		kernel_size: 1
1250 | 		pad: 0
1251 | 		stride: 1
1252 | 		bias_term: false
1253 | 	}
1254 | }
1255 | 
1256 | layer {
1257 | 	bottom: "res4b_branch2a"
1258 | 	top: "res4b_branch2a"
1259 | 	name: "bn4b_branch2a"
1260 | 	type: "BatchNorm"
1261 | 	batch_norm_param {
1262 | 		use_global_stats: true
1263 | 	}
1264 | }
1265 | 
1266 | layer {
1267 | 	bottom: "res4b_branch2a"
1268 | 	top: "res4b_branch2a"
1269 | 	name: "scale4b_branch2a"
1270 | 	type: "Scale"
1271 | 	scale_param {
1272 | 		bias_term: true
1273 | 	}
1274 | }
1275 | 
1276 | layer {
1277 | 	bottom: "res4b_branch2a"
1278 | 	top: "res4b_branch2a"
1279 | 	name: "res4b_branch2a_relu"
1280 | 	type: "ReLU"
1281 | }
1282 | 
1283 | layer {
1284 | 	bottom: "res4b_branch2a"
1285 | 	top: "res4b_branch2b"
1286 | 	name: "res4b_branch2b"
1287 | 	type: "Convolution"
1288 | 	convolution_param {
1289 | 		num_output: 256
1290 | 		kernel_size: 3
1291 | 		pad: 1
1292 | 		stride: 1
1293 | 		bias_term: false
1294 | 	}
1295 | }
1296 | 
1297 | layer {
1298 | 	bottom: "res4b_branch2b"
1299 | 	top: "res4b_branch2b"
1300 | 	name: "bn4b_branch2b"
1301 | 	type: "BatchNorm"
1302 | 	batch_norm_param {
1303 | 		use_global_stats: true
1304 | 	}
1305 | }
1306 | 
1307 | layer {
1308 | 	bottom: "res4b_branch2b"
1309 | 	top: "res4b_branch2b"
1310 | 	name: "scale4b_branch2b"
1311 | 	type: "Scale"
1312 | 	scale_param {
1313 | 		bias_term: true
1314 | 	}
1315 | }
1316 | 
1317 | layer {
1318 | 	bottom: "res4b_branch2b"
1319 | 	top: "res4b_branch2b"
1320 | 	name: "res4b_branch2b_relu"
1321 | 	type: "ReLU"
1322 | }
1323 | 
1324 | layer {
1325 | 	bottom: "res4b_branch2b"
1326 | 	top: "res4b_branch2c"
1327 | 	name: "res4b_branch2c"
1328 | 	type: "Convolution"
1329 | 	convolution_param {
1330 | 		num_output: 1024
1331 | 		kernel_size: 1
1332 | 		pad: 0
1333 | 		stride: 1
1334 | 		bias_term: false
1335 | 	}
1336 | }
1337 | 
1338 | layer {
1339 | 	bottom: "res4b_branch2c"
1340 | 	top: "res4b_branch2c"
1341 | 	name: "bn4b_branch2c"
1342 | 	type: "BatchNorm"
1343 | 	batch_norm_param {
1344 | 		use_global_stats: true
1345 | 	}
1346 | }
1347 | 
1348 | layer {
1349 | 	bottom: "res4b_branch2c"
1350 | 	top: "res4b_branch2c"
1351 | 	name: "scale4b_branch2c"
1352 | 	type: "Scale"
1353 | 	scale_param {
1354 | 		bias_term: true
1355 | 	}
1356 | }
1357 | 
1358 | layer {
1359 | 	bottom: "res4a"
1360 | 	bottom: "res4b_branch2c"
1361 | 	top: "res4b"
1362 | 	name: "res4b"
1363 | 	type: "Eltwise"
1364 | }
1365 | 
1366 | layer {
1367 | 	bottom: "res4b"
1368 | 	top: "res4b"
1369 | 	name: "res4b_relu"
1370 | 	type: "ReLU"
1371 | }
1372 | 
1373 | layer {
1374 | 	bottom: "res4b"
1375 | 	top: "res4c_branch2a"
1376 | 	name: "res4c_branch2a"
1377 | 	type: "Convolution"
1378 | 	convolution_param {
1379 | 		num_output: 256
1380 | 		kernel_size: 1
1381 | 		pad: 0
1382 | 		stride: 1
1383 | 		bias_term: false
1384 | 	}
1385 | }
1386 | 
1387 | layer {
1388 | 	bottom: "res4c_branch2a"
1389 | 	top: "res4c_branch2a"
1390 | 	name: "bn4c_branch2a"
1391 | 	type: "BatchNorm"
1392 | 	batch_norm_param {
1393 | 		use_global_stats: true
1394 | 	}
1395 | }
1396 | 
1397 | layer {
1398 | 	bottom: "res4c_branch2a"
1399 | 	top: "res4c_branch2a"
1400 | 	name: "scale4c_branch2a"
1401 | 	type: "Scale"
1402 | 	scale_param {
1403 | 		bias_term: true
1404 | 	}
1405 | }
1406 | 
1407 | layer {
1408 | 	bottom: "res4c_branch2a"
1409 | 	top: "res4c_branch2a"
1410 | 	name: "res4c_branch2a_relu"
1411 | 	type: "ReLU"
1412 | }
1413 | 
1414 | layer {
1415 | 	bottom: "res4c_branch2a"
1416 | 	top: "res4c_branch2b"
1417 | 	name: "res4c_branch2b"
1418 | 	type: "Convolution"
1419 | 	convolution_param {
1420 | 		num_output: 256
1421 | 		kernel_size: 3
1422 | 		pad: 1
1423 | 		stride: 1
1424 | 		bias_term: false
1425 | 	}
1426 | }
1427 | 
1428 | layer {
1429 | 	bottom: "res4c_branch2b"
1430 | 	top: "res4c_branch2b"
1431 | 	name: "bn4c_branch2b"
1432 | 	type: "BatchNorm"
1433 | 	batch_norm_param {
1434 | 		use_global_stats: true
1435 | 	}
1436 | }
1437 | 
1438 | layer {
1439 | 	bottom: "res4c_branch2b"
1440 | 	top: "res4c_branch2b"
1441 | 	name: "scale4c_branch2b"
1442 | 	type: "Scale"
1443 | 	scale_param {
1444 | 		bias_term: true
1445 | 	}
1446 | }
1447 | 
1448 | layer {
1449 | 	bottom: "res4c_branch2b"
1450 | 	top: "res4c_branch2b"
1451 | 	name: "res4c_branch2b_relu"
1452 | 	type: "ReLU"
1453 | }
1454 | 
1455 | layer {
1456 | 	bottom: "res4c_branch2b"
1457 | 	top: "res4c_branch2c"
1458 | 	name: "res4c_branch2c"
1459 | 	type: "Convolution"
1460 | 	convolution_param {
1461 | 		num_output: 1024
1462 | 		kernel_size: 1
1463 | 		pad: 0
1464 | 		stride: 1
1465 | 		bias_term: false
1466 | 	}
1467 | }
1468 | 
1469 | layer {
1470 | 	bottom: "res4c_branch2c"
1471 | 	top: "res4c_branch2c"
1472 | 	name: "bn4c_branch2c"
1473 | 	type: "BatchNorm"
1474 | 	batch_norm_param {
1475 | 		use_global_stats: true
1476 | 	}
1477 | }
1478 | 
1479 | layer {
1480 | 	bottom: "res4c_branch2c"
1481 | 	top: "res4c_branch2c"
1482 | 	name: "scale4c_branch2c"
1483 | 	type: "Scale"
1484 | 	scale_param {
1485 | 		bias_term: true
1486 | 	}
1487 | }
1488 | 
1489 | layer {
1490 | 	bottom: "res4b"
1491 | 	bottom: "res4c_branch2c"
1492 | 	top: "res4c"
1493 | 	name: "res4c"
1494 | 	type: "Eltwise"
1495 | }
1496 | 
1497 | layer {
1498 | 	bottom: "res4c"
1499 | 	top: "res4c"
1500 | 	name: "res4c_relu"
1501 | 	type: "ReLU"
1502 | }
1503 | 
1504 | layer {
1505 | 	bottom: "res4c"
1506 | 	top: "res4d_branch2a"
1507 | 	name: "res4d_branch2a"
1508 | 	type: "Convolution"
1509 | 	convolution_param {
1510 | 		num_output: 256
1511 | 		kernel_size: 1
1512 | 		pad: 0
1513 | 		stride: 1
1514 | 		bias_term: false
1515 | 	}
1516 | }
1517 | 
1518 | layer {
1519 | 	bottom: "res4d_branch2a"
1520 | 	top: "res4d_branch2a"
1521 | 	name: "bn4d_branch2a"
1522 | 	type: "BatchNorm"
1523 | 	batch_norm_param {
1524 | 		use_global_stats: true
1525 | 	}
1526 | }
1527 | 
1528 | layer {
1529 | 	bottom: "res4d_branch2a"
1530 | 	top: "res4d_branch2a"
1531 | 	name: "scale4d_branch2a"
1532 | 	type: "Scale"
1533 | 	scale_param {
1534 | 		bias_term: true
1535 | 	}
1536 | }
1537 | 
1538 | layer {
1539 | 	bottom: "res4d_branch2a"
1540 | 	top: "res4d_branch2a"
1541 | 	name: "res4d_branch2a_relu"
1542 | 	type: "ReLU"
1543 | }
1544 | 
1545 | layer {
1546 | 	bottom: "res4d_branch2a"
1547 | 	top: "res4d_branch2b"
1548 | 	name: "res4d_branch2b"
1549 | 	type: "Convolution"
1550 | 	convolution_param {
1551 | 		num_output: 256
1552 | 		kernel_size: 3
1553 | 		pad: 1
1554 | 		stride: 1
1555 | 		bias_term: false
1556 | 	}
1557 | }
1558 | 
1559 | layer {
1560 | 	bottom: "res4d_branch2b"
1561 | 	top: "res4d_branch2b"
1562 | 	name: "bn4d_branch2b"
1563 | 	type: "BatchNorm"
1564 | 	batch_norm_param {
1565 | 		use_global_stats: true
1566 | 	}
1567 | }
1568 | 
1569 | layer {
1570 | 	bottom: "res4d_branch2b"
1571 | 	top: "res4d_branch2b"
1572 | 	name: "scale4d_branch2b"
1573 | 	type: "Scale"
1574 | 	scale_param {
1575 | 		bias_term: true
1576 | 	}
1577 | }
1578 | 
1579 | layer {
1580 | 	bottom: "res4d_branch2b"
1581 | 	top: "res4d_branch2b"
1582 | 	name: "res4d_branch2b_relu"
1583 | 	type: "ReLU"
1584 | }
1585 | 
1586 | layer {
1587 | 	bottom: "res4d_branch2b"
1588 | 	top: "res4d_branch2c"
1589 | 	name: "res4d_branch2c"
1590 | 	type: "Convolution"
1591 | 	convolution_param {
1592 | 		num_output: 1024
1593 | 		kernel_size: 1
1594 | 		pad: 0
1595 | 		stride: 1
1596 | 		bias_term: false
1597 | 	}
1598 | }
1599 | 
1600 | layer {
1601 | 	bottom: "res4d_branch2c"
1602 | 	top: "res4d_branch2c"
1603 | 	name: "bn4d_branch2c"
1604 | 	type: "BatchNorm"
1605 | 	batch_norm_param {
1606 | 		use_global_stats: true
1607 | 	}
1608 | }
1609 | 
1610 | layer {
1611 | 	bottom: "res4d_branch2c"
1612 | 	top: "res4d_branch2c"
1613 | 	name: "scale4d_branch2c"
1614 | 	type: "Scale"
1615 | 	scale_param {
1616 | 		bias_term: true
1617 | 	}
1618 | }
1619 | 
1620 | layer {
1621 | 	bottom: "res4c"
1622 | 	bottom: "res4d_branch2c"
1623 | 	top: "res4d"
1624 | 	name: "res4d"
1625 | 	type: "Eltwise"
1626 | }
1627 | 
1628 | layer {
1629 | 	bottom: "res4d"
1630 | 	top: "res4d"
1631 | 	name: "res4d_relu"
1632 | 	type: "ReLU"
1633 | }
1634 | 
1635 | layer {
1636 | 	bottom: "res4d"
1637 | 	top: "res4e_branch2a"
1638 | 	name: "res4e_branch2a"
1639 | 	type: "Convolution"
1640 | 	convolution_param {
1641 | 		num_output: 256
1642 | 		kernel_size: 1
1643 | 		pad: 0
1644 | 		stride: 1
1645 | 		bias_term: false
1646 | 	}
1647 | }
1648 | 
1649 | layer {
1650 | 	bottom: "res4e_branch2a"
1651 | 	top: "res4e_branch2a"
1652 | 	name: "bn4e_branch2a"
1653 | 	type: "BatchNorm"
1654 | 	batch_norm_param {
1655 | 		use_global_stats: true
1656 | 	}
1657 | }
1658 | 
1659 | layer {
1660 | 	bottom: "res4e_branch2a"
1661 | 	top: "res4e_branch2a"
1662 | 	name: "scale4e_branch2a"
1663 | 	type: "Scale"
1664 | 	scale_param {
1665 | 		bias_term: true
1666 | 	}
1667 | }
1668 | 
1669 | layer {
1670 | 	bottom: "res4e_branch2a"
1671 | 	top: "res4e_branch2a"
1672 | 	name: "res4e_branch2a_relu"
1673 | 	type: "ReLU"
1674 | }
1675 | 
1676 | layer {
1677 | 	bottom: "res4e_branch2a"
1678 | 	top: "res4e_branch2b"
1679 | 	name: "res4e_branch2b"
1680 | 	type: "Convolution"
1681 | 	convolution_param {
1682 | 		num_output: 256
1683 | 		kernel_size: 3
1684 | 		pad: 1
1685 | 		stride: 1
1686 | 		bias_term: false
1687 | 	}
1688 | }
1689 | 
1690 | layer {
1691 | 	bottom: "res4e_branch2b"
1692 | 	top: "res4e_branch2b"
1693 | 	name: "bn4e_branch2b"
1694 | 	type: "BatchNorm"
1695 | 	batch_norm_param {
1696 | 		use_global_stats: true
1697 | 	}
1698 | }
1699 | 
1700 | layer {
1701 | 	bottom: "res4e_branch2b"
1702 | 	top: "res4e_branch2b"
1703 | 	name: "scale4e_branch2b"
1704 | 	type: "Scale"
1705 | 	scale_param {
1706 | 		bias_term: true
1707 | 	}
1708 | }
1709 | 
1710 | layer {
1711 | 	bottom: "res4e_branch2b"
1712 | 	top: "res4e_branch2b"
1713 | 	name: "res4e_branch2b_relu"
1714 | 	type: "ReLU"
1715 | }
1716 | 
1717 | layer {
1718 | 	bottom: "res4e_branch2b"
1719 | 	top: "res4e_branch2c"
1720 | 	name: "res4e_branch2c"
1721 | 	type: "Convolution"
1722 | 	convolution_param {
1723 | 		num_output: 1024
1724 | 		kernel_size: 1
1725 | 		pad: 0
1726 | 		stride: 1
1727 | 		bias_term: false
1728 | 	}
1729 | }
1730 | 
1731 | layer {
1732 | 	bottom: "res4e_branch2c"
1733 | 	top: "res4e_branch2c"
1734 | 	name: "bn4e_branch2c"
1735 | 	type: "BatchNorm"
1736 | 	batch_norm_param {
1737 | 		use_global_stats: true
1738 | 	}
1739 | }
1740 | 
1741 | layer {
1742 | 	bottom: "res4e_branch2c"
1743 | 	top: "res4e_branch2c"
1744 | 	name: "scale4e_branch2c"
1745 | 	type: "Scale"
1746 | 	scale_param {
1747 | 		bias_term: true
1748 | 	}
1749 | }
1750 | 
1751 | layer {
1752 | 	bottom: "res4d"
1753 | 	bottom: "res4e_branch2c"
1754 | 	top: "res4e"
1755 | 	name: "res4e"
1756 | 	type: "Eltwise"
1757 | }
1758 | 
1759 | layer {
1760 | 	bottom: "res4e"
1761 | 	top: "res4e"
1762 | 	name: "res4e_relu"
1763 | 	type: "ReLU"
1764 | }
1765 | 
1766 | layer {
1767 | 	bottom: "res4e"
1768 | 	top: "res4f_branch2a"
1769 | 	name: "res4f_branch2a"
1770 | 	type: "Convolution"
1771 | 	convolution_param {
1772 | 		num_output: 256
1773 | 		kernel_size: 1
1774 | 		pad: 0
1775 | 		stride: 1
