├── README.md
├── matlab
    ├── change_input_blob.m
    ├── data
    │   ├── deploy_caffenet.prototxt
    │   ├── deploy_googlenet.prototxt
    │   ├── deploy_resnet.prototxt
    │   ├── deploy_vgg_16.prototxt
    │   ├── deploy_vgg_19.prototxt
    │   ├── deploy_vgg_f.prototxt
    │   ├── ilsvrc_2012_mean.mat
    │   ├── synset_words.txt
    │   └── test_img.png
    ├── demo_caffe.m
    ├── makeImagenetData.m
    ├── predict_caffe.m
    ├── predict_matconvnet.m
    ├── preprocess_img.m
    ├── readme.md
    ├── universal_perturbation.m
    └── vl_argparse.m
├── precomputed
    ├── CaffeNet.mat
    ├── GoogLeNet.mat
    ├── README.md
    ├── ResNet-152.mat
    ├── VGG-16.mat
    ├── VGG-19.mat
    └── VGG-F.mat
└── python
    ├── README.md
    ├── data
        ├── labels.txt
        ├── test_img.png
        └── universal.npy
    ├── deepfool.py
    ├── demo_inception.py
    ├── prepare_imagenet_data.py
    └── universal_pert.py


/README.md:
--------------------------------------------------------------------------------
 1 | # Universal adversarial perturbations
 2 | 
 3 | *matlab*: MATLAB code to generate universal perturbations, using [Caffe](https://github.com/BVLC/caffe) or [MatConvNet](https://github.com/vlfeat/matconvnet).
 4 | 
 5 | *python*: Python code to generate universal perturbations using [TensorFlow](https://github.com/tensorflow/tensorflow).
 6 | 
 7 | *precomputed*: Precomputed universal perturbations, for pre-trained models on ImageNet.
 8 | 
 9 | ## Reference
10 | [1] S. Moosavi-Dezfooli\*, A. Fawzi\*, O. Fawzi, P. Frossard:
11 | [*Universal adversarial perturbations*](http://arxiv.org/pdf/1610.08401), CVPR 2017
12 | 


--------------------------------------------------------------------------------
/matlab/change_input_blob.m:
--------------------------------------------------------------------------------
1 | function change_input_blob(net,n)
2 | blob_shape = net.blobs('data').shape();
3 | blob_shape(4) = n;
4 | net.blobs('data').reshape(blob_shape);
5 | 
6 | net.reshape();
7 | end
8 | 


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


--------------------------------------------------------------------------------
/matlab/data/deploy_googlenet.prototxt:
--------------------------------------------------------------------------------
   1 | force_backward: true
   2 | name: "GoogleNet"
   3 | layer {
   4 |   name: "data"
   5 |   type: "Input"
   6 |   top: "data"
   7 |   input_param { shape: { dim: 1 dim: 3 dim: 224 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 |       std: 0.1
  30 |     }
  31 |     bias_filler {
  32 |       type: "constant"
  33 |       value: 0.2
  34 |     }
  35 |   }
  36 | }
  37 | layer {
  38 |   name: "conv1/relu_7x7"
  39 |   type: "ReLU"
  40 |   bottom: "conv1/7x7_s2"
  41 |   top: "conv1/7x7_s2"
  42 | }
  43 | layer {
  44 |   name: "pool1/3x3_s2"
  45 |   type: "Pooling"
  46 |   bottom: "conv1/7x7_s2"
  47 |   top: "pool1/3x3_s2"
  48 |   pooling_param {
  49 |     pool: MAX
  50 |     kernel_size: 3
  51 |     stride: 2
  52 |   }
  53 | }
  54 | layer {
  55 |   name: "pool1/norm1"
  56 |   type: "LRN"
  57 |   bottom: "pool1/3x3_s2"
  58 |   top: "pool1/norm1"
  59 |   lrn_param {
  60 |     local_size: 5
  61 |     alpha: 0.0001
  62 |     beta: 0.75
  63 |   }
  64 | }
  65 | layer {
  66 |   name: "conv2/3x3_reduce"
  67 |   type: "Convolution"
  68 |   bottom: "pool1/norm1"
  69 |   top: "conv2/3x3_reduce"
  70 |   param {
  71 |     lr_mult: 1
  72 |     decay_mult: 1
  73 |   }
  74 |   param {
  75 |     lr_mult: 2
  76 |     decay_mult: 0
  77 |   }
  78 |   convolution_param {
  79 |     num_output: 64
  80 |     kernel_size: 1
  81 |     weight_filler {
  82 |       type: "xavier"
  83 |       std: 0.1
  84 |     }
  85 |     bias_filler {
  86 |       type: "constant"
  87 |       value: 0.2
  88 |     }
  89 |   }
  90 | }
  91 | layer {
  92 |   name: "conv2/relu_3x3_reduce"
  93 |   type: "ReLU"
  94 |   bottom: "conv2/3x3_reduce"
  95 |   top: "conv2/3x3_reduce"
  96 | }
  97 | layer {
  98 |   name: "conv2/3x3"
  99 |   type: "Convolution"
 100 |   bottom: "conv2/3x3_reduce"
 101 |   top: "conv2/3x3"
 102 |   param {
 103 |     lr_mult: 1
 104 |     decay_mult: 1
 105 |   }
 106 |   param {
 107 |     lr_mult: 2
 108 |     decay_mult: 0
 109 |   }
 110 |   convolution_param {
 111 |     num_output: 192
 112 |     pad: 1
 113 |     kernel_size: 3
 114 |     weight_filler {
 115 |       type: "xavier"
 116 |       std: 0.03
 117 |     }
 118 |     bias_filler {
 119 |       type: "constant"
 120 |       value: 0.2
 121 |     }
 122 |   }
 123 | }
 124 | layer {
 125 |   name: "conv2/relu_3x3"
 126 |   type: "ReLU"
 127 |   bottom: "conv2/3x3"
 128 |   top: "conv2/3x3"
 129 | }
 130 | layer {
 131 |   name: "conv2/norm2"
 132 |   type: "LRN"
 133 |   bottom: "conv2/3x3"
 134 |   top: "conv2/norm2"
 135 |   lrn_param {
 136 |     local_size: 5
 137 |     alpha: 0.0001
 138 |     beta: 0.75
 139 |   }
 140 | }
 141 | layer {
 142 |   name: "pool2/3x3_s2"
 143 |   type: "Pooling"
 144 |   bottom: "conv2/norm2"
 145 |   top: "pool2/3x3_s2"
 146 |   pooling_param {
 147 |     pool: MAX
 148 |     kernel_size: 3
 149 |     stride: 2
 150 |   }
 151 | }
 152 | layer {
 153 |   name: "inception_3a/1x1"
 154 |   type: "Convolution"