1776 | 		bias_term: false
1777 | 	}
1778 | }
1779 | 
1780 | layer {
1781 | 	bottom: "res4f_branch2a"
1782 | 	top: "res4f_branch2a"
1783 | 	name: "bn4f_branch2a"
1784 | 	type: "BatchNorm"
1785 | 	batch_norm_param {
1786 | 		use_global_stats: true
1787 | 	}
1788 | }
1789 | 
1790 | layer {
1791 | 	bottom: "res4f_branch2a"
1792 | 	top: "res4f_branch2a"
1793 | 	name: "scale4f_branch2a"
1794 | 	type: "Scale"
1795 | 	scale_param {
1796 | 		bias_term: true
1797 | 	}
1798 | }
1799 | 
1800 | layer {
1801 | 	bottom: "res4f_branch2a"
1802 | 	top: "res4f_branch2a"
1803 | 	name: "res4f_branch2a_relu"
1804 | 	type: "ReLU"
1805 | }
1806 | 
1807 | layer {
1808 | 	bottom: "res4f_branch2a"
1809 | 	top: "res4f_branch2b"
1810 | 	name: "res4f_branch2b"
1811 | 	type: "Convolution"
1812 | 	convolution_param {
1813 | 		num_output: 256
1814 | 		kernel_size: 3
1815 | 		pad: 1
1816 | 		stride: 1
1817 | 		bias_term: false
1818 | 	}
1819 | }
1820 | 
1821 | layer {
1822 | 	bottom: "res4f_branch2b"
1823 | 	top: "res4f_branch2b"
1824 | 	name: "bn4f_branch2b"
1825 | 	type: "BatchNorm"
1826 | 	batch_norm_param {
1827 | 		use_global_stats: true
1828 | 	}
1829 | }
1830 | 
1831 | layer {
1832 | 	bottom: "res4f_branch2b"
1833 | 	top: "res4f_branch2b"
1834 | 	name: "scale4f_branch2b"
1835 | 	type: "Scale"
1836 | 	scale_param {
1837 | 		bias_term: true
1838 | 	}
1839 | }
1840 | 
1841 | layer {
1842 | 	bottom: "res4f_branch2b"
1843 | 	top: "res4f_branch2b"
1844 | 	name: "res4f_branch2b_relu"
1845 | 	type: "ReLU"
1846 | }
1847 | 
1848 | layer {
1849 | 	bottom: "res4f_branch2b"
1850 | 	top: "res4f_branch2c"
1851 | 	name: "res4f_branch2c"
1852 | 	type: "Convolution"
1853 | 	convolution_param {
1854 | 		num_output: 1024
1855 | 		kernel_size: 1
1856 | 		pad: 0
1857 | 		stride: 1
1858 | 		bias_term: false
1859 | 	}
1860 | }
1861 | 
1862 | layer {
1863 | 	bottom: "res4f_branch2c"
1864 | 	top: "res4f_branch2c"
1865 | 	name: "bn4f_branch2c"
1866 | 	type: "BatchNorm"
1867 | 	batch_norm_param {
1868 | 		use_global_stats: true
1869 | 	}
1870 | }
1871 | 
1872 | layer {
1873 | 	bottom: "res4f_branch2c"
1874 | 	top: "res4f_branch2c"
1875 | 	name: "scale4f_branch2c"
1876 | 	type: "Scale"
1877 | 	scale_param {
1878 | 		bias_term: true
1879 | 	}
1880 | }
1881 | 
1882 | layer {
1883 | 	bottom: "res4e"
1884 | 	bottom: "res4f_branch2c"
1885 | 	top: "res4f"
1886 | 	name: "res4f"
1887 | 	type: "Eltwise"
1888 | }
1889 | 
1890 | layer {
1891 | 	bottom: "res4f"
1892 | 	top: "res4f"
1893 | 	name: "res4f_relu"
1894 | 	type: "ReLU"
1895 | }
1896 | 
1897 | layer {
1898 | 	bottom: "res4f"
1899 | 	top: "res5a_branch1"
1900 | 	name: "res5a_branch1"
1901 | 	type: "Convolution"
1902 | 	convolution_param {
1903 | 		num_output: 2048
1904 | 		kernel_size: 1
1905 | 		pad: 0
1906 | 		stride: 2
1907 | 		bias_term: false
1908 | 	}
1909 | }
1910 | 
1911 | layer {
1912 | 	bottom: "res5a_branch1"
1913 | 	top: "res5a_branch1"
1914 | 	name: "bn5a_branch1"
1915 | 	type: "BatchNorm"
1916 | 	batch_norm_param {
1917 | 		use_global_stats: true
1918 | 	}
1919 | }
1920 | 
1921 | layer {
1922 | 	bottom: "res5a_branch1"
1923 | 	top: "res5a_branch1"
1924 | 	name: "scale5a_branch1"
1925 | 	type: "Scale"
1926 | 	scale_param {
1927 | 		bias_term: true
1928 | 	}
1929 | }
1930 | 
1931 | layer {
1932 | 	bottom: "res4f"
1933 | 	top: "res5a_branch2a"
1934 | 	name: "res5a_branch2a"
1935 | 	type: "Convolution"
1936 | 	convolution_param {
1937 | 		num_output: 512
1938 | 		kernel_size: 1
1939 | 		pad: 0
1940 | 		stride: 2
1941 | 		bias_term: false
1942 | 	}
1943 | }
1944 | 
1945 | layer {
1946 | 	bottom: "res5a_branch2a"
1947 | 	top: "res5a_branch2a"
1948 | 	name: "bn5a_branch2a"
1949 | 	type: "BatchNorm"
1950 | 	batch_norm_param {
1951 | 		use_global_stats: true
1952 | 	}
1953 | }
1954 | 
1955 | layer {
1956 | 	bottom: "res5a_branch2a"
1957 | 	top: "res5a_branch2a"
1958 | 	name: "scale5a_branch2a"
1959 | 	type: "Scale"
1960 | 	scale_param {
1961 | 		bias_term: true
1962 | 	}
1963 | }
1964 | 
1965 | layer {
1966 | 	bottom: "res5a_branch2a"
1967 | 	top: "res5a_branch2a"
1968 | 	name: "res5a_branch2a_relu"
1969 | 	type: "ReLU"
1970 | }
1971 | 
1972 | layer {
1973 | 	bottom: "res5a_branch2a"
1974 | 	top: "res5a_branch2b"
1975 | 	name: "res5a_branch2b"
1976 | 	type: "Convolution"
1977 | 	convolution_param {
1978 | 		num_output: 512
1979 | 		kernel_size: 3
1980 | 		pad: 1
1981 | 		stride: 1
1982 | 		bias_term: false
1983 | 	}
1984 | }
1985 | 
1986 | layer {
1987 | 	bottom: "res5a_branch2b"
1988 | 	top: "res5a_branch2b"
1989 | 	name: "bn5a_branch2b"
1990 | 	type: "BatchNorm"
1991 | 	batch_norm_param {
1992 | 		use_global_stats: true
1993 | 	}
1994 | }
1995 | 
1996 | layer {
1997 | 	bottom: "res5a_branch2b"
1998 | 	top: "res5a_branch2b"
1999 | 	name: "scale5a_branch2b"
2000 | 	type: "Scale"
2001 | 	scale_param {
2002 | 		bias_term: true
2003 | 	}
2004 | }
2005 | 
2006 | layer {
2007 | 	bottom: "res5a_branch2b"
2008 | 	top: "res5a_branch2b"
2009 | 	name: "res5a_branch2b_relu"
2010 | 	type: "ReLU"
2011 | }
2012 | 
2013 | layer {
2014 | 	bottom: "res5a_branch2b"
2015 | 	top: "res5a_branch2c"
2016 | 	name: "res5a_branch2c"
2017 | 	type: "Convolution"
2018 | 	convolution_param {
2019 | 		num_output: 2048
2020 | 		kernel_size: 1
2021 | 		pad: 0
2022 | 		stride: 1
2023 | 		bias_term: false
2024 | 	}
2025 | }
2026 | 
2027 | layer {
2028 | 	bottom: "res5a_branch2c"
2029 | 	top: "res5a_branch2c"
2030 | 	name: "bn5a_branch2c"
2031 | 	type: "BatchNorm"
2032 | 	batch_norm_param {
2033 | 		use_global_stats: true
2034 | 	}
2035 | }
2036 | 
2037 | layer {
2038 | 	bottom: "res5a_branch2c"
2039 | 	top: "res5a_branch2c"
2040 | 	name: "scale5a_branch2c"
2041 | 	type: "Scale"
2042 | 	scale_param {
2043 | 		bias_term: true
2044 | 	}
2045 | }
2046 | 
2047 | layer {
2048 | 	bottom: "res5a_branch1"
2049 | 	bottom: "res5a_branch2c"
2050 | 	top: "res5a"
2051 | 	name: "res5a"
2052 | 	type: "Eltwise"
2053 | }
2054 | 
2055 | layer {
2056 | 	bottom: "res5a"
2057 | 	top: "res5a"
2058 | 	name: "res5a_relu"
2059 | 	type: "ReLU"
2060 | }
2061 | 
2062 | layer {
2063 | 	bottom: "res5a"
2064 | 	top: "res5b_branch2a"
2065 | 	name: "res5b_branch2a"
2066 | 	type: "Convolution"
2067 | 	convolution_param {
2068 | 		num_output: 512
2069 | 		kernel_size: 1
2070 | 		pad: 0
2071 | 		stride: 1
2072 | 		bias_term: false
2073 | 	}
2074 | }
2075 | 
2076 | layer {
2077 | 	bottom: "res5b_branch2a"
2078 | 	top: "res5b_branch2a"
2079 | 	name: "bn5b_branch2a"
2080 | 	type: "BatchNorm"
2081 | 	batch_norm_param {
2082 | 		use_global_stats: true
2083 | 	}
2084 | }
2085 | 
2086 | layer {
2087 | 	bottom: "res5b_branch2a"
2088 | 	top: "res5b_branch2a"
2089 | 	name: "scale5b_branch2a"
2090 | 	type: "Scale"
2091 | 	scale_param {
2092 | 		bias_term: true
2093 | 	}
2094 | }
2095 | 
2096 | layer {
2097 | 	bottom: "res5b_branch2a"
2098 | 	top: "res5b_branch2a"
2099 | 	name: "res5b_branch2a_relu"
2100 | 	type: "ReLU"
2101 | }
2102 | 
2103 | layer {
2104 | 	bottom: "res5b_branch2a"
2105 | 	top: "res5b_branch2b"
2106 | 	name: "res5b_branch2b"
2107 | 	type: "Convolution"
2108 | 	convolution_param {
2109 | 		num_output: 512
2110 | 		kernel_size: 3
2111 | 		pad: 1
2112 | 		stride: 1
2113 | 		bias_term: false
2114 | 	}
2115 | }
2116 | 
2117 | layer {
2118 | 	bottom: "res5b_branch2b"
2119 | 	top: "res5b_branch2b"
2120 | 	name: "bn5b_branch2b"
2121 | 	type: "BatchNorm"
2122 | 	batch_norm_param {
2123 | 		use_global_stats: true
2124 | 	}
2125 | }
2126 | 
2127 | layer {
2128 | 	bottom: "res5b_branch2b"
2129 | 	top: "res5b_branch2b"
2130 | 	name: "scale5b_branch2b"
2131 | 	type: "Scale"
2132 | 	scale_param {
2133 | 		bias_term: true
2134 | 	}
2135 | }
2136 | 
2137 | layer {
2138 | 	bottom: "res5b_branch2b"
2139 | 	top: "res5b_branch2b"
2140 | 	name: "res5b_branch2b_relu"
2141 | 	type: "ReLU"
2142 | }
2143 | 
2144 | layer {
2145 | 	bottom: "res5b_branch2b"
2146 | 	top: "res5b_branch2c"
2147 | 	name: "res5b_branch2c"
2148 | 	type: "Convolution"
2149 | 	convolution_param {
2150 | 		num_output: 2048
2151 | 		kernel_size: 1
2152 | 		pad: 0
2153 | 		stride: 1
2154 | 		bias_term: false
2155 | 	}
2156 | }
2157 | 
2158 | layer {
2159 | 	bottom: "res5b_branch2c"
2160 | 	top: "res5b_branch2c"
2161 | 	name: "bn5b_branch2c"
2162 | 	type: "BatchNorm"
2163 | 	batch_norm_param {
2164 | 		use_global_stats: true
2165 | 	}
2166 | }
2167 | 
2168 | layer {
2169 | 	bottom: "res5b_branch2c"
2170 | 	top: "res5b_branch2c"
2171 | 	name: "scale5b_branch2c"
2172 | 	type: "Scale"
2173 | 	scale_param {
2174 | 		bias_term: true
2175 | 	}
2176 | }
2177 | 
2178 | layer {
2179 | 	bottom: "res5a"
2180 | 	bottom: "res5b_branch2c"
2181 | 	top: "res5b"
2182 | 	name: "res5b"
2183 | 	type: "Eltwise"
2184 | }
2185 | 
2186 | layer {
2187 | 	bottom: "res5b"
2188 | 	top: "res5b"
2189 | 	name: "res5b_relu"
2190 | 	type: "ReLU"
2191 | }
2192 | 
2193 | layer {
2194 | 	bottom: "res5b"
2195 | 	top: "res5c_branch2a"
2196 | 	name: "res5c_branch2a"
2197 | 	type: "Convolution"
2198 | 	convolution_param {
2199 | 		num_output: 512
2200 | 		kernel_size: 1
2201 | 		pad: 0
2202 | 		stride: 1
2203 | 		bias_term: false
2204 | 	}
2205 | }
2206 | 
2207 | layer {
2208 | 	bottom: "res5c_branch2a"
2209 | 	top: "res5c_branch2a"
2210 | 	name: "bn5c_branch2a"
2211 | 	type: "BatchNorm"
2212 | 	batch_norm_param {
2213 | 		use_global_stats: true
2214 | 	}
2215 | }
2216 | 
2217 | layer {
2218 | 	bottom: "res5c_branch2a"
2219 | 	top: "res5c_branch2a"
2220 | 	name: "scale5c_branch2a"
2221 | 	type: "Scale"
2222 | 	scale_param {
2223 | 		bias_term: true
2224 | 	}
2225 | }
2226 | 
2227 | layer {
2228 | 	bottom: "res5c_branch2a"
2229 | 	top: "res5c_branch2a"
2230 | 	name: "res5c_branch2a_relu"
2231 | 	type: "ReLU"
2232 | }
2233 | 
2234 | layer {
2235 | 	bottom: "res5c_branch2a"
2236 | 	top: "res5c_branch2b"
2237 | 	name: "res5c_branch2b"
2238 | 	type: "Convolution"
2239 | 	convolution_param {
2240 | 		num_output: 512
2241 | 		kernel_size: 3
2242 | 		pad: 1
2243 | 		stride: 1
2244 | 		bias_term: false
2245 | 	}
2246 | }
2247 | 
2248 | layer {
2249 | 	bottom: "res5c_branch2b"
2250 | 	top: "res5c_branch2b"
2251 | 	name: "bn5c_branch2b"
2252 | 	type: "BatchNorm"
2253 | 	batch_norm_param {
2254 | 		use_global_stats: true
2255 | 	}
2256 | }
2257 | 
2258 | layer {
2259 | 	bottom: "res5c_branch2b"
2260 | 	top: "res5c_branch2b"
2261 | 	name: "scale5c_branch2b"
2262 | 	type: "Scale"
2263 | 	scale_param {
2264 | 		bias_term: true
2265 | 	}
2266 | }
2267 | 
2268 | layer {
2269 | 	bottom: "res5c_branch2b"
2270 | 	top: "res5c_branch2b"
2271 | 	name: "res5c_branch2b_relu"
2272 | 	type: "ReLU"
2273 | }
2274 | 
2275 | layer {
2276 | 	bottom: "res5c_branch2b"
2277 | 	top: "res5c_branch2c"
2278 | 	name: "res5c_branch2c"
2279 | 	type: "Convolution"
2280 | 	convolution_param {
2281 | 		num_output: 2048
2282 | 		kernel_size: 1
2283 | 		pad: 0
2284 | 		stride: 1
2285 | 		bias_term: false
2286 | 	}
2287 | }
2288 | 
2289 | layer {
2290 | 	bottom: "res5c_branch2c"
2291 | 	top: "res5c_branch2c"
2292 | 	name: "bn5c_branch2c"
2293 | 	type: "BatchNorm"
2294 | 	batch_norm_param {
2295 | 		use_global_stats: true
2296 | 	}
2297 | }
2298 | 
2299 | layer {
2300 | 	bottom: "res5c_branch2c"
2301 | 	top: "res5c_branch2c"
2302 | 	name: "scale5c_branch2c"
2303 | 	type: "Scale"
2304 | 	scale_param {
2305 | 		bias_term: true
2306 | 	}
2307 | }
2308 | 
2309 | layer {
2310 | 	bottom: "res5b"
2311 | 	bottom: "res5c_branch2c"
2312 | 	top: "res5c"
2313 | 	name: "res5c"
2314 | 	type: "Eltwise"
2315 | }
2316 | 
2317 | layer {
2318 | 	bottom: "res5c"
2319 | 	top: "res5c"
2320 | 	name: "res5c_relu"
2321 | 	type: "ReLU"
2322 | }
2323 | 
2324 | layer {
2325 | 	bottom: "res5c"
2326 | 	top: "pool5"
2327 | 	name: "pool5"
2328 | 	type: "Pooling"
2329 | 	pooling_param {
2330 | 		kernel_size: 7
2331 | 		stride: 1
2332 | 		pool: AVE
2333 | 	}
2334 | }
2335 | 
2336 | layer {
2337 | 	bottom: "pool5"
2338 | 	top: "my-classifier"
2339 | 	name: "my-classifier"
2340 | 	type: "InnerProduct"
2341 | 	param {
2342 |         lr_mult: 10
2343 |         decay_mult: 1
2344 |     }
2345 |     param {
2346 |         lr_mult: 20
2347 |         decay_mult: 0
2348 |     }
2349 |     inner_product_param {
2350 | 		num_output: 3
2351 | 	}
2352 | }
2353 | layer {
2354 |   name: "loss1/loss1"
2355 |   type: "SoftmaxWithLoss"
2356 |   bottom: "my-classifier"
2357 |   bottom: "label"
2358 |   top: "loss1/loss1"
2359 |   loss_weight: 0.3
2360 | }
2361 | layer {
2362 |   name: "test/loss"
2363 |   type: "SoftmaxWithLoss"
2364 |   bottom: "my-classifier"
2365 |   bottom: "label"
2366 |   top: "test/loss"
2367 |   include {
2368 |     phase: TEST
2369 |   }
2370 | }
2371 | layer {
2372 |   name: "accuracy"
2373 |   type: "Accuracy"
2374 |   bottom: "my-classifier"
2375 |   bottom: "label"
2376 |   top: "accuracy"
2377 |   include {
2378 |     phase: TEST
2379 |   }
2380 | }
2381 | 


--------------------------------------------------------------------------------
/SE-BN-Inception/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | #test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | #test_interval: 10000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | #average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 5000
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 25
26 | # save path
27 | snapshot_prefix: "se-bn-inception-dog"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/SE-ResNet-50/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | #test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | #test_interval: 10000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | #average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 5000
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 50
26 | # save path
27 | snapshot_prefix: "se-resnet-50-dog"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/SqueezeNet/deploy.prototxt:
--------------------------------------------------------------------------------
  1 | input: "data"
  2 | input_shape {
  3 |   dim: 1
  4 |   dim: 3
  5 |   dim: 227
  6 |   dim: 227
  7 | }
  8 | layer {
  9 |   name: "conv1"
 10 |   type: "Convolution"
 11 |   bottom: "data"
 12 |   top: "conv1"
 13 |   convolution_param {
 14 |     num_output: 64
 15 |     kernel_size: 3
 16 |     stride: 2
 17 |     weight_filler {
 18 |       type: "xavier"
 19 |     }
 20 |   }
 21 | }
 22 | layer {
 23 |   name: "relu_conv1"
 24 |   type: "ReLU"
 25 |   bottom: "conv1"
 26 |   top: "conv1"
 27 | }
 28 | layer {
 29 |   name: "pool1"
 30 |   type: "Pooling"
 31 |   bottom: "conv1"
 32 |   top: "pool1"
 33 |   pooling_param {
 34 |     pool: MAX
 35 |     kernel_size: 3
 36 |     stride: 2
 37 |   }
 38 | }
 39 | layer {
 40 |   name: "fire2/squeeze1x1"
 41 |   type: "Convolution"
 42 |   bottom: "pool1"
 43 |   top: "fire2/squeeze1x1"
 44 |   convolution_param {
 45 |     num_output: 16
 46 |     kernel_size: 1
 47 |     weight_filler {
 48 |       type: "xavier"
 49 |     }
 50 |   }
 51 | }
 52 | layer {
 53 |   name: "fire2/relu_squeeze1x1"
 54 |   type: "ReLU"
 55 |   bottom: "fire2/squeeze1x1"
 56 |   top: "fire2/squeeze1x1"
 57 | }
 58 | layer {
 59 |   name: "fire2/expand1x1"
 60 |   type: "Convolution"
 61 |   bottom: "fire2/squeeze1x1"
 62 |   top: "fire2/expand1x1"
 63 |   convolution_param {
 64 |     num_output: 64
 65 |     kernel_size: 1
 66 |     weight_filler {
 67 |       type: "xavier"
 68 |     }
 69 |   }
 70 | }
 71 | layer {
 72 |   name: "fire2/relu_expand1x1"
 73 |   type: "ReLU"
 74 |   bottom: "fire2/expand1x1"
 75 |   top: "fire2/expand1x1"
 76 | }
 77 | layer {
 78 |   name: "fire2/expand3x3"
 79 |   type: "Convolution"
 80 |   bottom: "fire2/squeeze1x1"
 81 |   top: "fire2/expand3x3"
 82 |   convolution_param {
 83 |     num_output: 64
 84 |     pad: 1
 85 |     kernel_size: 3
 86 |     weight_filler {
 87 |       type: "xavier"
 88 |     }
 89 |   }
 90 | }
 91 | layer {
 92 |   name: "fire2/relu_expand3x3"
 93 |   type: "ReLU"
 94 |   bottom: "fire2/expand3x3"
 95 |   top: "fire2/expand3x3"
 96 | }
 97 | layer {
 98 |   name: "fire2/concat"
 99 |   type: "Concat"
100 |   bottom: "fire2/expand1x1"
101 |   bottom: "fire2/expand3x3"
102 |   top: "fire2/concat"
103 | }
104 | layer {
105 |   name: "fire3/squeeze1x1"
106 |   type: "Convolution"
107 |   bottom: "fire2/concat"
108 |   top: "fire3/squeeze1x1"
109 |   convolution_param {
110 |     num_output: 16
111 |     kernel_size: 1
112 |     weight_filler {
113 |       type: "xavier"
114 |     }
115 |   }
116 | }
117 | layer {
118 |   name: "fire3/relu_squeeze1x1"
119 |   type: "ReLU"
120 |   bottom: "fire3/squeeze1x1"
121 |   top: "fire3/squeeze1x1"
122 | }
123 | layer {
124 |   name: "fire3/expand1x1"
125 |   type: "Convolution"
126 |   bottom: "fire3/squeeze1x1"
127 |   top: "fire3/expand1x1"
128 |   convolution_param {
129 |     num_output: 64
130 |     kernel_size: 1
131 |     weight_filler {
132 |       type: "xavier"
133 |     }
134 |   }
135 | }
136 | layer {
137 |   name: "fire3/relu_expand1x1"
138 |   type: "ReLU"
139 |   bottom: "fire3/expand1x1"
140 |   top: "fire3/expand1x1"
141 | }
142 | layer {
143 |   name: "fire3/expand3x3"
144 |   type: "Convolution"
145 |   bottom: "fire3/squeeze1x1"
146 |   top: "fire3/expand3x3"
147 |   convolution_param {
148 |     num_output: 64
149 |     pad: 1
150 |     kernel_size: 3
151 |     weight_filler {
152 |       type: "xavier"
153 |     }
154 |   }
155 | }
156 | layer {
157 |   name: "fire3/relu_expand3x3"
158 |   type: "ReLU"
159 |   bottom: "fire3/expand3x3"
160 |   top: "fire3/expand3x3"
161 | }
162 | layer {
163 |   name: "fire3/concat"
164 |   type: "Concat"
165 |   bottom: "fire3/expand1x1"
166 |   bottom: "fire3/expand3x3"
167 |   top: "fire3/concat"
168 | }
169 | layer {
170 |   name: "pool3"
171 |   type: "Pooling"
172 |   bottom: "fire3/concat"
173 |   top: "pool3"
174 |   pooling_param {
175 |     pool: MAX
176 |     kernel_size: 3
177 |     stride: 2
178 |   }
179 | }
180 | layer {
181 |   name: "fire4/squeeze1x1"
182 |   type: "Convolution"
183 |   bottom: "pool3"
184 |   top: "fire4/squeeze1x1"
185 |   convolution_param {
186 |     num_output: 32
187 |     kernel_size: 1
188 |     weight_filler {
189 |       type: "xavier"
190 |     }
191 |   }
192 | }
193 | layer {
194 |   name: "fire4/relu_squeeze1x1"
195 |   type: "ReLU"
196 |   bottom: "fire4/squeeze1x1"
197 |   top: "fire4/squeeze1x1"
198 | }
199 | layer {
200 |   name: "fire4/expand1x1"
201 |   type: "Convolution"
202 |   bottom: "fire4/squeeze1x1"
203 |   top: "fire4/expand1x1"
204 |   convolution_param {
205 |     num_output: 128
206 |     kernel_size: 1
207 |     weight_filler {
208 |       type: "xavier"
209 |     }
210 |   }
211 | }
212 | layer {
213 |   name: "fire4/relu_expand1x1"
214 |   type: "ReLU"
215 |   bottom: "fire4/expand1x1"
216 |   top: "fire4/expand1x1"
217 | }
218 | layer {
219 |   name: "fire4/expand3x3"
220 |   type: "Convolution"
221 |   bottom: "fire4/squeeze1x1"
222 |   top: "fire4/expand3x3"
223 |   convolution_param {
224 |     num_output: 128
225 |     pad: 1
226 |     kernel_size: 3
227 |     weight_filler {
228 |       type: "xavier"
229 |     }
230 |   }
231 | }
232 | layer {
233 |   name: "fire4/relu_expand3x3"
234 |   type: "ReLU"
235 |   bottom: "fire4/expand3x3"
236 |   top: "fire4/expand3x3"
237 | }
238 | layer {
239 |   name: "fire4/concat"
240 |   type: "Concat"
241 |   bottom: "fire4/expand1x1"
242 |   bottom: "fire4/expand3x3"
243 |   top: "fire4/concat"
244 | }
245 | layer {
246 |   name: "fire5/squeeze1x1"
247 |   type: "Convolution"
248 |   bottom: "fire4/concat"
249 |   top: "fire5/squeeze1x1"
250 |   convolution_param {
251 |     num_output: 32
252 |     kernel_size: 1
253 |     weight_filler {
254 |       type: "xavier"
255 |     }
256 |   }
257 | }
258 | layer {
259 |   name: "fire5/relu_squeeze1x1"
260 |   type: "ReLU"
261 |   bottom: "fire5/squeeze1x1"
262 |   top: "fire5/squeeze1x1"
263 | }
264 | layer {
265 |   name: "fire5/expand1x1"
266 |   type: "Convolution"
267 |   bottom: "fire5/squeeze1x1"
268 |   top: "fire5/expand1x1"
269 |   convolution_param {
270 |     num_output: 128
271 |     kernel_size: 1
272 |     weight_filler {
273 |       type: "xavier"
274 |     }
275 |   }
276 | }
277 | layer {
278 |   name: "fire5/relu_expand1x1"
279 |   type: "ReLU"
280 |   bottom: "fire5/expand1x1"
281 |   top: "fire5/expand1x1"
282 | }
283 | layer {
284 |   name: "fire5/expand3x3"
285 |   type: "Convolution"
286 |   bottom: "fire5/squeeze1x1"
287 |   top: "fire5/expand3x3"
288 |   convolution_param {
289 |     num_output: 128
290 |     pad: 1
291 |     kernel_size: 3
292 |     weight_filler {
293 |       type: "xavier"
294 |     }
295 |   }
296 | }
297 | layer {
298 |   name: "fire5/relu_expand3x3"
299 |   type: "ReLU"
300 |   bottom: "fire5/expand3x3"
301 |   top: "fire5/expand3x3"
302 | }
303 | layer {
304 |   name: "fire5/concat"
305 |   type: "Concat"
306 |   bottom: "fire5/expand1x1"
307 |   bottom: "fire5/expand3x3"
308 |   top: "fire5/concat"
309 | }
310 | layer {
311 |   name: "pool5"
312 |   type: "Pooling"
313 |   bottom: "fire5/concat"
314 |   top: "pool5"
315 |   pooling_param {
316 |     pool: MAX
317 |     kernel_size: 3
318 |     stride: 2
319 |   }
320 | }
321 | layer {
322 |   name: "fire6/squeeze1x1"
323 |   type: "Convolution"
324 |   bottom: "pool5"
325 |   top: "fire6/squeeze1x1"
326 |   convolution_param {
327 |     num_output: 48
328 |     kernel_size: 1
329 |     weight_filler {
330 |       type: "xavier"
331 |     }
332 |   }
333 | }
334 | layer {
335 |   name: "fire6/relu_squeeze1x1"
336 |   type: "ReLU"
337 |   bottom: "fire6/squeeze1x1"
338 |   top: "fire6/squeeze1x1"
339 | }
340 | layer {
341 |   name: "fire6/expand1x1"
342 |   type: "Convolution"
343 |   bottom: "fire6/squeeze1x1"
344 |   top: "fire6/expand1x1"
345 |   convolution_param {
346 |     num_output: 192
347 |     kernel_size: 1
348 |     weight_filler {
349 |       type: "xavier"
350 |     }
351 |   }
352 | }
353 | layer {
354 |   name: "fire6/relu_expand1x1"
355 |   type: "ReLU"
356 |   bottom: "fire6/expand1x1"
357 |   top: "fire6/expand1x1"
358 | }
359 | layer {
360 |   name: "fire6/expand3x3"
361 |   type: "Convolution"
362 |   bottom: "fire6/squeeze1x1"
363 |   top: "fire6/expand3x3"
364 |   convolution_param {
365 |     num_output: 192
366 |     pad: 1
367 |     kernel_size: 3