 155 |   bottom: "pool2/3x3_s2"
 156 |   top: "inception_3a/1x1"
 157 |   param {
 158 |     lr_mult: 1
 159 |     decay_mult: 1
 160 |   }
 161 |   param {
 162 |     lr_mult: 2
 163 |     decay_mult: 0
 164 |   }
 165 |   convolution_param {
 166 |     num_output: 64
 167 |     kernel_size: 1
 168 |     weight_filler {
 169 |       type: "xavier"
 170 |       std: 0.03
 171 |     }
 172 |     bias_filler {
 173 |       type: "constant"
 174 |       value: 0.2
 175 |     }
 176 |   }
 177 | }
 178 | layer {
 179 |   name: "inception_3a/relu_1x1"
 180 |   type: "ReLU"
 181 |   bottom: "inception_3a/1x1"
 182 |   top: "inception_3a/1x1"
 183 | }
 184 | layer {
 185 |   name: "inception_3a/3x3_reduce"
 186 |   type: "Convolution"
 187 |   bottom: "pool2/3x3_s2"
 188 |   top: "inception_3a/3x3_reduce"
 189 |   param {
 190 |     lr_mult: 1
 191 |     decay_mult: 1
 192 |   }
 193 |   param {
 194 |     lr_mult: 2
 195 |     decay_mult: 0
 196 |   }
 197 |   convolution_param {
 198 |     num_output: 96
 199 |     kernel_size: 1
 200 |     weight_filler {
 201 |       type: "xavier"
 202 |       std: 0.09
 203 |     }
 204 |     bias_filler {
 205 |       type: "constant"
 206 |       value: 0.2
 207 |     }
 208 |   }
 209 | }
 210 | layer {
 211 |   name: "inception_3a/relu_3x3_reduce"
 212 |   type: "ReLU"
 213 |   bottom: "inception_3a/3x3_reduce"
 214 |   top: "inception_3a/3x3_reduce"
 215 | }
 216 | layer {
 217 |   name: "inception_3a/3x3"
 218 |   type: "Convolution"
 219 |   bottom: "inception_3a/3x3_reduce"
 220 |   top: "inception_3a/3x3"
 221 |   param {
 222 |     lr_mult: 1
 223 |     decay_mult: 1
 224 |   }
 225 |   param {
 226 |     lr_mult: 2
 227 |     decay_mult: 0
 228 |   }
 229 |   convolution_param {
 230 |     num_output: 128
 231 |     pad: 1
 232 |     kernel_size: 3
 233 |     weight_filler {
 234 |       type: "xavier"
 235 |       std: 0.03
 236 |     }
 237 |     bias_filler {
 238 |       type: "constant"
 239 |       value: 0.2
 240 |     }
 241 |   }
 242 | }
 243 | layer {
 244 |   name: "inception_3a/relu_3x3"
 245 |   type: "ReLU"
 246 |   bottom: "inception_3a/3x3"
 247 |   top: "inception_3a/3x3"
 248 | }
 249 | layer {
 250 |   name: "inception_3a/5x5_reduce"
 251 |   type: "Convolution"
 252 |   bottom: "pool2/3x3_s2"
 253 |   top: "inception_3a/5x5_reduce"
 254 |   param {
 255 |     lr_mult: 1
 256 |     decay_mult: 1
 257 |   }
 258 |   param {
 259 |     lr_mult: 2
 260 |     decay_mult: 0
 261 |   }
 262 |   convolution_param {
 263 |     num_output: 16
 264 |     kernel_size: 1
 265 |     weight_filler {
 266 |       type: "xavier"
 267 |       std: 0.2
 268 |     }
 269 |     bias_filler {
 270 |       type: "constant"
 271 |       value: 0.2
 272 |     }
 273 |   }
 274 | }
 275 | layer {
 276 |   name: "inception_3a/relu_5x5_reduce"
 277 |   type: "ReLU"
 278 |   bottom: "inception_3a/5x5_reduce"
 279 |   top: "inception_3a/5x5_reduce"
 280 | }
 281 | layer {
 282 |   name: "inception_3a/5x5"
 283 |   type: "Convolution"
 284 |   bottom: "inception_3a/5x5_reduce"
 285 |   top: "inception_3a/5x5"
 286 |   param {
 287 |     lr_mult: 1
 288 |     decay_mult: 1
 289 |   }
 290 |   param {
 291 |     lr_mult: 2
 292 |     decay_mult: 0
 293 |   }
 294 |   convolution_param {
 295 |     num_output: 32
 296 |     pad: 2
 297 |     kernel_size: 5
 298 |     weight_filler {
 299 |       type: "xavier"
 300 |       std: 0.03
 301 |     }
 302 |     bias_filler {
 303 |       type: "constant"
 304 |       value: 0.2
 305 |     }
 306 |   }
 307 | }
 308 | layer {
 309 |   name: "inception_3a/relu_5x5"
 310 |   type: "ReLU"
 311 |   bottom: "inception_3a/5x5"
 312 |   top: "inception_3a/5x5"
 313 | }
 314 | layer {
 315 |   name: "inception_3a/pool"
 316 |   type: "Pooling"
 317 |   bottom: "pool2/3x3_s2"
 318 |   top: "inception_3a/pool"
 319 |   pooling_param {
 320 |     pool: MAX
 321 |     kernel_size: 3
 322 |     stride: 1
 323 |     pad: 1
 324 |   }
 325 | }
 326 | layer {
 327 |   name: "inception_3a/pool_proj"
 328 |   type: "Convolution"
 329 |   bottom: "inception_3a/pool"
 330 |   top: "inception_3a/pool_proj"
 331 |   param {
 332 |     lr_mult: 1
 333 |     decay_mult: 1
 334 |   }
 335 |   param {
 336 |     lr_mult: 2
 337 |     decay_mult: 0
 338 |   }
 339 |   convolution_param {
 340 |     num_output: 32
 341 |     kernel_size: 1
 342 |     weight_filler {
 343 |       type: "xavier"
 344 |       std: 0.1
 345 |     }
 346 |     bias_filler {
 347 |       type: "constant"
 348 |       value: 0.2
 349 |     }
 350 |   }
 351 | }
 352 | layer {
 353 |   name: "inception_3a/relu_pool_proj"
 354 |   type: "ReLU"
 355 |   bottom: "inception_3a/pool_proj"
 356 |   top: "inception_3a/pool_proj"
 357 | }
 358 | layer {
 359 |   name: "inception_3a/output"
 360 |   type: "Concat"
 361 |   bottom: "inception_3a/1x1"
 362 |   bottom: "inception_3a/3x3"
 363 |   bottom: "inception_3a/5x5"
 364 |   bottom: "inception_3a/pool_proj"
 365 |   top: "inception_3a/output"
 366 | }
 367 | layer {
 368 |   name: "inception_3b/1x1"
 369 |   type: "Convolution"
 370 |   bottom: "inception_3a/output"
 371 |   top: "inception_3b/1x1"
 372 |   param {
 373 |     lr_mult: 1
 374 |     decay_mult: 1
 375 |   }
 376 |   param {
 377 |     lr_mult: 2
 378 |     decay_mult: 0
 379 |   }
 380 |   convolution_param {
 381 |     num_output: 128
 382 |     kernel_size: 1
 383 |     weight_filler {
 384 |       type: "xavier"
 385 |       std: 0.03
 386 |     }
 387 |     bias_filler {
 388 |       type: "constant"
 389 |       value: 0.2
 390 |     }
 391 |   }
 392 | }
 393 | layer {
 394 |   name: "inception_3b/relu_1x1"
 395 |   type: "ReLU"
 396 |   bottom: "inception_3b/1x1"
 397 |   top: "inception_3b/1x1"
 398 | }
 399 | layer {
 400 |   name: "inception_3b/3x3_reduce"
 401 |   type: "Convolution"
 402 |   bottom: "inception_3a/output"
 403 |   top: "inception_3b/3x3_reduce"
 404 |   param {
 405 |     lr_mult: 1
 406 |     decay_mult: 1
 407 |   }
 408 |   param {
 409 |     lr_mult: 2
 410 |     decay_mult: 0
 411 |   }
 412 |   convolution_param {
 413 |     num_output: 128
 414 |     kernel_size: 1
 415 |     weight_filler {
 416 |       type: "xavier"
 417 |       std: 0.09
 418 |     }
 419 |     bias_filler {
 420 |       type: "constant"
 421 |       value: 0.2
 422 |     }
 423 |   }
 424 | }
 425 | layer {
 426 |   name: "inception_3b/relu_3x3_reduce"
 427 |   type: "ReLU"
 428 |   bottom: "inception_3b/3x3_reduce"
 429 |   top: "inception_3b/3x3_reduce"
 430 | }
 431 | layer {
 432 |   name: "inception_3b/3x3"
 433 |   type: "Convolution"
 434 |   bottom: "inception_3b/3x3_reduce"
 435 |   top: "inception_3b/3x3"
 436 |   param {
 437 |     lr_mult: 1
 438 |     decay_mult: 1
 439 |   }
 440 |   param {
 441 |     lr_mult: 2
 442 |     decay_mult: 0
 443 |   }
 444 |   convolution_param {
 445 |     num_output: 192
 446 |     pad: 1
 447 |     kernel_size: 3
 448 |     weight_filler {
 449 |       type: "xavier"
 450 |       std: 0.03
 451 |     }
 452 |     bias_filler {
 453 |       type: "constant"
 454 |       value: 0.2
 455 |     }
 456 |   }
 457 | }
 458 | layer {
 459 |   name: "inception_3b/relu_3x3"
 460 |   type: "ReLU"
 461 |   bottom: "inception_3b/3x3"
 462 |   top: "inception_3b/3x3"
 463 | }
 464 | layer {
 465 |   name: "inception_3b/5x5_reduce"
 466 |   type: "Convolution"
 467 |   bottom: "inception_3a/output"
 468 |   top: "inception_3b/5x5_reduce"
 469 |   param {
 470 |     lr_mult: 1
 471 |     decay_mult: 1
 472 |   }
 473 |   param {
 474 |     lr_mult: 2
 475 |     decay_mult: 0
 476 |   }
 477 |   convolution_param {
 478 |     num_output: 32
 479 |     kernel_size: 1
 480 |     weight_filler {
 481 |       type: "xavier"
 482 |       std: 0.2
 483 |     }
 484 |     bias_filler {
 485 |       type: "constant"
 486 |       value: 0.2
 487 |     }
 488 |   }
 489 | }
 490 | layer {
 491 |   name: "inception_3b/relu_5x5_reduce"
 492 |   type: "ReLU"
 493 |   bottom: "inception_3b/5x5_reduce"
 494 |   top: "inception_3b/5x5_reduce"
 495 | }
 496 | layer {
 497 |   name: "inception_3b/5x5"
 498 |   type: "Convolution"
 499 |   bottom: "inception_3b/5x5_reduce"
 500 |   top: "inception_3b/5x5"
 501 |   param {
 502 |     lr_mult: 1
 503 |     decay_mult: 1
 504 |   }
 505 |   param {
 506 |     lr_mult: 2
 507 |     decay_mult: 0
 508 |   }
 509 |   convolution_param {
 510 |     num_output: 96
 511 |     pad: 2
 512 |     kernel_size: 5
 513 |     weight_filler {
 514 |       type: "xavier"
 515 |       std: 0.03
 516 |     }
 517 |     bias_filler {
 518 |       type: "constant"
 519 |       value: 0.2
 520 |     }
 521 |   }
 522 | }
 523 | layer {
 524 |   name: "inception_3b/relu_5x5"
 525 |   type: "ReLU"
 526 |   bottom: "inception_3b/5x5"
 527 |   top: "inception_3b/5x5"
 528 | }
 529 | layer {
 530 |   name: "inception_3b/pool"
 531 |   type: "Pooling"
 532 |   bottom: "inception_3a/output"
 533 |   top: "inception_3b/pool"
 534 |   pooling_param {
 535 |     pool: MAX
 536 |     kernel_size: 3
 537 |     stride: 1
 538 |     pad: 1
 539 |   }
 540 | }
 541 | layer {
 542 |   name: "inception_3b/pool_proj"
 543 |   type: "Convolution"
 544 |   bottom: "inception_3b/pool"
 545 |   top: "inception_3b/pool_proj"
 546 |   param {
 547 |     lr_mult: 1
 548 |     decay_mult: 1
 549 |   }
 550 |   param {
 551 |     lr_mult: 2
 552 |     decay_mult: 0
 553 |   }
 554 |   convolution_param {
 555 |     num_output: 64
 556 |     kernel_size: 1
 557 |     weight_filler {
 558 |       type: "xavier"
 559 |       std: 0.1
 560 |     }
 561 |     bias_filler {
 562 |       type: "constant"
 563 |       value: 0.2
 564 |     }
 565 |   }
 566 | }
 567 | layer {
 568 |   name: "inception_3b/relu_pool_proj"
 569 |   type: "ReLU"
 570 |   bottom: "inception_3b/pool_proj"
 571 |   top: "inception_3b/pool_proj"
 572 | }
 573 | layer {
 574 |   name: "inception_3b/output"
 575 |   type: "Concat"
 576 |   bottom: "inception_3b/1x1"
 577 |   bottom: "inception_3b/3x3"
 578 |   bottom: "inception_3b/5x5"
 579 |   bottom: "inception_3b/pool_proj"
 580 |   top: "inception_3b/output"
 581 | }
 582 | layer {
 583 |   name: "pool3/3x3_s2"
 584 |   type: "Pooling"
 585 |   bottom: "inception_3b/output"
 586 |   top: "pool3/3x3_s2"
 587 |   pooling_param {
 588 |     pool: MAX
 589 |     kernel_size: 3
 590 |     stride: 2
 591 |   }
 592 | }
 593 | layer {
 594 |   name: "inception_4a/1x1"
 595 |   type: "Convolution"
 596 |   bottom: "pool3/3x3_s2"
 597 |   top: "inception_4a/1x1"
 598 |   param {
 599 |     lr_mult: 1
 600 |     decay_mult: 1
 601 |   }