368 |     weight_filler {
369 |       type: "xavier"
370 |     }
371 |   }
372 | }
373 | layer {
374 |   name: "fire6/relu_expand3x3"
375 |   type: "ReLU"
376 |   bottom: "fire6/expand3x3"
377 |   top: "fire6/expand3x3"
378 | }
379 | layer {
380 |   name: "fire6/concat"
381 |   type: "Concat"
382 |   bottom: "fire6/expand1x1"
383 |   bottom: "fire6/expand3x3"
384 |   top: "fire6/concat"
385 | }
386 | layer {
387 |   name: "fire7/squeeze1x1"
388 |   type: "Convolution"
389 |   bottom: "fire6/concat"
390 |   top: "fire7/squeeze1x1"
391 |   convolution_param {
392 |     num_output: 48
393 |     kernel_size: 1
394 |     weight_filler {
395 |       type: "xavier"
396 |     }
397 |   }
398 | }
399 | layer {
400 |   name: "fire7/relu_squeeze1x1"
401 |   type: "ReLU"
402 |   bottom: "fire7/squeeze1x1"
403 |   top: "fire7/squeeze1x1"
404 | }
405 | layer {
406 |   name: "fire7/expand1x1"
407 |   type: "Convolution"
408 |   bottom: "fire7/squeeze1x1"
409 |   top: "fire7/expand1x1"
410 |   convolution_param {
411 |     num_output: 192
412 |     kernel_size: 1
413 |     weight_filler {
414 |       type: "xavier"
415 |     }
416 |   }
417 | }
418 | layer {
419 |   name: "fire7/relu_expand1x1"
420 |   type: "ReLU"
421 |   bottom: "fire7/expand1x1"
422 |   top: "fire7/expand1x1"
423 | }
424 | layer {
425 |   name: "fire7/expand3x3"
426 |   type: "Convolution"
427 |   bottom: "fire7/squeeze1x1"
428 |   top: "fire7/expand3x3"
429 |   convolution_param {
430 |     num_output: 192
431 |     pad: 1
432 |     kernel_size: 3
433 |     weight_filler {
434 |       type: "xavier"
435 |     }
436 |   }
437 | }
438 | layer {
439 |   name: "fire7/relu_expand3x3"
440 |   type: "ReLU"
441 |   bottom: "fire7/expand3x3"
442 |   top: "fire7/expand3x3"
443 | }
444 | layer {
445 |   name: "fire7/concat"
446 |   type: "Concat"
447 |   bottom: "fire7/expand1x1"
448 |   bottom: "fire7/expand3x3"
449 |   top: "fire7/concat"
450 | }
451 | layer {
452 |   name: "fire8/squeeze1x1"
453 |   type: "Convolution"
454 |   bottom: "fire7/concat"
455 |   top: "fire8/squeeze1x1"
456 |   convolution_param {
457 |     num_output: 64
458 |     kernel_size: 1
459 |     weight_filler {
460 |       type: "xavier"
461 |     }
462 |   }
463 | }
464 | layer {
465 |   name: "fire8/relu_squeeze1x1"
466 |   type: "ReLU"
467 |   bottom: "fire8/squeeze1x1"
468 |   top: "fire8/squeeze1x1"
469 | }
470 | layer {
471 |   name: "fire8/expand1x1"
472 |   type: "Convolution"
473 |   bottom: "fire8/squeeze1x1"
474 |   top: "fire8/expand1x1"
475 |   convolution_param {
476 |     num_output: 256
477 |     kernel_size: 1
478 |     weight_filler {
479 |       type: "xavier"
480 |     }
481 |   }
482 | }
483 | layer {
484 |   name: "fire8/relu_expand1x1"
485 |   type: "ReLU"
486 |   bottom: "fire8/expand1x1"
487 |   top: "fire8/expand1x1"
488 | }
489 | layer {
490 |   name: "fire8/expand3x3"
491 |   type: "Convolution"
492 |   bottom: "fire8/squeeze1x1"
493 |   top: "fire8/expand3x3"
494 |   convolution_param {
495 |     num_output: 256
496 |     pad: 1
497 |     kernel_size: 3
498 |     weight_filler {
499 |       type: "xavier"
500 |     }
501 |   }
502 | }
503 | layer {
504 |   name: "fire8/relu_expand3x3"
505 |   type: "ReLU"
506 |   bottom: "fire8/expand3x3"
507 |   top: "fire8/expand3x3"
508 | }
509 | layer {
510 |   name: "fire8/concat"
511 |   type: "Concat"
512 |   bottom: "fire8/expand1x1"
513 |   bottom: "fire8/expand3x3"
514 |   top: "fire8/concat"
515 | }
516 | layer {
517 |   name: "fire9/squeeze1x1"
518 |   type: "Convolution"
519 |   bottom: "fire8/concat"
520 |   top: "fire9/squeeze1x1"
521 |   convolution_param {
522 |     num_output: 64
523 |     kernel_size: 1
524 |     weight_filler {
525 |       type: "xavier"
526 |     }
527 |   }
528 | }
529 | layer {
530 |   name: "fire9/relu_squeeze1x1"
531 |   type: "ReLU"
532 |   bottom: "fire9/squeeze1x1"
533 |   top: "fire9/squeeze1x1"
534 | }
535 | layer {
536 |   name: "fire9/expand1x1"
537 |   type: "Convolution"
538 |   bottom: "fire9/squeeze1x1"
539 |   top: "fire9/expand1x1"
540 |   convolution_param {
541 |     num_output: 256
542 |     kernel_size: 1
543 |     weight_filler {
544 |       type: "xavier"
545 |     }
546 |   }
547 | }
548 | layer {
549 |   name: "fire9/relu_expand1x1"
550 |   type: "ReLU"
551 |   bottom: "fire9/expand1x1"
552 |   top: "fire9/expand1x1"
553 | }
554 | layer {
555 |   name: "fire9/expand3x3"
556 |   type: "Convolution"
557 |   bottom: "fire9/squeeze1x1"
558 |   top: "fire9/expand3x3"
559 |   convolution_param {
560 |     num_output: 256
561 |     pad: 1
562 |     kernel_size: 3
563 |     weight_filler {
564 |       type: "xavier"
565 |     }
566 |   }
567 | }
568 | layer {
569 |   name: "fire9/relu_expand3x3"
570 |   type: "ReLU"
571 |   bottom: "fire9/expand3x3"
572 |   top: "fire9/expand3x3"
573 | }
574 | layer {
575 |   name: "fire9/concat"
576 |   type: "Concat"
577 |   bottom: "fire9/expand1x1"
578 |   bottom: "fire9/expand3x3"
579 |   top: "fire9/concat"
580 | }
581 | layer {
582 |   name: "drop9"
583 |   type: "Dropout"
584 |   bottom: "fire9/concat"
585 |   top: "fire9/concat"
586 |   dropout_param {
587 |     dropout_ratio: 0.5
588 |   }
589 | }
590 | layer {
591 |   name: "my-conv10"
592 |   type: "Convolution"
593 |   bottom: "fire9/concat"
594 |   top: "my-conv10"
595 |   convolution_param {
596 |     num_output: 3
597 |     kernel_size: 1
598 |     weight_filler {
599 |       type: "gaussian"
600 |       mean: 0.0
601 |       std: 0.01
602 |     }
603 |   }
604 | }
605 | layer {
606 |   name: "relu_conv10"
607 |   type: "ReLU"
608 |   bottom: "my-conv10"
609 |   top: "conv10"
610 | }
611 | layer {
612 |   name: "pool10"
613 |   type: "Pooling"
614 |   bottom: "conv10"
615 |   top: "pool10"
616 |   pooling_param {
617 |     pool: AVE
618 |     global_pooling: true
619 |   }
620 | }
621 | layer {
622 |   name: "prob"
623 |   type: "Softmax"
624 |   bottom: "pool10"
625 |   top: "prob"
626 | }
627 | 


--------------------------------------------------------------------------------
/SqueezeNet/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 10000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 10001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 20
26 | # save path
27 | snapshot_prefix: "squeezenet"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/SqueezeNet/squeezenet_v1.1.caffemodel:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/SnailTyan/caffe-model-zoo/83fb66248583f20fc7ce2f9d40fc6fe09ef4d4b9/SqueezeNet/squeezenet_v1.1.caffemodel


--------------------------------------------------------------------------------
/SqueezeNet/train_val.prototxt:
--------------------------------------------------------------------------------
  1 | name: "SqueezeNet"
  2 | layer {
  3 |   name: "data"
  4 |   type: "Data"
  5 |   top: "data"
  6 |   top: "label"
  7 |   include {
  8 |     phase: TRAIN
  9 |   }
 10 |   transform_param {
 11 |     mirror: true
 12 |     crop_size: 227
 13 |     mean_file: "mean.binaryproto"
 14 |   }
 15 |   data_param {
 16 |     source: "train_lmdb"
 17 |     batch_size: 64
 18 |     backend: LMDB
 19 |   }
 20 | }
 21 | layer {
 22 |   name: "data"
 23 |   type: "Data"
 24 |   top: "data"
 25 |   top: "label"
 26 |   include {
 27 |     phase: TEST
 28 |   }
 29 |   transform_param {
 30 |     mirror: false
 31 |     crop_size: 227
 32 |     mean_file: "mean.binaryproto"
 33 |   }
 34 |   data_param {
 35 |     source: "val_lmdb"
 36 |     batch_size: 1
 37 |     backend: LMDB
 38 |   }
 39 | }
 40 | layer {
 41 |   name: "conv1"
 42 |   type: "Convolution"
 43 |   bottom: "data"
 44 |   top: "conv1"
 45 |   convolution_param {
 46 |     num_output: 64
 47 |     kernel_size: 3
 48 |     stride: 2
 49 |     weight_filler {
 50 |       type: "xavier"
 51 |     }
 52 |   }
 53 | }
 54 | layer {
 55 |   name: "relu_conv1"
 56 |   type: "ReLU"
 57 |   bottom: "conv1"
 58 |   top: "conv1"
 59 | }
 60 | layer {
 61 |   name: "pool1"
 62 |   type: "Pooling"
 63 |   bottom: "conv1"
 64 |   top: "pool1"
 65 |   pooling_param {
 66 |     pool: MAX
 67 |     kernel_size: 3
 68 |     stride: 2
 69 |   }
 70 | }
 71 | layer {
 72 |   name: "fire2/squeeze1x1"
 73 |   type: "Convolution"
 74 |   bottom: "pool1"
 75 |   top: "fire2/squeeze1x1"
 76 |   convolution_param {
 77 |     num_output: 16
 78 |     kernel_size: 1
 79 |     weight_filler {
 80 |       type: "xavier"
 81 |     }
 82 |   }
 83 | }
 84 | layer {
 85 |   name: "fire2/relu_squeeze1x1"
 86 |   type: "ReLU"
 87 |   bottom: "fire2/squeeze1x1"
 88 |   top: "fire2/squeeze1x1"
 89 | }
 90 | layer {
 91 |   name: "fire2/expand1x1"
 92 |   type: "Convolution"
 93 |   bottom: "fire2/squeeze1x1"
 94 |   top: "fire2/expand1x1"
 95 |   convolution_param {
 96 |     num_output: 64
 97 |     kernel_size: 1
 98 |     weight_filler {
 99 |       type: "xavier"
100 |     }
101 |   }
102 | }
103 | layer {
104 |   name: "fire2/relu_expand1x1"
105 |   type: "ReLU"
106 |   bottom: "fire2/expand1x1"
107 |   top: "fire2/expand1x1"
108 | }
109 | layer {
110 |   name: "fire2/expand3x3"
111 |   type: "Convolution"
112 |   bottom: "fire2/squeeze1x1"
113 |   top: "fire2/expand3x3"
114 |   convolution_param {
115 |     num_output: 64
116 |     pad: 1
117 |     kernel_size: 3
118 |     weight_filler {
119 |       type: "xavier"
120 |     }
121 |   }
122 | }
123 | layer {
124 |   name: "fire2/relu_expand3x3"
125 |   type: "ReLU"
126 |   bottom: "fire2/expand3x3"
127 |   top: "fire2/expand3x3"
128 | }
129 | layer {
130 |   name: "fire2/concat"
131 |   type: "Concat"
132 |   bottom: "fire2/expand1x1"
133 |   bottom: "fire2/expand3x3"
134 |   top: "fire2/concat"
135 | }
136 | layer {
137 |   name: "fire3/squeeze1x1"
138 |   type: "Convolution"
139 |   bottom: "fire2/concat"
140 |   top: "fire3/squeeze1x1"
141 |   convolution_param {
142 |     num_output: 16
143 |     kernel_size: 1
144 |     weight_filler {
145 |       type: "xavier"
146 |     }
147 |   }
148 | }
149 | layer {
150 |   name: "fire3/relu_squeeze1x1"
151 |   type: "ReLU"
152 |   bottom: "fire3/squeeze1x1"
153 |   top: "fire3/squeeze1x1"
154 | }
155 | layer {
156 |   name: "fire3/expand1x1"
157 |   type: "Convolution"
158 |   bottom: "fire3/squeeze1x1"
159 |   top: "fire3/expand1x1"
160 |   convolution_param {
161 |     num_output: 64
162 |     kernel_size: 1
163 |     weight_filler {
164 |       type: "xavier"
165 |     }
166 |   }
167 | }
168 | layer {
169 |   name: "fire3/relu_expand1x1"
170 |   type: "ReLU"
171 |   bottom: "fire3/expand1x1"
172 |   top: "fire3/expand1x1"
173 | }
174 | layer {
175 |   name: "fire3/expand3x3"
176 |   type: "Convolution"
177 |   bottom: "fire3/squeeze1x1"
178 |   top: "fire3/expand3x3"
179 |   convolution_param {
180 |     num_output: 64
181 |     pad: 1
182 |     kernel_size: 3
183 |     weight_filler {
184 |       type: "xavier"
185 |     }
186 |   }
187 | }
188 | layer {
189 |   name: "fire3/relu_expand3x3"
190 |   type: "ReLU"
191 |   bottom: "fire3/expand3x3"
192 |   top: "fire3/expand3x3"
193 | }
194 | layer {
195 |   name: "fire3/concat"
196 |   type: "Concat"
197 |   bottom: "fire3/expand1x1"