 602 |   param {
 603 |     lr_mult: 2
 604 |     decay_mult: 0
 605 |   }
 606 |   convolution_param {
 607 |     num_output: 192
 608 |     kernel_size: 1
 609 |     weight_filler {
 610 |       type: "xavier"
 611 |       std: 0.03
 612 |     }
 613 |     bias_filler {
 614 |       type: "constant"
 615 |       value: 0.2
 616 |     }
 617 |   }
 618 | }
 619 | layer {
 620 |   name: "inception_4a/relu_1x1"
 621 |   type: "ReLU"
 622 |   bottom: "inception_4a/1x1"
 623 |   top: "inception_4a/1x1"
 624 | }
 625 | layer {
 626 |   name: "inception_4a/3x3_reduce"
 627 |   type: "Convolution"
 628 |   bottom: "pool3/3x3_s2"
 629 |   top: "inception_4a/3x3_reduce"
 630 |   param {
 631 |     lr_mult: 1
 632 |     decay_mult: 1
 633 |   }
 634 |   param {
 635 |     lr_mult: 2
 636 |     decay_mult: 0
 637 |   }
 638 |   convolution_param {
 639 |     num_output: 96
 640 |     kernel_size: 1
 641 |     weight_filler {
 642 |       type: "xavier"
 643 |       std: 0.09
 644 |     }
 645 |     bias_filler {
 646 |       type: "constant"
 647 |       value: 0.2
 648 |     }
 649 |   }
 650 | }
 651 | layer {
 652 |   name: "inception_4a/relu_3x3_reduce"
 653 |   type: "ReLU"
 654 |   bottom: "inception_4a/3x3_reduce"
 655 |   top: "inception_4a/3x3_reduce"
 656 | }
 657 | layer {
 658 |   name: "inception_4a/3x3"
 659 |   type: "Convolution"
 660 |   bottom: "inception_4a/3x3_reduce"
 661 |   top: "inception_4a/3x3"
 662 |   param {
 663 |     lr_mult: 1
 664 |     decay_mult: 1
 665 |   }
 666 |   param {
 667 |     lr_mult: 2
 668 |     decay_mult: 0
 669 |   }
 670 |   convolution_param {
 671 |     num_output: 208
 672 |     pad: 1
 673 |     kernel_size: 3
 674 |     weight_filler {
 675 |       type: "xavier"
 676 |       std: 0.03
 677 |     }
 678 |     bias_filler {
 679 |       type: "constant"
 680 |       value: 0.2
 681 |     }
 682 |   }
 683 | }
 684 | layer {
 685 |   name: "inception_4a/relu_3x3"
 686 |   type: "ReLU"
 687 |   bottom: "inception_4a/3x3"
 688 |   top: "inception_4a/3x3"
 689 | }
 690 | layer {
 691 |   name: "inception_4a/5x5_reduce"
 692 |   type: "Convolution"
 693 |   bottom: "pool3/3x3_s2"
 694 |   top: "inception_4a/5x5_reduce"
 695 |   param {
 696 |     lr_mult: 1
 697 |     decay_mult: 1
 698 |   }
 699 |   param {
 700 |     lr_mult: 2
 701 |     decay_mult: 0
 702 |   }
 703 |   convolution_param {
 704 |     num_output: 16
 705 |     kernel_size: 1
 706 |     weight_filler {
 707 |       type: "xavier"
 708 |       std: 0.2
 709 |     }
 710 |     bias_filler {
 711 |       type: "constant"
 712 |       value: 0.2
 713 |     }
 714 |   }
 715 | }
 716 | layer {
 717 |   name: "inception_4a/relu_5x5_reduce"
 718 |   type: "ReLU"
 719 |   bottom: "inception_4a/5x5_reduce"
 720 |   top: "inception_4a/5x5_reduce"
 721 | }
 722 | layer {
 723 |   name: "inception_4a/5x5"
 724 |   type: "Convolution"
 725 |   bottom: "inception_4a/5x5_reduce"
 726 |   top: "inception_4a/5x5"
 727 |   param {
 728 |     lr_mult: 1
 729 |     decay_mult: 1
 730 |   }
 731 |   param {
 732 |     lr_mult: 2
 733 |     decay_mult: 0
 734 |   }
 735 |   convolution_param {
 736 |     num_output: 48
 737 |     pad: 2
 738 |     kernel_size: 5
 739 |     weight_filler {
 740 |       type: "xavier"
 741 |       std: 0.03
 742 |     }
 743 |     bias_filler {
 744 |       type: "constant"
 745 |       value: 0.2
 746 |     }
 747 |   }
 748 | }
 749 | layer {
 750 |   name: "inception_4a/relu_5x5"
 751 |   type: "ReLU"
 752 |   bottom: "inception_4a/5x5"
 753 |   top: "inception_4a/5x5"
 754 | }
 755 | layer {
 756 |   name: "inception_4a/pool"
 757 |   type: "Pooling"
 758 |   bottom: "pool3/3x3_s2"
 759 |   top: "inception_4a/pool"
 760 |   pooling_param {
 761 |     pool: MAX
 762 |     kernel_size: 3
 763 |     stride: 1
 764 |     pad: 1
 765 |   }
 766 | }
 767 | layer {
 768 |   name: "inception_4a/pool_proj"
 769 |   type: "Convolution"
 770 |   bottom: "inception_4a/pool"
 771 |   top: "inception_4a/pool_proj"
 772 |   param {
 773 |     lr_mult: 1
 774 |     decay_mult: 1
 775 |   }
 776 |   param {
 777 |     lr_mult: 2
 778 |     decay_mult: 0
 779 |   }
 780 |   convolution_param {
 781 |     num_output: 64
 782 |     kernel_size: 1
 783 |     weight_filler {
 784 |       type: "xavier"
 785 |       std: 0.1
 786 |     }
 787 |     bias_filler {
 788 |       type: "constant"
 789 |       value: 0.2
 790 |     }
 791 |   }
 792 | }
 793 | layer {
 794 |   name: "inception_4a/relu_pool_proj"
 795 |   type: "ReLU"
 796 |   bottom: "inception_4a/pool_proj"
 797 |   top: "inception_4a/pool_proj"
 798 | }
 799 | layer {
 800 |   name: "inception_4a/output"
 801 |   type: "Concat"
 802 |   bottom: "inception_4a/1x1"
 803 |   bottom: "inception_4a/3x3"
 804 |   bottom: "inception_4a/5x5"
 805 |   bottom: "inception_4a/pool_proj"
 806 |   top: "inception_4a/output"
 807 | }
 808 | layer {
 809 |   name: "inception_4b/1x1"
 810 |   type: "Convolution"
 811 |   bottom: "inception_4a/output"
 812 |   top: "inception_4b/1x1"
 813 |   param {
 814 |     lr_mult: 1
 815 |     decay_mult: 1
 816 |   }
 817 |   param {
 818 |     lr_mult: 2
 819 |     decay_mult: 0
 820 |   }
 821 |   convolution_param {
 822 |     num_output: 160
 823 |     kernel_size: 1
 824 |     weight_filler {
 825 |       type: "xavier"