198 |   bottom: "fire3/expand3x3"
199 |   top: "fire3/concat"
200 | }
201 | layer {
202 |   name: "pool3"
203 |   type: "Pooling"
204 |   bottom: "fire3/concat"
205 |   top: "pool3"
206 |   pooling_param {
207 |     pool: MAX
208 |     kernel_size: 3
209 |     stride: 2
210 |   }
211 | }
212 | layer {
213 |   name: "fire4/squeeze1x1"
214 |   type: "Convolution"
215 |   bottom: "pool3"
216 |   top: "fire4/squeeze1x1"
217 |   convolution_param {
218 |     num_output: 32
219 |     kernel_size: 1
220 |     weight_filler {
221 |       type: "xavier"
222 |     }
223 |   }
224 | }
225 | layer {
226 |   name: "fire4/relu_squeeze1x1"
227 |   type: "ReLU"
228 |   bottom: "fire4/squeeze1x1"
229 |   top: "fire4/squeeze1x1"
230 | }
231 | layer {
232 |   name: "fire4/expand1x1"
233 |   type: "Convolution"
234 |   bottom: "fire4/squeeze1x1"
235 |   top: "fire4/expand1x1"
236 |   convolution_param {
237 |     num_output: 128
238 |     kernel_size: 1
239 |     weight_filler {
240 |       type: "xavier"
241 |     }
242 |   }
243 | }
244 | layer {
245 |   name: "fire4/relu_expand1x1"
246 |   type: "ReLU"
247 |   bottom: "fire4/expand1x1"
248 |   top: "fire4/expand1x1"
249 | }
250 | layer {
251 |   name: "fire4/expand3x3"
252 |   type: "Convolution"
253 |   bottom: "fire4/squeeze1x1"
254 |   top: "fire4/expand3x3"
255 |   convolution_param {
256 |     num_output: 128
257 |     pad: 1
258 |     kernel_size: 3
259 |     weight_filler {
260 |       type: "xavier"
261 |     }
262 |   }
263 | }
264 | layer {
265 |   name: "fire4/relu_expand3x3"
266 |   type: "ReLU"
267 |   bottom: "fire4/expand3x3"
268 |   top: "fire4/expand3x3"
269 | }
270 | layer {
271 |   name: "fire4/concat"
272 |   type: "Concat"
273 |   bottom: "fire4/expand1x1"
274 |   bottom: "fire4/expand3x3"
275 |   top: "fire4/concat"
276 | }
277 | layer {
278 |   name: "fire5/squeeze1x1"
279 |   type: "Convolution"
280 |   bottom: "fire4/concat"
281 |   top: "fire5/squeeze1x1"
282 |   convolution_param {
283 |     num_output: 32
284 |     kernel_size: 1
285 |     weight_filler {
286 |       type: "xavier"
287 |     }
288 |   }
289 | }
290 | layer {
291 |   name: "fire5/relu_squeeze1x1"
292 |   type: "ReLU"
293 |   bottom: "fire5/squeeze1x1"
294 |   top: "fire5/squeeze1x1"
295 | }
296 | layer {
297 |   name: "fire5/expand1x1"
298 |   type: "Convolution"
299 |   bottom: "fire5/squeeze1x1"
300 |   top: "fire5/expand1x1"
301 |   convolution_param {
302 |     num_output: 128
303 |     kernel_size: 1
304 |     weight_filler {
305 |       type: "xavier"
306 |     }
307 |   }
308 | }
309 | layer {
310 |   name: "fire5/relu_expand1x1"
311 |   type: "ReLU"
312 |   bottom: "fire5/expand1x1"
313 |   top: "fire5/expand1x1"
314 | }
315 | layer {
316 |   name: "fire5/expand3x3"
317 |   type: "Convolution"
318 |   bottom: "fire5/squeeze1x1"
319 |   top: "fire5/expand3x3"
320 |   convolution_param {
321 |     num_output: 128
322 |     pad: 1
323 |     kernel_size: 3
324 |     weight_filler {
325 |       type: "xavier"
326 |     }
327 |   }
328 | }
329 | layer {
330 |   name: "fire5/relu_expand3x3"
331 |   type: "ReLU"
332 |   bottom: "fire5/expand3x3"
333 |   top: "fire5/expand3x3"
334 | }
335 | layer {
336 |   name: "fire5/concat"
337 |   type: "Concat"
338 |   bottom: "fire5/expand1x1"
339 |   bottom: "fire5/expand3x3"
340 |   top: "fire5/concat"
341 | }
342 | layer {
343 |   name: "pool5"
344 |   type: "Pooling"
345 |   bottom: "fire5/concat"
346 |   top: "pool5"
347 |   pooling_param {
348 |     pool: MAX
349 |     kernel_size: 3
350 |     stride: 2
351 |   }
352 | }
353 | layer {
354 |   name: "fire6/squeeze1x1"
355 |   type: "Convolution"
356 |   bottom: "pool5"
357 |   top: "fire6/squeeze1x1"
358 |   convolution_param {
359 |     num_output: 48
360 |     kernel_size: 1
361 |     weight_filler {
362 |       type: "xavier"
363 |     }
364 |   }
365 | }
366 | layer {
367 |   name: "fire6/relu_squeeze1x1"
368 |   type: "ReLU"
369 |   bottom: "fire6/squeeze1x1"
370 |   top: "fire6/squeeze1x1"
371 | }
372 | layer {
373 |   name: "fire6/expand1x1"
374 |   type: "Convolution"
375 |   bottom: "fire6/squeeze1x1"
376 |   top: "fire6/expand1x1"
377 |   convolution_param {
378 |     num_output: 192
379 |     kernel_size: 1
380 |     weight_filler {
381 |       type: "xavier"
382 |     }
383 |   }
384 | }
385 | layer {
386 |   name: "fire6/relu_expand1x1"
387 |   type: "ReLU"
388 |   bottom: "fire6/expand1x1"
389 |   top: "fire6/expand1x1"
390 | }
391 | layer {
392 |   name: "fire6/expand3x3"
393 |   type: "Convolution"
394 |   bottom: "fire6/squeeze1x1"
395 |   top: "fire6/expand3x3"
396 |   convolution_param {
397 |     num_output: 192
398 |     pad: 1
399 |     kernel_size: 3
400 |     weight_filler {
401 |       type: "xavier"
402 |     }
403 |   }
404 | }
405 | layer {
406 |   name: "fire6/relu_expand3x3"
407 |   type: "ReLU"
408 |   bottom: "fire6/expand3x3"
409 |   top: "fire6/expand3x3"
410 | }
411 | layer {
412 |   name: "fire6/concat"
413 |   type: "Concat"
414 |   bottom: "fire6/expand1x1"
415 |   bottom: "fire6/expand3x3"
416 |   top: "fire6/concat"
417 | }
418 | layer {
419 |   name: "fire7/squeeze1x1"
420 |   type: "Convolution"
421 |   bottom: "fire6/concat"
422 |   top: "fire7/squeeze1x1"
423 |   convolution_param {
424 |     num_output: 48
425 |     kernel_size: 1
426 |     weight_filler {
427 |       type: "xavier"
428 |     }
429 |   }
430 | }
431 | layer {
432 |   name: "fire7/relu_squeeze1x1"
433 |   type: "ReLU"
434 |   bottom: "fire7/squeeze1x1"
435 |   top: "fire7/squeeze1x1"
436 | }
437 | layer {
438 |   name: "fire7/expand1x1"
439 |   type: "Convolution"
440 |   bottom: "fire7/squeeze1x1"
441 |   top: "fire7/expand1x1"
442 |   convolution_param {
443 |     num_output: 192
444 |     kernel_size: 1
445 |     weight_filler {
446 |       type: "xavier"
447 |     }
448 |   }
449 | }
450 | layer {
451 |   name: "fire7/relu_expand1x1"
452 |   type: "ReLU"
453 |   bottom: "fire7/expand1x1"
454 |   top: "fire7/expand1x1"
455 | }
456 | layer {
457 |   name: "fire7/expand3x3"
458 |   type: "Convolution"
459 |   bottom: "fire7/squeeze1x1"
460 |   top: "fire7/expand3x3"
461 |   convolution_param {
462 |     num_output: 192
463 |     pad: 1
464 |     kernel_size: 3
465 |     weight_filler {
466 |       type: "xavier"
467 |     }
468 |   }
469 | }
470 | layer {
471 |   name: "fire7/relu_expand3x3"
472 |   type: "ReLU"
473 |   bottom: "fire7/expand3x3"
474 |   top: "fire7/expand3x3"
475 | }
476 | layer {
477 |   name: "fire7/concat"
478 |   type: "Concat"
479 |   bottom: "fire7/expand1x1"
480 |   bottom: "fire7/expand3x3"
481 |   top: "fire7/concat"
482 | }
483 | layer {
484 |   name: "fire8/squeeze1x1"
485 |   type: "Convolution"
486 |   bottom: "fire7/concat"
487 |   top: "fire8/squeeze1x1"
488 |   convolution_param {
489 |     num_output: 64
490 |     kernel_size: 1
491 |     weight_filler {
492 |       type: "xavier"
493 |     }
494 |   }
495 | }
496 | layer {
497 |   name: "fire8/relu_squeeze1x1"
498 |   type: "ReLU"
499 |   bottom: "fire8/squeeze1x1"
500 |   top: "fire8/squeeze1x1"
501 | }
502 | layer {
503 |   name: "fire8/expand1x1"
504 |   type: "Convolution"
505 |   bottom: "fire8/squeeze1x1"
506 |   top: "fire8/expand1x1"
507 |   convolution_param {
508 |     num_output: 256
509 |     kernel_size: 1
510 |     weight_filler {
511 |       type: "xavier"
512 |     }
513 |   }
514 | }
515 | layer {
516 |   name: "fire8/relu_expand1x1"
517 |   type: "ReLU"
518 |   bottom: "fire8/expand1x1"
519 |   top: "fire8/expand1x1"
520 | }
521 | layer {
522 |   name: "fire8/expand3x3"
523 |   type: "Convolution"
524 |   bottom: "fire8/squeeze1x1"
525 |   top: "fire8/expand3x3"
526 |   convolution_param {
527 |     num_output: 256
528 |     pad: 1
529 |     kernel_size: 3
530 |     weight_filler {
531 |       type: "xavier"
532 |     }
533 |   }
534 | }
535 | layer {
536 |   name: "fire8/relu_expand3x3"
537 |   type: "ReLU"
538 |   bottom: "fire8/expand3x3"
539 |   top: "fire8/expand3x3"
540 | }
541 | layer {
542 |   name: "fire8/concat"
543 |   type: "Concat"
544 |   bottom: "fire8/expand1x1"
545 |   bottom: "fire8/expand3x3"
546 |   top: "fire8/concat"
547 | }
548 | layer {
549 |   name: "fire9/squeeze1x1"
550 |   type: "Convolution"
551 |   bottom: "fire8/concat"
552 |   top: "fire9/squeeze1x1"
553 |   convolution_param {
554 |     num_output: 64
555 |     kernel_size: 1
556 |     weight_filler {
557 |       type: "xavier"
558 |     }
559 |   }
560 | }
561 | layer {
562 |   name: "fire9/relu_squeeze1x1"
563 |   type: "ReLU"
564 |   bottom: "fire9/squeeze1x1"
565 |   top: "fire9/squeeze1x1"
566 | }
567 | layer {
568 |   name: "fire9/expand1x1"
569 |   type: "Convolution"
570 |   bottom: "fire9/squeeze1x1"
571 |   top: "fire9/expand1x1"
572 |   convolution_param {
573 |     num_output: 256
574 |     kernel_size: 1
575 |     weight_filler {
576 |       type: "xavier"
577 |     }
578 |   }
579 | }
580 | layer {
581 |   name: "fire9/relu_expand1x1"
582 |   type: "ReLU"
583 |   bottom: "fire9/expand1x1"
584 |   top: "fire9/expand1x1"
585 | }
586 | layer {
587 |   name: "fire9/expand3x3"
588 |   type: "Convolution"
589 |   bottom: "fire9/squeeze1x1"
590 |   top: "fire9/expand3x3"
591 |   convolution_param {
592 |     num_output: 256
593 |     pad: 1
594 |     kernel_size: 3
595 |     weight_filler {
596 |       type: "xavier"
597 |     }
598 |   }
599 | }
600 | layer {
601 |   name: "fire9/relu_expand3x3"
602 |   type: "ReLU"
603 |   bottom: "fire9/expand3x3"
604 |   top: "fire9/expand3x3"
605 | }
606 | layer {
607 |   name: "fire9/concat"
608 |   type: "Concat"
609 |   bottom: "fire9/expand1x1"
610 |   bottom: "fire9/expand3x3"
611 |   top: "fire9/concat"
612 | }
613 | layer {
614 |   name: "drop9"
615 |   type: "Dropout"
616 |   bottom: "fire9/concat"
617 |   top: "fire9/concat"
618 |   dropout_param {
619 |     dropout_ratio: 0.5
620 |   }
621 | }
622 | layer {
623 |   name: "my-conv10"
624 |   type: "Convolution"
625 |   bottom: "fire9/concat"
626 |   top: "my-conv10"
627 |   convolution_param {
628 |     num_output: 3
629 |     kernel_size: 1
630 |     weight_filler {
631 |       type: "gaussian"
632 |       mean: 0.0
633 |       std: 0.01
634 |     }
635 |   }
636 | }
637 | layer {
638 |   name: "relu_conv10"
639 |   type: "ReLU"
640 |   bottom: "my-conv10"
641 |   top: "conv10"
642 | }
643 | layer {
644 |   name: "pool10"
645 |   type: "Pooling"
646 |   bottom: "conv10"
647 |   top: "pool10"
648 |   pooling_param {
649 |     pool: AVE
650 |     global_pooling: true
651 |   }
652 | }
653 | layer {
654 |   name: "loss"
655 |   type: "SoftmaxWithLoss"
656 |   bottom: "pool10"
657 |   bottom: "label"
658 |   top: "loss"
659 |   #include {
660 |   #  phase: TRAIN
661 |   #}
662 | }
663 | 


--------------------------------------------------------------------------------
/VGG16/deploy.prototxt:
--------------------------------------------------------------------------------
  1 | name: "VGG_ILSVRC_16_layer"
  2 | input: "data"
  3 | input_dim: 10
  4 | input_dim: 3
  5 | input_dim: 224
  6 | input_dim: 224
  7 | layer {
  8 |   bottom: "data"
  9 |   top: "conv1_1"
 10 |   name: "conv1_1"