 826 |       std: 0.03
 827 |     }
 828 |     bias_filler {
 829 |       type: "constant"
 830 |       value: 0.2
 831 |     }
 832 |   }
 833 | }
 834 | layer {
 835 |   name: "inception_4b/relu_1x1"
 836 |   type: "ReLU"
 837 |   bottom: "inception_4b/1x1"
 838 |   top: "inception_4b/1x1"
 839 | }
 840 | layer {
 841 |   name: "inception_4b/3x3_reduce"
 842 |   type: "Convolution"
 843 |   bottom: "inception_4a/output"
 844 |   top: "inception_4b/3x3_reduce"
 845 |   param {
 846 |     lr_mult: 1
 847 |     decay_mult: 1
 848 |   }
 849 |   param {
 850 |     lr_mult: 2
 851 |     decay_mult: 0
 852 |   }
 853 |   convolution_param {
 854 |     num_output: 112
 855 |     kernel_size: 1
 856 |     weight_filler {
 857 |       type: "xavier"
 858 |       std: 0.09
 859 |     }
 860 |     bias_filler {
 861 |       type: "constant"
 862 |       value: 0.2
 863 |     }
 864 |   }
 865 | }
 866 | layer {
 867 |   name: "inception_4b/relu_3x3_reduce"
 868 |   type: "ReLU"
 869 |   bottom: "inception_4b/3x3_reduce"
 870 |   top: "inception_4b/3x3_reduce"
 871 | }
 872 | layer {
 873 |   name: "inception_4b/3x3"
 874 |   type: "Convolution"
 875 |   bottom: "inception_4b/3x3_reduce"
 876 |   top: "inception_4b/3x3"
 877 |   param {
 878 |     lr_mult: 1
 879 |     decay_mult: 1
 880 |   }
 881 |   param {
 882 |     lr_mult: 2
 883 |     decay_mult: 0
 884 |   }
 885 |   convolution_param {
 886 |     num_output: 224
 887 |     pad: 1
 888 |     kernel_size: 3
 889 |     weight_filler {
 890 |       type: "xavier"
 891 |       std: 0.03
 892 |     }
 893 |     bias_filler {
 894 |       type: "constant"
 895 |       value: 0.2
 896 |     }
 897 |   }
 898 | }
 899 | layer {
 900 |   name: "inception_4b/relu_3x3"
 901 |   type: "ReLU"
 902 |   bottom: "inception_4b/3x3"
 903 |   top: "inception_4b/3x3"
 904 | }
 905 | layer {
 906 |   name: "inception_4b/5x5_reduce"
 907 |   type: "Convolution"
 908 |   bottom: "inception_4a/output"
 909 |   top: "inception_4b/5x5_reduce"
 910 |   param {
 911 |     lr_mult: 1
 912 |     decay_mult: 1
 913 |   }
 914 |   param {
 915 |     lr_mult: 2
 916 |     decay_mult: 0
 917 |   }
 918 |   convolution_param {
 919 |     num_output: 24
 920 |     kernel_size: 1
 921 |     weight_filler {
 922 |       type: "xavier"
 923 |       std: 0.2
 924 |     }
 925 |     bias_filler {
 926 |       type: "constant"
 927 |       value: 0.2
 928 |     }
 929 |   }
 930 | }
 931 | layer {
 932 |   name: "inception_4b/relu_5x5_reduce"
 933 |   type: "ReLU"
 934 |   bottom: "inception_4b/5x5_reduce"
 935 |   top: "inception_4b/5x5_reduce"
 936 | }
 937 | layer {
 938 |   name: "inception_4b/5x5"
 939 |   type: "Convolution"
 940 |   bottom: "inception_4b/5x5_reduce"
 941 |   top: "inception_4b/5x5"
 942 |   param {
 943 |     lr_mult: 1
 944 |     decay_mult: 1
 945 |   }
 946 |   param {
 947 |     lr_mult: 2
 948 |     decay_mult: 0
 949 |   }
 950 |   convolution_param {
 951 |     num_output: 64
 952 |     pad: 2
 953 |     kernel_size: 5
 954 |     weight_filler {
 955 |       type: "xavier"
 956 |       std: 0.03
 957 |     }
 958 |     bias_filler {
 959 |       type: "constant"
 960 |       value: 0.2
 961 |     }
 962 |   }
 963 | }
 964 | layer {
 965 |   name: "inception_4b/relu_5x5"
 966 |   type: "ReLU"
 967 |   bottom: "inception_4b/5x5"
 968 |   top: "inception_4b/5x5"
 969 | }
 970 | layer {
 971 |   name: "inception_4b/pool"
 972 |   type: "Pooling"
 973 |   bottom: "inception_4a/output"
 974 |   top: "inception_4b/pool"
 975 |   pooling_param {
 976 |     pool: MAX
 977 |     kernel_size: 3
 978 |     stride: 1
 979 |     pad: 1
 980 |   }
 981 | }
 982 | layer {
 983 |   name: "inception_4b/pool_proj"
 984 |   type: "Convolution"
 985 |   bottom: "inception_4b/pool"
 986 |   top: "inception_4b/pool_proj"
 987 |   param {
 988 |     lr_mult: 1
 989 |     decay_mult: 1
 990 |   }
 991 |   param {
 992 |     lr_mult: 2
 993 |     decay_mult: 0
 994 |   }
 995 |   convolution_param {
 996 |     num_output: 64
 997 |     kernel_size: 1
 998 |     weight_filler {
 999 |       type: "xavier"
1000 |       std: 0.1
1001 |     }
1002 |     bias_filler {
1003 |       type: "constant"
1004 |       value: 0.2
1005 |     }
1006 |   }
1007 | }
1008 | layer {
1009 |   name: "inception_4b/relu_pool_proj"
1010 |   type: "ReLU"
1011 |   bottom: "inception_4b/pool_proj"
1012 |   top: "inception_4b/pool_proj"
1013 | }
1014 | layer {
1015 |   name: "inception_4b/output"
1016 |   type: "Concat"
1017 |   bottom: "inception_4b/1x1"
1018 |   bottom: "inception_4b/3x3"
1019 |   bottom: "inception_4b/5x5"
1020 |   bottom: "inception_4b/pool_proj"
1021 |   top: "inception_4b/output"
1022 | }
1023 | layer {
1024 |   name: "inception_4c/1x1"
1025 |   type: "Convolution"
1026 |   bottom: "inception_4b/output"
1027 |   top: "inception_4c/1x1"
1028 |   param {
1029 |     lr_mult: 1
1030 |     decay_mult: 1
1031 |   }
1032 |   param {
1033 |     lr_mult: 2
1034 |     decay_mult: 0
1035 |   }
1036 |   convolution_param {
1037 |     num_output: 128
1038 |     kernel_size: 1
1039 |     weight_filler {
1040 |       type: "xavier"
1041 |       std: 0.03
1042 |     }
1043 |     bias_filler {
1044 |       type: "constant"
1045 |       value: 0.2
1046 |     }
1047 |   }
1048 | }
1049 | layer {
1050 |   name: "inception_4c/relu_1x1"
1051 |   type: "ReLU"
1052 |   bottom: "inception_4c/1x1"
1053 |   top: "inception_4c/1x1"
1054 | }
1055 | layer {
1056 |   name: "inception_4c/3x3_reduce"
1057 |   type: "Convolution"
1058 |   bottom: "inception_4b/output"
1059 |   top: "inception_4c/3x3_reduce"
1060 |   param {
1061 |     lr_mult: 1
1062 |     decay_mult: 1
1063 |   }
1064 |   param {
1065 |     lr_mult: 2
1066 |     decay_mult: 0
1067 |   }
1068 |   convolution_param {
1069 |     num_output: 128
1070 |     kernel_size: 1
1071 |     weight_filler {
1072 |       type: "xavier"
1073 |       std: 0.09
1074 |     }
1075 |     bias_filler {
1076 |       type: "constant"
1077 |       value: 0.2
1078 |     }
1079 |   }
1080 | }
1081 | layer {
1082 |   name: "inception_4c/relu_3x3_reduce"
1083 |   type: "ReLU"
1084 |   bottom: "inception_4c/3x3_reduce"
1085 |   top: "inception_4c/3x3_reduce"
1086 | }
1087 | layer {
1088 |   name: "inception_4c/3x3"
1089 |   type: "Convolution"
1090 |   bottom: "inception_4c/3x3_reduce"
1091 |   top: "inception_4c/3x3"
1092 |   param {
1093 |     lr_mult: 1
1094 |     decay_mult: 1
1095 |   }
1096 |   param {
1097 |     lr_mult: 2
1098 |     decay_mult: 0
1099 |   }
1100 |   convolution_param {
1101 |     num_output: 256
1102 |     pad: 1
1103 |     kernel_size: 3
1104 |     weight_filler {
1105 |       type: "xavier"
1106 |       std: 0.03
1107 |     }
1108 |     bias_filler {
1109 |       type: "constant"
1110 |       value: 0.2
1111 |     }
1112 |   }
1113 | }
1114 | layer {
1115 |   name: "inception_4c/relu_3x3"
1116 |   type: "ReLU"
1117 |   bottom: "inception_4c/3x3"
1118 |   top: "inception_4c/3x3"
1119 | }
1120 | layer {
1121 |   name: "inception_4c/5x5_reduce"
1122 |   type: "Convolution"
1123 |   bottom: "inception_4b/output"
1124 |   top: "inception_4c/5x5_reduce"
1125 |   param {
1126 |     lr_mult: 1
1127 |     decay_mult: 1
1128 |   }
1129 |   param {
1130 |     lr_mult: 2
1131 |     decay_mult: 0
1132 |   }
1133 |   convolution_param {
1134 |     num_output: 24
1135 |     kernel_size: 1
1136 |     weight_filler {
1137 |       type: "xavier"
1138 |       std: 0.2
1139 |     }
1140 |     bias_filler {
1141 |       type: "constant"
1142 |       value: 0.2
1143 |     }
1144 |   }
1145 | }
1146 | layer {
1147 |   name: "inception_4c/relu_5x5_reduce"
1148 |   type: "ReLU"
1149 |   bottom: "inception_4c/5x5_reduce"
1150 |   top: "inception_4c/5x5_reduce"
1151 | }
1152 | layer {
1153 |   name: "inception_4c/5x5"
1154 |   type: "Convolution"
1155 |   bottom: "inception_4c/5x5_reduce"
1156 |   top: "inception_4c/5x5"
1157 |   param {
1158 |     lr_mult: 1
1159 |     decay_mult: 1
1160 |   }
1161 |   param {
1162 |     lr_mult: 2
1163 |     decay_mult: 0
1164 |   }
1165 |   convolution_param {
1166 |     num_output: 64
1167 |     pad: 2
1168 |     kernel_size: 5
1169 |     weight_filler {
1170 |       type: "xavier"
1171 |       std: 0.03
1172 |     }
1173 |     bias_filler {
1174 |       type: "constant"
1175 |       value: 0.2
1176 |     }
1177 |   }
1178 | }
1179 | layer {
1180 |   name: "inception_4c/relu_5x5"
1181 |   type: "ReLU"
1182 |   bottom: "inception_4c/5x5"
1183 |   top: "inception_4c/5x5"
1184 | }
1185 | layer {
1186 |   name: "inception_4c/pool"
1187 |   type: "Pooling"
1188 |   bottom: "inception_4b/output"
1189 |   top: "inception_4c/pool"
1190 |   pooling_param {
1191 |     pool: MAX
1192 |     kernel_size: 3
1193 |     stride: 1
1194 |     pad: 1
1195 |   }
1196 | }
1197 | layer {
1198 |   name: "inception_4c/pool_proj"
1199 |   type: "Convolution"
1200 |   bottom: "inception_4c/pool"
1201 |   top: "inception_4c/pool_proj"
1202 |   param {
1203 |     lr_mult: 1
1204 |     decay_mult: 1
1205 |   }
1206 |   param {
1207 |     lr_mult: 2
1208 |     decay_mult: 0
1209 |   }
1210 |   convolution_param {
1211 |     num_output: 64
1212 |     kernel_size: 1
1213 |     weight_filler {
1214 |       type: "xavier"
1215 |       std: 0.1
1216 |     }
1217 |     bias_filler {
1218 |       type: "constant"
1219 |       value: 0.2
1220 |     }
1221 |   }
1222 | }
1223 | layer {
1224 |   name: "inception_4c/relu_pool_proj"
1225 |   type: "ReLU"
1226 |   bottom: "inception_4c/pool_proj"
1227 |   top: "inception_4c/pool_proj"
1228 | }
1229 | layer {
1230 |   name: "inception_4c/output"
1231 |   type: "Concat"
1232 |   bottom: "inception_4c/1x1"
1233 |   bottom: "inception_4c/3x3"
1234 |   bottom: "inception_4c/5x5"
1235 |   bottom: "inception_4c/pool_proj"
1236 |   top: "inception_4c/output"
1237 | }
1238 | layer {
1239 |   name: "inception_4d/1x1"
1240 |   type: "Convolution"
1241 |   bottom: "inception_4c/output"
1242 |   top: "inception_4d/1x1"
1243 |   param {
1244 |     lr_mult: 1
1245 |     decay_mult: 1
1246 |   }
1247 |   param {
1248 |     lr_mult: 2
1249 |     decay_mult: 0
1250 |   }
1251 |   convolution_param {
1252 |     num_output: 112
1253 |     kernel_size: 1
1254 |     weight_filler {
1255 |       type: "xavier"
1256 |       std: 0.03
1257 |     }
1258 |     bias_filler {
1259 |       type: "constant"
1260 |       value: 0.2
1261 |     }
1262 |   }
1263 | }
1264 | layer {
1265 |   name: "inception_4d/relu_1x1"
1266 |   type: "ReLU"
1267 |   bottom: "inception_4d/1x1"
1268 |   top: "inception_4d/1x1"
1269 | }
1270 | layer {
1271 |   name: "inception_4d/3x3_reduce"