 11 |   type: "Convolution"
 12 |   convolution_param {
 13 |     num_output: 64
 14 |     pad: 1
 15 |     kernel_size: 3
 16 |   }
 17 | }
 18 | layer {
 19 |   bottom: "conv1_1"
 20 |   top: "conv1_1"
 21 |   name: "relu1_1"
 22 |   type: "ReLU"
 23 | }
 24 | layer {
 25 |   bottom: "conv1_1"
 26 |   top: "conv1_2"
 27 |   name: "conv1_2"
 28 |   type: "Convolution"
 29 |   convolution_param {
 30 |     num_output: 64
 31 |     pad: 1
 32 |     kernel_size: 3
 33 |   }
 34 | }
 35 | layer {
 36 |   bottom: "conv1_2"
 37 |   top: "conv1_2"
 38 |   name: "relu1_2"
 39 |   type: "ReLU"
 40 | }
 41 | layer {
 42 |   bottom: "conv1_2"
 43 |   top: "pool1"
 44 |   name: "pool1"
 45 |   type: "Pooling"
 46 |   pooling_param {
 47 |     pool: MAX
 48 |     kernel_size: 2
 49 |     stride: 2
 50 |   }
 51 | }
 52 | layer {
 53 |   bottom: "pool1"
 54 |   top: "conv2_1"
 55 |   name: "conv2_1"
 56 |   type: "Convolution"
 57 |   convolution_param {
 58 |     num_output: 128
 59 |     pad: 1
 60 |     kernel_size: 3
 61 |   }
 62 | }
 63 | layer {
 64 |   bottom: "conv2_1"
 65 |   top: "conv2_1"
 66 |   name: "relu2_1"
 67 |   type: "ReLU"
 68 | }
 69 | layer {
 70 |   bottom: "conv2_1"
 71 |   top: "conv2_2"
 72 |   name: "conv2_2"
 73 |   type: "Convolution"
 74 |   convolution_param {
 75 |     num_output: 128
 76 |     pad: 1
 77 |     kernel_size: 3
 78 |   }
 79 | }
 80 | layer {
 81 |   bottom: "conv2_2"
 82 |   top: "conv2_2"
 83 |   name: "relu2_2"
 84 |   type: "ReLU"
 85 | }
 86 | layer {
 87 |   bottom: "conv2_2"
 88 |   top: "pool2"
 89 |   name: "pool2"
 90 |   type: "Pooling"
 91 |   pooling_param {
 92 |     pool: MAX
 93 |     kernel_size: 2
 94 |     stride: 2
 95 |   }
 96 | }
 97 | layer {
 98 |   bottom: "pool2"
 99 |   top: "conv3_1"
100 |   name: "conv3_1"
101 |   type: "Convolution"
102 |   convolution_param {
103 |     num_output: 256
104 |     pad: 1
105 |     kernel_size: 3
106 |   }
107 | }
108 | layer {
109 |   bottom: "conv3_1"
110 |   top: "conv3_1"
111 |   name: "relu3_1"
112 |   type: "ReLU"
113 | }
114 | layer {
115 |   bottom: "conv3_1"
116 |   top: "conv3_2"
117 |   name: "conv3_2"
118 |   type: "Convolution"
119 |   convolution_param {
120 |     num_output: 256
121 |     pad: 1
122 |     kernel_size: 3
123 |   }
124 | }
125 | layer {
126 |   bottom: "conv3_2"
127 |   top: "conv3_2"
128 |   name: "relu3_2"
129 |   type: "ReLU"
130 | }
131 | layer {
132 |   bottom: "conv3_2"
133 |   top: "conv3_3"
134 |   name: "conv3_3"
135 |   type: "Convolution"
136 |   convolution_param {
137 |     num_output: 256
138 |     pad: 1
139 |     kernel_size: 3
140 |   }
141 | }
142 | layer {
143 |   bottom: "conv3_3"
144 |   top: "conv3_3"
145 |   name: "relu3_3"
146 |   type: "ReLU"
147 | }
148 | layer {
149 |   bottom: "conv3_3"
150 |   top: "pool3"
151 |   name: "pool3"
152 |   type: "Pooling"
153 |   pooling_param {
154 |     pool: MAX
155 |     kernel_size: 2
156 |     stride: 2
157 |   }
158 | }
159 | layer {
160 |   bottom: "pool3"
161 |   top: "conv4_1"
162 |   name: "conv4_1"
163 |   type: "Convolution"
164 |   convolution_param {
165 |     num_output: 512
166 |     pad: 1
167 |     kernel_size: 3
168 |   }
169 | }
170 | layer {
171 |   bottom: "conv4_1"
172 |   top: "conv4_1"
173 |   name: "relu4_1"
174 |   type: "ReLU"
175 | }
176 | layer {
177 |   bottom: "conv4_1"
178 |   top: "conv4_2"
179 |   name: "conv4_2"
180 |   type: "Convolution"
181 |   convolution_param {
182 |     num_output: 512
183 |     pad: 1
184 |     kernel_size: 3
185 |   }
186 | }
187 | layer {
188 |   bottom: "conv4_2"
189 |   top: "conv4_2"
190 |   name: "relu4_2"
191 |   type: "ReLU"
192 | }
193 | layer {
194 |   bottom: "conv4_2"
195 |   top: "conv4_3"
196 |   name: "conv4_3"
197 |   type: "Convolution"
198 |   convolution_param {
199 |     num_output: 512
200 |     pad: 1
201 |     kernel_size: 3
202 |   }
203 | }
204 | layer {
205 |   bottom: "conv4_3"
206 |   top: "conv4_3"
207 |   name: "relu4_3"
208 |   type: "ReLU"
209 | }
210 | layer {
211 |   bottom: "conv4_3"
212 |   top: "pool4"
213 |   name: "pool4"
214 |   type: "Pooling"
215 |   pooling_param {
216 |     pool: MAX
217 |     kernel_size: 2
218 |     stride: 2
219 |   }
220 | }
221 | layer {
222 |   bottom: "pool4"
223 |   top: "conv5_1"
224 |   name: "conv5_1"
225 |   type: "Convolution"
226 |   convolution_param {
227 |     num_output: 512
228 |     pad: 1
229 |     kernel_size: 3
230 |   }
231 | }
232 | layer {
233 |   bottom: "conv5_1"
234 |   top: "conv5_1"
235 |   name: "relu5_1"
236 |   type: "ReLU"
237 | }
238 | layer {
239 |   bottom: "conv5_1"
240 |   top: "conv5_2"
241 |   name: "conv5_2"
242 |   type: "Convolution"
243 |   convolution_param {
244 |     num_output: 512
245 |     pad: 1
246 |     kernel_size: 3
247 |   }
248 | }
249 | layer {
250 |   bottom: "conv5_2"
251 |   top: "conv5_2"
252 |   name: "relu5_2"
253 |   type: "ReLU"
254 | }
255 | layer {
256 |   bottom: "conv5_2"
257 |   top: "conv5_3"
258 |   name: "conv5_3"
259 |   type: "Convolution"
260 |   convolution_param {
261 |     num_output: 512
262 |     pad: 1
263 |     kernel_size: 3
264 |   }
265 | }
266 | layer {
267 |   bottom: "conv5_3"
268 |   top: "conv5_3"
269 |   name: "relu5_3"
270 |   type: "ReLU"
271 | }
272 | layer {
273 |   bottom: "conv5_3"
274 |   top: "pool5"
275 |   name: "pool5"
276 |   type: "Pooling"
277 |   pooling_param {
278 |     pool: MAX
279 |     kernel_size: 2
280 |     stride: 2
281 |   }
282 | }
283 | layer {
284 |   bottom: "pool5"
285 |   top: "fc6"
286 |   name: "fc6"
287 |   type: "InnerProduct"
288 |   inner_product_param {
289 |     num_output: 4096
290 |   }
291 | }
292 | layer {
293 |   bottom: "fc6"
294 |   top: "fc6"
295 |   name: "relu6"
296 |   type: "ReLU"
297 | }
298 | layer {
299 |   bottom: "fc6"
300 |   top: "fc6"
301 |   name: "drop6"
302 |   type: "Dropout"
303 |   dropout_param {
304 |     dropout_ratio: 0.5
305 |   }
306 | }
307 | layer {
308 |   bottom: "fc6"
309 |   top: "fc7"
310 |   name: "fc7"
311 |   type: "InnerProduct"
312 |   inner_product_param {
313 |     num_output: 4096
314 |   }
315 | }
316 | layer {
317 |   bottom: "fc7"
318 |   top: "fc7"
319 |   name: "relu7"
320 |   type: "ReLU"
321 | }
322 | layer {
323 |   bottom: "fc7"
324 |   top: "fc7"
325 |   name: "drop7"
326 |   type: "Dropout"
327 |   dropout_param {
328 |     dropout_ratio: 0.5
329 |   }
330 | }
331 | layer {
332 |   bottom: "fc7"
333 |   top: "my-fc8"
334 |   name: "my-fc8"
335 |   type: "InnerProduct"
336 |   inner_product_param {
337 |     num_output: 3
338 |   }
339 | }
340 | layer {
341 |   bottom: "my-fc8"
342 |   top: "prob"
343 |   name: "prob"
344 |   type: "Softmax"
345 | }
346 | 


--------------------------------------------------------------------------------
/VGG16/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 1000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 5000001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 20
26 | # save path
27 | snapshot_prefix: "vgg16"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/VGG16/train_val.prototxt:
--------------------------------------------------------------------------------
  1 | name: "VGG_ILSVRC_16_layer"
  2 | layer {
  3 |   name: "data"
  4 |   type: "Data"
  5 |   top: "data"
  6 |   top: "label"
  7 |   include {
  8 |     phase: TRAIN
  9 |   }
 10 |   transform_param {
 11 |     mirror: true
 12 |     crop_size: 224
 13 |     mean_file: "mean.binaryproto"
 14 |   }
 15 |   data_param {
 16 |     source: "train_lmdb"
 17 |     batch_size: 32
 18 |     backend: LMDB
 19 |   }
 20 | }
 21 | layer {
 22 |   name: "data"
 23 |   type: "Data"
 24 |   top: "data"
 25 |   top: "label"
 26 |   include {
 27 |     phase: TEST
 28 |   }
 29 |   transform_param {
 30 |     mirror: false
 31 |     crop_size: 224
 32 |     mean_file: "mean.binaryproto"
 33 |   }
 34 |   data_param {
 35 |     source: "val_lmdb"
 36 |     batch_size: 1
 37 |     backend: LMDB
 38 |   }
 39 | }
 40 | layer {
 41 |   bottom: "data"
 42 |   top: "conv1_1"
 43 |   name: "conv1_1"
 44 |   type: "Convolution"
 45 |   convolution_param {
 46 |     num_output: 64
 47 |     pad: 1
 48 |     kernel_size: 3
 49 |   }
 50 | }
 51 | layer {
 52 |   bottom: "conv1_1"
 53 |   top: "conv1_1"
 54 |   name: "relu1_1"
 55 |   type: "ReLU"
 56 | }
 57 | layer {
 58 |   bottom: "conv1_1"
 59 |   top: "conv1_2"
 60 |   name: "conv1_2"
 61 |   type: "Convolution"
 62 |   convolution_param {
 63 |     num_output: 64
 64 |     pad: 1
 65 |     kernel_size: 3
 66 |   }
 67 | }
 68 | layer {
 69 |   bottom: "conv1_2"
 70 |   top: "conv1_2"
 71 |   name: "relu1_2"
 72 |   type: "ReLU"
 73 | }
 74 | layer {
 75 |   bottom: "conv1_2"
 76 |   top: "pool1"
 77 |   name: "pool1"
 78 |   type: "Pooling"
 79 |   pooling_param {
 80 |     pool: MAX
 81 |     kernel_size: 2
 82 |     stride: 2
 83 |   }
 84 | }
 85 | layer {
 86 |   bottom: "pool1"
 87 |   top: "conv2_1"
 88 |   name: "conv2_1"
 89 |   type: "Convolution"
 90 |   convolution_param {
 91 |     num_output: 128
 92 |     pad: 1
 93 |     kernel_size: 3
 94 |   }
 95 | }
 96 | layer {
 97 |   bottom: "conv2_1"
 98 |   top: "conv2_1"
 99 |   name: "relu2_1"
100 |   type: "ReLU"
101 | }
102 | layer {
103 |   bottom: "conv2_1"
104 |   top: "conv2_2"
105 |   name: "conv2_2"
106 |   type: "Convolution"
107 |   convolution_param {
108 |     num_output: 128
109 |     pad: 1
110 |     kernel_size: 3
111 |   }
112 | }
113 | layer {
114 |   bottom: "conv2_2"
115 |   top: "conv2_2"
116 |   name: "relu2_2"
117 |   type: "ReLU"
118 | }
119 | layer {
120 |   bottom: "conv2_2"
121 |   top: "pool2"
122 |   name: "pool2"
123 |   type: "Pooling"
124 |   pooling_param {
125 |     pool: MAX
126 |     kernel_size: 2
127 |     stride: 2
128 |   }
129 | }
130 | layer {
131 |   bottom: "pool2"
132 |   top: "conv3_1"
133 |   name: "conv3_1"
134 |   type: "Convolution"
135 |   convolution_param {
136 |     num_output: 256
137 |     pad: 1
138 |     kernel_size: 3
139 |   }
140 | }
141 | layer {
142 |   bottom: "conv3_1"
143 |   top: "conv3_1"
144 |   name: "relu3_1"
145 |   type: "ReLU"
146 | }
147 | layer {
148 |   bottom: "conv3_1"
149 |   top: "conv3_2"
150 |   name: "conv3_2"
151 |   type: "Convolution"
152 |   convolution_param {
153 |     num_output: 256
154 |     pad: 1
155 |     kernel_size: 3
156 |   }
157 | }
158 | layer {
159 |   bottom: "conv3_2"
160 |   top: "conv3_2"
161 |   name: "relu3_2"
162 |   type: "ReLU"
163 | }
164 | layer {
165 |   bottom: "conv3_2"
166 |   top: "conv3_3"
167 |   name: "conv3_3"
168 |   type: "Convolution"
169 |   convolution_param {
170 |     num_output: 256
171 |     pad: 1
172 |     kernel_size: 3
173 |   }
174 | }
175 | layer {
176 |   bottom: "conv3_3"
177 |   top: "conv3_3"
178 |   name: "relu3_3"
179 |   type: "ReLU"
180 | }
181 | layer {
182 |   bottom: "conv3_3"
183 |   top: "pool3"
184 |   name: "pool3"
185 |   type: "Pooling"
186 |   pooling_param {
187 |     pool: MAX
188 |     kernel_size: 2
189 |     stride: 2
190 |   }
191 | }
192 | layer {
193 |   bottom: "pool3"
194 |   top: "conv4_1"
195 |   name: "conv4_1"