1272 |   type: "Convolution"
1273 |   bottom: "inception_4c/output"
1274 |   top: "inception_4d/3x3_reduce"
1275 |   param {
1276 |     lr_mult: 1
1277 |     decay_mult: 1
1278 |   }
1279 |   param {
1280 |     lr_mult: 2
1281 |     decay_mult: 0
1282 |   }
1283 |   convolution_param {
1284 |     num_output: 144
1285 |     kernel_size: 1
1286 |     weight_filler {
1287 |       type: "xavier"
1288 |       std: 0.09
1289 |     }
1290 |     bias_filler {
1291 |       type: "constant"
1292 |       value: 0.2
1293 |     }
1294 |   }
1295 | }
1296 | layer {
1297 |   name: "inception_4d/relu_3x3_reduce"
1298 |   type: "ReLU"
1299 |   bottom: "inception_4d/3x3_reduce"
1300 |   top: "inception_4d/3x3_reduce"
1301 | }
1302 | layer {
1303 |   name: "inception_4d/3x3"
1304 |   type: "Convolution"
1305 |   bottom: "inception_4d/3x3_reduce"
1306 |   top: "inception_4d/3x3"
1307 |   param {
1308 |     lr_mult: 1
1309 |     decay_mult: 1
1310 |   }
1311 |   param {
1312 |     lr_mult: 2
1313 |     decay_mult: 0
1314 |   }
1315 |   convolution_param {
1316 |     num_output: 288
1317 |     pad: 1
1318 |     kernel_size: 3
1319 |     weight_filler {
1320 |       type: "xavier"
1321 |       std: 0.03
1322 |     }
1323 |     bias_filler {
1324 |       type: "constant"
1325 |       value: 0.2
1326 |     }
1327 |   }
1328 | }
1329 | layer {
1330 |   name: "inception_4d/relu_3x3"
1331 |   type: "ReLU"
1332 |   bottom: "inception_4d/3x3"
1333 |   top: "inception_4d/3x3"
1334 | }
1335 | layer {
1336 |   name: "inception_4d/5x5_reduce"
1337 |   type: "Convolution"
1338 |   bottom: "inception_4c/output"
1339 |   top: "inception_4d/5x5_reduce"
1340 |   param {
1341 |     lr_mult: 1
1342 |     decay_mult: 1
1343 |   }
1344 |   param {
1345 |     lr_mult: 2
1346 |     decay_mult: 0
1347 |   }
1348 |   convolution_param {
1349 |     num_output: 32
1350 |     kernel_size: 1
1351 |     weight_filler {
1352 |       type: "xavier"
1353 |       std: 0.2
1354 |     }
1355 |     bias_filler {
1356 |       type: "constant"
1357 |       value: 0.2
1358 |     }
1359 |   }
1360 | }
1361 | layer {
1362 |   name: "inception_4d/relu_5x5_reduce"
1363 |   type: "ReLU"
1364 |   bottom: "inception_4d/5x5_reduce"
1365 |   top: "inception_4d/5x5_reduce"
1366 | }
1367 | layer {
1368 |   name: "inception_4d/5x5"
1369 |   type: "Convolution"
1370 |   bottom: "inception_4d/5x5_reduce"
1371 |   top: "inception_4d/5x5"
1372 |   param {
1373 |     lr_mult: 1
1374 |     decay_mult: 1
1375 |   }
1376 |   param {
1377 |     lr_mult: 2
1378 |     decay_mult: 0
1379 |   }
1380 |   convolution_param {
1381 |     num_output: 64
1382 |     pad: 2
1383 |     kernel_size: 5
1384 |     weight_filler {
1385 |       type: "xavier"
1386 |       std: 0.03
1387 |     }
1388 |     bias_filler {
1389 |       type: "constant"
1390 |       value: 0.2
1391 |     }
1392 |   }
1393 | }
1394 | layer {
1395 |   name: "inception_4d/relu_5x5"
1396 |   type: "ReLU"
1397 |   bottom: "inception_4d/5x5"
1398 |   top: "inception_4d/5x5"
1399 | }
1400 | layer {
1401 |   name: "inception_4d/pool"
1402 |   type: "Pooling"
1403 |   bottom: "inception_4c/output"
1404 |   top: "inception_4d/pool"
1405 |   pooling_param {
1406 |     pool: MAX
1407 |     kernel_size: 3
1408 |     stride: 1
1409 |     pad: 1
1410 |   }
1411 | }
1412 | layer {
1413 |   name: "inception_4d/pool_proj"
1414 |   type: "Convolution"
1415 |   bottom: "inception_4d/pool"
1416 |   top: "inception_4d/pool_proj"
1417 |   param {
1418 |     lr_mult: 1
1419 |     decay_mult: 1
1420 |   }
1421 |   param {
1422 |     lr_mult: 2
1423 |     decay_mult: 0
1424 |   }
1425 |   convolution_param {
1426 |     num_output: 64
1427 |     kernel_size: 1
1428 |     weight_filler {
1429 |       type: "xavier"
1430 |       std: 0.1
1431 |     }
1432 |     bias_filler {
1433 |       type: "constant"
1434 |       value: 0.2
1435 |     }
1436 |   }
1437 | }
1438 | layer {
1439 |   name: "inception_4d/relu_pool_proj"
1440 |   type: "ReLU"
1441 |   bottom: "inception_4d/pool_proj"
1442 |   top: "inception_4d/pool_proj"
1443 | }
1444 | layer {
1445 |   name: "inception_4d/output"
1446 |   type: "Concat"
1447 |   bottom: "inception_4d/1x1"
1448 |   bottom: "inception_4d/3x3"
1449 |   bottom: "inception_4d/5x5"
1450 |   bottom: "inception_4d/pool_proj"
1451 |   top: "inception_4d/output"
1452 | }
1453 | layer {
1454 |   name: "inception_4e/1x1"
1455 |   type: "Convolution"
1456 |   bottom: "inception_4d/output"
1457 |   top: "inception_4e/1x1"
1458 |   param {
1459 |     lr_mult: 1
1460 |     decay_mult: 1
1461 |   }
1462 |   param {
1463 |     lr_mult: 2
1464 |     decay_mult: 0
1465 |   }
1466 |   convolution_param {
1467 |     num_output: 256
1468 |     kernel_size: 1
1469 |     weight_filler {
1470 |       type: "xavier"
1471 |       std: 0.03
1472 |     }
1473 |     bias_filler {
1474 |       type: "constant"
1475 |       value: 0.2
1476 |     }
1477 |   }
1478 | }
1479 | layer {
1480 |   name: "inception_4e/relu_1x1"
1481 |   type: "ReLU"
1482 |   bottom: "inception_4e/1x1"
1483 |   top: "inception_4e/1x1"
1484 | }
1485 | layer {
1486 |   name: "inception_4e/3x3_reduce"
1487 |   type: "Convolution"
1488 |   bottom: "inception_4d/output"
1489 |   top: "inception_4e/3x3_reduce"
1490 |   param {
1491 |     lr_mult: 1
1492 |     decay_mult: 1
1493 |   }
1494 |   param {