196 |   type: "Convolution"
197 |   convolution_param {
198 |     num_output: 512
199 |     pad: 1
200 |     kernel_size: 3
201 |   }
202 | }
203 | layer {
204 |   bottom: "conv4_1"
205 |   top: "conv4_1"
206 |   name: "relu4_1"
207 |   type: "ReLU"
208 | }
209 | layer {
210 |   bottom: "conv4_1"
211 |   top: "conv4_2"
212 |   name: "conv4_2"
213 |   type: "Convolution"
214 |   convolution_param {
215 |     num_output: 512
216 |     pad: 1
217 |     kernel_size: 3
218 |   }
219 | }
220 | layer {
221 |   bottom: "conv4_2"
222 |   top: "conv4_2"
223 |   name: "relu4_2"
224 |   type: "ReLU"
225 | }
226 | layer {
227 |   bottom: "conv4_2"
228 |   top: "conv4_3"
229 |   name: "conv4_3"
230 |   type: "Convolution"
231 |   convolution_param {
232 |     num_output: 512
233 |     pad: 1
234 |     kernel_size: 3
235 |   }
236 | }
237 | layer {
238 |   bottom: "conv4_3"
239 |   top: "conv4_3"
240 |   name: "relu4_3"
241 |   type: "ReLU"
242 | }
243 | layer {
244 |   bottom: "conv4_3"
245 |   top: "pool4"
246 |   name: "pool4"
247 |   type: "Pooling"
248 |   pooling_param {
249 |     pool: MAX
250 |     kernel_size: 2
251 |     stride: 2
252 |   }
253 | }
254 | layer {
255 |   bottom: "pool4"
256 |   top: "conv5_1"
257 |   name: "conv5_1"
258 |   type: "Convolution"
259 |   convolution_param {
260 |     num_output: 512
261 |     pad: 1
262 |     kernel_size: 3
263 |   }
264 | }
265 | layer {
266 |   bottom: "conv5_1"
267 |   top: "conv5_1"
268 |   name: "relu5_1"
269 |   type: "ReLU"
270 | }
271 | layer {
272 |   bottom: "conv5_1"
273 |   top: "conv5_2"
274 |   name: "conv5_2"
275 |   type: "Convolution"
276 |   convolution_param {
277 |     num_output: 512
278 |     pad: 1
279 |     kernel_size: 3
280 |   }
281 | }
282 | layer {
283 |   bottom: "conv5_2"
284 |   top: "conv5_2"
285 |   name: "relu5_2"
286 |   type: "ReLU"
287 | }
288 | layer {
289 |   bottom: "conv5_2"
290 |   top: "conv5_3"
291 |   name: "conv5_3"
292 |   type: "Convolution"
293 |   convolution_param {
294 |     num_output: 512
295 |     pad: 1
296 |     kernel_size: 3
297 |   }
298 | }
299 | layer {
300 |   bottom: "conv5_3"
301 |   top: "conv5_3"
302 |   name: "relu5_3"
303 |   type: "ReLU"
304 | }
305 | layer {
306 |   bottom: "conv5_3"
307 |   top: "pool5"
308 |   name: "pool5"
309 |   type: "Pooling"
310 |   pooling_param {
311 |     pool: MAX
312 |     kernel_size: 2
313 |     stride: 2
314 |   }
315 | }
316 | layer {
317 |   bottom: "pool5"
318 |   top: "fc6"
319 |   name: "fc6"
320 |   type: "InnerProduct"
321 |   inner_product_param {
322 |     num_output: 4096
323 |   }
324 | }
325 | layer {
326 |   bottom: "fc6"
327 |   top: "fc6"
328 |   name: "relu6"
329 |   type: "ReLU"
330 | }
331 | layer {
332 |   bottom: "fc6"
333 |   top: "fc6"
334 |   name: "drop6"
335 |   type: "Dropout"
336 |   dropout_param {
337 |     dropout_ratio: 0.5
338 |   }
339 | }
340 | layer {
341 |   bottom: "fc6"
342 |   top: "fc7"
343 |   name: "fc7"
344 |   type: "InnerProduct"
345 |   inner_product_param {
346 |     num_output: 4096
347 |   }
348 | }
349 | layer {
350 |   bottom: "fc7"
351 |   top: "fc7"
352 |   name: "relu7"
353 |   type: "ReLU"
354 | }
355 | layer {
356 |   bottom: "fc7"
357 |   top: "fc7"
358 |   name: "drop7"
359 |   type: "Dropout"
360 |   dropout_param {
361 |     dropout_ratio: 0.5
362 |   }
363 | }
364 | layer {
365 |   bottom: "fc7"
366 |   top: "my-fc8"
367 |   name: "my-fc8"
368 |   type: "InnerProduct"
369 |   inner_product_param {
370 |     num_output: 3
371 |   }
372 | }
373 | layer {
374 |   name: "loss1/loss1"
375 |   type: "SoftmaxWithLoss"
376 |   bottom: "my-fc8"
377 |   bottom: "label"
378 |   top: "loss1/loss1"
379 |   loss_weight: 0.3
380 | }
381 | layer {
382 |   name: "accuracy"
383 |   type: "Accuracy"
384 |   bottom: "my-fc8"
385 |   bottom: "label"
386 |   top: "accuracy"
387 |   include {
388 |     phase: TEST
389 |   }
390 | }
391 | 


--------------------------------------------------------------------------------
/VGG19/deploy.prototxt:
--------------------------------------------------------------------------------
  1 | name: "VGG_19"
  2 | input: "data"
  3 | input_shape {
  4 |   dim: 1
  5 |   dim: 3
  6 |   dim: 224
  7 |   dim: 224
  8 | }
  9 | layer {
 10 |   bottom: "data"
 11 |   top: "conv1_1"
 12 |   name: "conv1_1"
 13 |   type: "Convolution"
 14 |   convolution_param {
 15 |     num_output: 64
 16 |     pad: 1
 17 |     kernel_size: 3
 18 |   }
 19 | }
 20 | layer {
 21 |   bottom: "conv1_1"
 22 |   top: "conv1_1"
 23 |   name: "relu1_1"
 24 |   type: "ReLU"
 25 | }
 26 | layer {
 27 |   bottom: "conv1_1"
 28 |   top: "conv1_2"
 29 |   name: "conv1_2"
 30 |   type: "Convolution"
 31 |   convolution_param {
 32 |     num_output: 64
 33 |     pad: 1
 34 |     kernel_size: 3
 35 |   }
 36 | }
 37 | layer {
 38 |   bottom: "conv1_2"
 39 |   top: "conv1_2"
 40 |   name: "relu1_2"
 41 |   type: "ReLU"
 42 | }
 43 | layer {
 44 |   bottom: "conv1_2"
 45 |   top: "pool1"
 46 |   name: "pool1"
 47 |   type: "Pooling"
 48 |   pooling_param {
 49 |     pool: MAX
 50 |     kernel_size: 2
 51 |     stride: 2
 52 |   }
 53 | }
 54 | layer {
 55 |   bottom: "pool1"
 56 |   top: "conv2_1"
 57 |   name: "conv2_1"
 58 |   type: "Convolution"
 59 |   convolution_param {
 60 |     num_output: 128
 61 |     pad: 1
 62 |     kernel_size: 3
 63 |   }
 64 | }
 65 | layer {
 66 |   bottom: "conv2_1"
 67 |   top: "conv2_1"
 68 |   name: "relu2_1"
 69 |   type: "ReLU"
 70 | }
 71 | layer {
 72 |   bottom: "conv2_1"
 73 |   top: "conv2_2"
 74 |   name: "conv2_2"
 75 |   type: "Convolution"
 76 |   convolution_param {
 77 |     num_output: 128
 78 |     pad: 1
 79 |     kernel_size: 3
 80 |   }
 81 | }
 82 | layer {
 83 |   bottom: "conv2_2"
 84 |   top: "conv2_2"
 85 |   name: "relu2_2"
 86 |   type: "ReLU"
 87 | }
 88 | layer {
 89 |   bottom: "conv2_2"
 90 |   top: "pool2"
 91 |   name: "pool2"
 92 |   type: "Pooling"
 93 |   pooling_param {
 94 |     pool: MAX
 95 |     kernel_size: 2
 96 |     stride: 2
 97 |   }
 98 | }
 99 | layer {
100 |   bottom: "pool2"
101 |   top: "conv3_1"
102 |   name: "conv3_1"
103 |   type: "Convolution"
104 |   convolution_param {
105 |     num_output: 256
106 |     pad: 1
107 |     kernel_size: 3
108 |   }
109 | }
110 | layer {
111 |   bottom: "conv3_1"
112 |   top: "conv3_1"
113 |   name: "relu3_1"
114 |   type: "ReLU"
115 | }
116 | layer {
117 |   bottom: "conv3_1"
118 |   top: "conv3_2"
119 |   name: "conv3_2"
120 |   type: "Convolution"
121 |   convolution_param {
122 |     num_output: 256
123 |     pad: 1
124 |     kernel_size: 3
125 |   }
126 | }
127 | layer {
128 |   bottom: "conv3_2"
129 |   top: "conv3_2"
130 |   name: "relu3_2"
131 |   type: "ReLU"
132 | }
133 | layer {
134 |   bottom: "conv3_2"
135 |   top: "conv3_3"
136 |   name: "conv3_3"
137 |   type: "Convolution"
138 |   convolution_param {
139 |     num_output: 256
140 |     pad: 1
141 |     kernel_size: 3
142 |   }
143 | }
144 | layer {
145 |   bottom: "conv3_3"
146 |   top: "conv3_3"
147 |   name: "relu3_3"
148 |   type: "ReLU"
149 | }
150 | layer {
151 |   bottom: "conv3_3"
152 |   top: "conv3_4"
153 |   name: "conv3_4"
154 |   type: "Convolution"
155 |   convolution_param {
156 |     num_output: 256
157 |     pad: 1
158 |     kernel_size: 3
159 |   }
160 | }
161 | layer {
162 |   bottom: "conv3_4"
163 |   top: "conv3_4"
164 |   name: "relu3_4"
165 |   type: "ReLU"
166 | }
167 | layer {
168 |   bottom: "conv3_4"
169 |   top: "pool3"
170 |   name: "pool3"
171 |   type: "Pooling"
172 |   pooling_param {
173 |     pool: MAX
174 |     kernel_size: 2
175 |     stride: 2
176 |   }
177 | }
178 | layer {
179 |   bottom: "pool3"
180 |   top: "conv4_1"
181 |   name: "conv4_1"
182 |   type: "Convolution"
183 |   convolution_param {
184 |     num_output: 512
185 |     pad: 1
186 |     kernel_size: 3
187 |   }
188 | }
189 | layer {
190 |   bottom: "conv4_1"
191 |   top: "conv4_1"
192 |   name: "relu4_1"
193 |   type: "ReLU"
194 | }
195 | layer {
196 |   bottom: "conv4_1"
197 |   top: "conv4_2"
198 |   name: "conv4_2"
199 |   type: "Convolution"
200 |   convolution_param {
201 |     num_output: 512
202 |     pad: 1
203 |     kernel_size: 3
204 |   }
205 | }
206 | layer {
207 |   bottom: "conv4_2"
208 |   top: "conv4_2"
209 |   name: "relu4_2"
210 |   type: "ReLU"
211 | }
212 | layer {
213 |   bottom: "conv4_2"
214 |   top: "conv4_3"
215 |   name: "conv4_3"
216 |   type: "Convolution"
217 |   convolution_param {
218 |     num_output: 512
219 |     pad: 1
220 |     kernel_size: 3
221 |   }
222 | }
223 | layer {
224 |   bottom: "conv4_3"
225 |   top: "conv4_3"
226 |   name: "relu4_3"
227 |   type: "ReLU"
228 | }
229 | layer {
230 |   bottom: "conv4_3"
231 |   top: "conv4_4"
232 |   name: "conv4_4"
233 |   type: "Convolution"
234 |   convolution_param {
235 |     num_output: 512
236 |     pad: 1
237 |     kernel_size: 3
238 |   }
239 | }
240 | layer {
241 |   bottom: "conv4_4"
242 |   top: "conv4_4"
243 |   name: "relu4_4"
244 |   type: "ReLU"
245 | }
246 | layer {
247 |   bottom: "conv4_4"
248 |   top: "pool4"
249 |   name: "pool4"
250 |   type: "Pooling"
251 |   pooling_param {
252 |     pool: MAX
253 |     kernel_size: 2
254 |     stride: 2
255 |   }
256 | }
257 | layer {
258 |   bottom: "pool4"
259 |   top: "conv5_1"
260 |   name: "conv5_1"
261 |   type: "Convolution"
262 |   convolution_param {
263 |     num_output: 512
264 |     pad: 1
265 |     kernel_size: 3
266 |   }
267 | }
268 | layer {
269 |   bottom: "conv5_1"
270 |   top: "conv5_1"
271 |   name: "relu5_1"
272 |   type: "ReLU"
273 | }
274 | layer {
275 |   bottom: "conv5_1"
276 |   top: "conv5_2"
277 |   name: "conv5_2"
278 |   type: "Convolution"
279 |   convolution_param {
280 |     num_output: 512
281 |     pad: 1
282 |     kernel_size: 3
283 |   }
284 | }
285 | layer {
286 |   bottom: "conv5_2"
287 |   top: "conv5_2"
288 |   name: "relu5_2"
289 |   type: "ReLU"
290 | }
291 | layer {
292 |   bottom: "conv5_2"
293 |   top: "conv5_3"
294 |   name: "conv5_3"
295 |   type: "Convolution"
296 |   convolution_param {
297 |     num_output: 512
298 |     pad: 1
299 |     kernel_size: 3
300 |   }
301 | }
302 | layer {
303 |   bottom: "conv5_3"
304 |   top: "conv5_3"
305 |   name: "relu5_3"
306 |   type: "ReLU"
307 | }
308 | layer {
309 |   bottom: "conv5_3"
310 |   top: "conv5_4"
311 |   name: "conv5_4"
312 |   type: "Convolution"
313 |   convolution_param {
314 |     num_output: 512
315 |     pad: 1
316 |     kernel_size: 3
317 |   }
318 | }
319 | layer {
320 |   bottom: "conv5_4"
321 |   top: "conv5_4"
322 |   name: "relu5_4"
323 |   type: "ReLU"
324 | }
325 | layer {
326 |   bottom: "conv5_4"
327 |   top: "pool5"
328 |   name: "pool5"
329 |   type: "Pooling"
330 |   pooling_param {
331 |     pool: MAX
332 |     kernel_size: 2
333 |     stride: 2
334 |   }
335 | }
336 | layer {
337 |   bottom: "pool5"
338 |   top: "fc6"
339 |   name: "fc6"
340 |   type: "InnerProduct"
341 |   inner_product_param {
342 |     num_output: 4096
343 |   }
344 | }
345 | layer {
346 |   bottom: "fc6"
347 |   top: "fc6"
348 |   name: "relu6"
349 |   type: "ReLU"
350 | }
351 | layer {
352 |   bottom: "fc6"
353 |   top: "fc6"
354 |   name: "drop6"
355 |   type: "Dropout"
356 |   dropout_param {
357 |     dropout_ratio: 0.5
358 |   }
359 | }
360 | layer {
361 |   bottom: "fc6"
362 |   top: "fc7"
363 |   name: "fc7"
364 |   type: "InnerProduct"
365 |   inner_product_param {
366 |     num_output: 4096
367 |   }
368 | }
369 | layer {
370 |   bottom: "fc7"
371 |   top: "fc7"
372 |   name: "relu7"
373 |   type: "ReLU"
374 | }
375 | layer {
376 |   bottom: "fc7"
377 |   top: "fc7"
378 |   name: "drop7"
379 |   type: "Dropout"
380 |   dropout_param {
381 |     dropout_ratio: 0.5
382 |   }
383 | }
384 | layer {
385 |   bottom: "fc7"
386 |   top: "my-fc8"
387 |   name: "my-fc8"
388 |   type: "InnerProduct"
389 |   inner_product_param {
390 |     num_output: 3
391 |   }
392 | }
393 | layer {
394 |   name: "prob"
395 |   type: "Softmax"
396 |   bottom: "my-fc8"
397 |   top: "prob"
398 | }
399 | 


--------------------------------------------------------------------------------
/VGG19/solver.prototxt:
--------------------------------------------------------------------------------
 1 | # the definition of neural network model
 2 | net: "train_val.prototxt"
 3 | # test_iter is related to batch_size in test layer, test_iter * batch_size = the number of test data
 4 | test_iter: 761
 5 | # carry out test once every 5 training iterations
 6 | test_interval: 1000
 7 | # exclude test phase when test_initialization = false
 8 | # test_initialization: false
 9 | # display information once every 10 training iterations
10 | display: 1
11 | # 
12 | average_loss: 40
13 | # the initial learning rate
14 | base_lr: 0.001
15 | lr_policy: "poly"
16 | stepsize: 320000
17 | gamma: 0.96
18 | # The max number of iterations
19 | max_iter: 5000001
20 | power: 1.0
21 | momentum: 0.9
22 | # weight decay item, in case of overfitting
23 | weight_decay: 0.0002
24 | # save once every 50 training iterations
25 | snapshot: 1000
26 | # save path
27 | snapshot_prefix: "vgg19"
28 | solver_mode: GPU
29 | 


--------------------------------------------------------------------------------
/VGG19/train_val.prototxt:
--------------------------------------------------------------------------------
  1 | name: "VGG_19"
  2 | layer {
  3 |   name: "data"
  4 |   type: "Data"
  5 |   top: "data"
  6 |   top: "label"
  7 |   include {
  8 |     phase: TRAIN
  9 |   }
 10 |   transform_param {
 11 |     mirror: true
 12 |     crop_size: 224
 13 |     mean_file: "mean.binaryproto"
 14 |   }
 15 |   data_param {
 16 |     source: "train_lmdb"
 17 |     batch_size: 36
 18 |     backend: LMDB
 19 |   }
 20 | }
 21 | layer {
 22 |   name: "data"
 23 |   type: "Data"
 24 |   top: "data"
 25 |   top: "label"
 26 |   include {
 27 |     phase: TEST
 28 |   }
 29 |   transform_param {
 30 |     mirror: false
 31 |     crop_size: 224
 32 |     mean_file: "mean.binaryproto"
 33 |   }
 34 |   data_param {
 35 |     source: "val_lmdb"
 36 |     batch_size: 1
 37 |     backend: LMDB
 38 |   }
 39 | }
 40 | layer {
 41 |   bottom: "data"
 42 |   top: "conv1_1"
 43 |   name: "conv1_1"
 44 |   type: "Convolution"
 45 |   convolution_param {
 46 |     num_output: 64
 47 |     pad: 1
 48 |     kernel_size: 3
 49 |   }
 50 | }
 51 | layer {
 52 |   bottom: "conv1_1"
 53 |   top: "conv1_1"
 54 |   name: "relu1_1"
 55 |   type: "ReLU"
 56 | }
 57 | layer {
 58 |   bottom: "conv1_1"
 59 |   top: "conv1_2"
 60 |   name: "conv1_2"
 61 |   type: "Convolution"
 62 |   convolution_param {
 63 |     num_output: 64
 64 |     pad: 1
 65 |     kernel_size: 3
 66 |   }
 67 | }
 68 | layer {
 69 |   bottom: "conv1_2"
 70 |   top: "conv1_2"
 71 |   name: "relu1_2"
 72 |   type: "ReLU"
 73 | }
 74 | layer {
 75 |   bottom: "conv1_2"
 76 |   top: "pool1"
 77 |   name: "pool1"
 78 |   type: "Pooling"
 79 |   pooling_param {
 80 |     pool: MAX
 81 |     kernel_size: 2
 82 |     stride: 2
 83 |   }
 84 | }
 85 | layer {
 86 |   bottom: "pool1"
 87 |   top: "conv2_1"
 88 |   name: "conv2_1"
 89 |   type: "Convolution"
 90 |   convolution_param {
 91 |     num_output: 128
 92 |     pad: 1
 93 |     kernel_size: 3
 94 |   }
 95 | }
 96 | layer {
 97 |   bottom: "conv2_1"
 98 |   top: "conv2_1"
 99 |   name: "relu2_1"
100 |   type: "ReLU"
101 | }
102 | layer {
103 |   bottom: "conv2_1"
104 |   top: "conv2_2"
105 |   name: "conv2_2"
106 |   type: "Convolution"
107 |   convolution_param {
108 |     num_output: 128
109 |     pad: 1
110 |     kernel_size: 3
111 |   }
112 | }
113 | layer {
114 |   bottom: "conv2_2"
115 |   top: "conv2_2"
116 |   name: "relu2_2"
117 |   type: "ReLU"
118 | }
119 | layer {
120 |   bottom: "conv2_2"
121 |   top: "pool2"
122 |   name: "pool2"
123 |   type: "Pooling"
124 |   pooling_param {
125 |     pool: MAX
126 |     kernel_size: 2
127 |     stride: 2
128 |   }
129 | }
130 | layer {
131 |   bottom: "pool2"
132 |   top: "conv3_1"
133 |   name: "conv3_1"
134 |   type: "Convolution"
135 |   convolution_param {
136 |     num_output: 256
137 |     pad: 1
138 |     kernel_size: 3
139 |   }
140 | }
141 | layer {
142 |   bottom: "conv3_1"
143 |   top: "conv3_1"
144 |   name: "relu3_1"
145 |   type: "ReLU"
146 | }
147 | layer {
148 |   bottom: "conv3_1"
149 |   top: "conv3_2"
150 |   name: "conv3_2"
151 |   type: "Convolution"
152 |   convolution_param {
153 |     num_output: 256
154 |     pad: 1
155 |     kernel_size: 3
156 |   }
157 | }
158 | layer {
159 |   bottom: "conv3_2"
160 |   top: "conv3_2"
161 |   name: "relu3_2"
162 |   type: "ReLU"
163 | }
164 | layer {
165 |   bottom: "conv3_2"
166 |   top: "conv3_3"
167 |   name: "conv3_3"
168 |   type: "Convolution"
169 |   convolution_param {
170 |     num_output: 256
171 |     pad: 1
172 |     kernel_size: 3
173 |   }
174 | }
175 | layer {
176 |   bottom: "conv3_3"
177 |   top: "conv3_3"
178 |   name: "relu3_3"
179 |   type: "ReLU"
180 | }
181 | layer {
182 |   bottom: "conv3_3"
183 |   top: "conv3_4"
184 |   name: "conv3_4"
185 |   type: "Convolution"
186 |   convolution_param {
187 |     num_output: 256
188 |     pad: 1
189 |     kernel_size: 3
190 |   }
191 | }
192 | layer {
193 |   bottom: "conv3_4"
194 |   top: "conv3_4"
195 |   name: "relu3_4"
196 |   type: "ReLU"
197 | }
198 | layer {
199 |   bottom: "conv3_4"
200 |   top: "pool3"
201 |   name: "pool3"
202 |   type: "Pooling"
203 |   pooling_param {
204 |     pool: MAX
205 |     kernel_size: 2
206 |     stride: 2
207 |   }
208 | }
209 | layer {
210 |   bottom: "pool3"
211 |   top: "conv4_1"
212 |   name: "conv4_1"
213 |   type: "Convolution"
214 |   convolution_param {
215 |     num_output: 512
216 |     pad: 1
217 |     kernel_size: 3
218 |   }
219 | }
220 | layer {
221 |   bottom: "conv4_1"
222 |   top: "conv4_1"
223 |   name: "relu4_1"
224 |   type: "ReLU"
225 | }
226 | layer {
227 |   bottom: "conv4_1"
228 |   top: "conv4_2"
229 |   name: "conv4_2"
230 |   type: "Convolution"
231 |   convolution_param {
232 |     num_output: 512
233 |     pad: 1
234 |     kernel_size: 3
235 |   }
236 | }
237 | layer {
238 |   bottom: "conv4_2"
239 |   top: "conv4_2"
240 |   name: "relu4_2"
241 |   type: "ReLU"
242 | }
243 | layer {
244 |   bottom: "conv4_2"
245 |   top: "conv4_3"
246 |   name: "conv4_3"
247 |   type: "Convolution"
248 |   convolution_param {
249 |     num_output: 512
250 |     pad: 1
251 |     kernel_size: 3
252 |   }
253 | }
254 | layer {
255 |   bottom: "conv4_3"
256 |   top: "conv4_3"
257 |   name: "relu4_3"
258 |   type: "ReLU"
259 | }
260 | layer {
261 |   bottom: "conv4_3"
262 |   top: "conv4_4"
263 |   name: "conv4_4"
264 |   type: "Convolution"
265 |   convolution_param {
266 |     num_output: 512
267 |     pad: 1
268 |     kernel_size: 3
269 |   }
270 | }
271 | layer {
272 |   bottom: "conv4_4"
273 |   top: "conv4_4"
274 |   name: "relu4_4"
275 |   type: "ReLU"
276 | }
277 | layer {
278 |   bottom: "conv4_4"
279 |   top: "pool4"
280 |   name: "pool4"
281 |   type: "Pooling"
282 |   pooling_param {
283 |     pool: MAX
284 |     kernel_size: 2
285 |     stride: 2
286 |   }
287 | }
288 | layer {
289 |   bottom: "pool4"
290 |   top: "conv5_1"
291 |   name: "conv5_1"
292 |   type: "Convolution"
293 |   convolution_param {
294 |     num_output: 512
295 |     pad: 1
296 |     kernel_size: 3
297 |   }
298 | }
299 | layer {
300 |   bottom: "conv5_1"
301 |   top: "conv5_1"
302 |   name: "relu5_1"
303 |   type: "ReLU"
304 | }
305 | layer {
306 |   bottom: "conv5_1"
307 |   top: "conv5_2"
308 |   name: "conv5_2"
309 |   type: "Convolution"
310 |   convolution_param {
311 |     num_output: 512
312 |     pad: 1
313 |     kernel_size: 3
314 |   }
315 | }
316 | layer {
317 |   bottom: "conv5_2"
318 |   top: "conv5_2"
319 |   name: "relu5_2"
320 |   type: "ReLU"
321 | }
322 | layer {
323 |   bottom: "conv5_2"
324 |   top: "conv5_3"
325 |   name: "conv5_3"
326 |   type: "Convolution"
327 |   convolution_param {
328 |     num_output: 512
329 |     pad: 1
330 |     kernel_size: 3
331 |   }
332 | }
333 | layer {
334 |   bottom: "conv5_3"
335 |   top: "conv5_3"
336 |   name: "relu5_3"
337 |   type: "ReLU"
338 | }
339 | layer {
340 |   bottom: "conv5_3"
341 |   top: "conv5_4"
342 |   name: "conv5_4"
343 |   type: "Convolution"
344 |   convolution_param {
345 |     num_output: 512
346 |     pad: 1
347 |     kernel_size: 3
348 |   }
349 | }
350 | layer {
351 |   bottom: "conv5_4"
352 |   top: "conv5_4"
353 |   name: "relu5_4"
354 |   type: "ReLU"
355 | }
356 | layer {
357 |   bottom: "conv5_4"
358 |   top: "pool5"
359 |   name: "pool5"
360 |   type: "Pooling"
361 |   pooling_param {
362 |     pool: MAX
363 |     kernel_size: 2
364 |     stride: 2
365 |   }
366 | }
367 | layer {
368 |   bottom: "pool5"
369 |   top: "fc6"
370 |   name: "fc6"
371 |   type: "InnerProduct"
372 |   inner_product_param {
373 |     num_output: 4096
374 |   }
375 | }
376 | layer {
377 |   bottom: "fc6"
378 |   top: "fc6"
379 |   name: "relu6"
380 |   type: "ReLU"
381 | }
382 | layer {
383 |   bottom: "fc6"
384 |   top: "fc6"
385 |   name: "drop6"
386 |   type: "Dropout"
387 |   dropout_param {
388 |     dropout_ratio: 0.5
389 |   }
390 | }
391 | layer {
392 |   bottom: "fc6"
393 |   top: "fc7"
394 |   name: "fc7"
395 |   type: "InnerProduct"
396 |   inner_product_param {
397 |     num_output: 4096
398 |   }
399 | }
400 | layer {
401 |   bottom: "fc7"
402 |   top: "fc7"
403 |   name: "relu7"
404 |   type: "ReLU"
405 | }
406 | layer {
407 |   bottom: "fc7"
408 |   top: "fc7"
409 |   name: "drop7"
410 |   type: "Dropout"
411 |   dropout_param {
412 |     dropout_ratio: 0.5
413 |   }
414 | }
415 | layer {
416 |   bottom: "fc7"
417 |   top: "my-fc8"
418 |   name: "my-fc8"
419 |   type: "InnerProduct"
420 |   inner_product_param {
421 |     num_output: 3
422 |   }
423 | }
424 | layer {
425 |   name: "loss1/loss1"
426 |   type: "SoftmaxWithLoss"
427 |   bottom: "my-fc8"
428 |   bottom: "label"
429 |   top: "loss1/loss1"
430 |   loss_weight: 0.3
431 | }
432 | layer {
433 |   name: "accuracy"
434 |   type: "Accuracy"
435 |   bottom: "my-fc8"
436 |   bottom: "label"
437 |   top: "accuracy"
438 |   include {
439 |     phase: TEST
440 |   }
441 | }
442 | 


--------------------------------------------------------------------------------