1495 |     lr_mult: 2
1496 |     decay_mult: 0
1497 |   }
1498 |   convolution_param {
1499 |     num_output: 160
1500 |     kernel_size: 1
1501 |     weight_filler {
1502 |       type: "xavier"
1503 |       std: 0.09
1504 |     }
1505 |     bias_filler {
1506 |       type: "constant"
1507 |       value: 0.2
1508 |     }
1509 |   }
1510 | }
1511 | layer {
1512 |   name: "inception_4e/relu_3x3_reduce"
1513 |   type: "ReLU"
1514 |   bottom: "inception_4e/3x3_reduce"
1515 |   top: "inception_4e/3x3_reduce"
1516 | }
1517 | layer {
1518 |   name: "inception_4e/3x3"
1519 |   type: "Convolution"
1520 |   bottom: "inception_4e/3x3_reduce"
1521 |   top: "inception_4e/3x3"
1522 |   param {
1523 |     lr_mult: 1
1524 |     decay_mult: 1
1525 |   }
1526 |   param {
1527 |     lr_mult: 2
1528 |     decay_mult: 0
1529 |   }
1530 |   convolution_param {
1531 |     num_output: 320
1532 |     pad: 1
1533 |     kernel_size: 3
1534 |     weight_filler {
1535 |       type: "xavier"
1536 |       std: 0.03
1537 |     }
1538 |     bias_filler {
1539 |       type: "constant"
1540 |       value: 0.2
1541 |     }
1542 |   }
1543 | }
1544 | layer {
1545 |   name: "inception_4e/relu_3x3"
1546 |   type: "ReLU"
1547 |   bottom: "inception_4e/3x3"
1548 |   top: "inception_4e/3x3"
1549 | }
1550 | layer {
1551 |   name: "inception_4e/5x5_reduce"
1552 |   type: "Convolution"
1553 |   bottom: "inception_4d/output"
1554 |   top: "inception_4e/5x5_reduce"
1555 |   param {
1556 |     lr_mult: 1
1557 |     decay_mult: 1
1558 |   }
1559 |   param {
1560 |     lr_mult: 2
1561 |     decay_mult: 0
1562 |   }
1563 |   convolution_param {
1564 |     num_output: 32
1565 |     kernel_size: 1
1566 |     weight_filler {
1567 |       type: "xavier"
1568 |       std: 0.2
1569 |     }
1570 |     bias_filler {
1571 |       type: "constant"
1572 |       value: 0.2
1573 |     }
1574 |   }
1575 | }
1576 | layer {
1577 |   name: "inception_4e/relu_5x5_reduce"
1578 |   type: "ReLU"
1579 |   bottom: "inception_4e/5x5_reduce"
1580 |   top: "inception_4e/5x5_reduce"
1581 | }
1582 | layer {
1583 |   name: "inception_4e/5x5"
1584 |   type: "Convolution"
1585 |   bottom: "inception_4e/5x5_reduce"
1586 |   top: "inception_4e/5x5"
1587 |   param {
1588 |     lr_mult: 1
1589 |     decay_mult: 1
1590 |   }
1591 |   param {
1592 |     lr_mult: 2
1593 |     decay_mult: 0
1594 |   }
1595 |   convolution_param {
1596 |     num_output: 128
1597 |     pad: 2
1598 |     kernel_size: 5
1599 |     weight_filler {
1600 |       type: "xavier"
1601 |       std: 0.03
1602 |     }
1603 |     bias_filler {
1604 |       type: "constant"
1605 |       value: 0.2
1606 |     }
1607 |   }
1608 | }
1609 | layer {
1610 |   name: "inception_4e/relu_5x5"
1611 |   type: "ReLU"
1612 |   bottom: "inception_4e/5x5"
1613 |   top: "inception_4e/5x5"
1614 | }
1615 | layer {
1616 |   name: "inception_4e/pool"
1617 |   type: "Pooling"
1618 |   bottom: "inception_4d/output"
1619 |   top: "inception_4e/pool"
1620 |   pooling_param {
1621 |     pool: MAX
1622 |     kernel_size: 3
1623 |     stride: 1
1624 |     pad: 1
1625 |   }
1626 | }
1627 | layer {
1628 |   name: "inception_4e/pool_proj"
1629 |   type: "Convolution"
1630 |   bottom: "inception_4e/pool"
1631 |   top: "inception_4e/pool_proj"
1632 |   param {
1633 |     lr_mult: 1
1634 |     decay_mult: 1
1635 |   }
1636 |   param {
1637 |     lr_mult: 2
1638 |     decay_mult: 0
1639 |   }
1640 |   convolution_param {
1641 |     num_output: 128
1642 |     kernel_size: 1
1643 |     weight_filler {
1644 |       type: "xavier"
1645 |       std: 0.1
1646 |     }
1647 |     bias_filler {
1648 |       type: "constant"
1649 |       value: 0.2
1650 |     }
1651 |   }
1652 | }
1653 | layer {
1654 |   name: "inception_4e/relu_pool_proj"
1655 |   type: "ReLU"
1656 |   bottom: "inception_4e/pool_proj"
1657 |   top: "inception_4e/pool_proj"
1658 | }
1659 | layer {
1660 |   name: "inception_4e/output"
1661 |   type: "Concat"
1662 |   bottom: "inception_4e/1x1"
1663 |   bottom: "inception_4e/3x3"
1664 |   bottom: "inception_4e/5x5"
1665 |   bottom: "inception_4e/pool_proj"
1666 |   top: "inception_4e/output"
1667 | }
1668 | layer {
1669 |   name: "pool4/3x3_s2"
1670 |   type: "Pooling"
1671 |   bottom: "inception_4e/output"
1672 |   top: "pool4/3x3_s2"
1673 |   pooling_param {
1674 |     pool: MAX
1675 |     kernel_size: 3
1676 |     stride: 2
1677 |   }
1678 | }
1679 | layer {
1680 |   name: "inception_5a/1x1"
1681 |   type: "Convolution"
1682 |   bottom: "pool4/3x3_s2"
1683 |   top: "inception_5a/1x1"
1684 |   param {
1685 |     lr_mult: 1
1686 |     decay_mult: 1
1687 |   }
1688 |   param {
1689 |     lr_mult: 2
1690 |     decay_mult: 0
1691 |   }
1692 |   convolution_param {
1693 |     num_output: 256
1694 |     kernel_size: 1
1695 |     weight_filler {
1696 |       type: "xavier"
1697 |       std: 0.03
1698 |     }
1699 |     bias_filler {
1700 |       type: "constant"
1701 |       value: 0.2
1702 |     }
1703 |   }
1704 | }
1705 | layer {
1706 |   name: "inception_5a/relu_1x1"
1707 |   type: "ReLU"
1708 |   bottom: "inception_5a/1x1"
1709 |   top: "inception_5a/1x1"
1710 | }
1711 | layer {
1712 |   name: "inception_5a/3x3_reduce"
1713 |   type: "Convolution"
1714 |   bottom: "pool4/3x3_s2"
1715 |   top: "inception_5a/3x3_reduce"
1716 |   param {
1717 |     lr_mult: 1
1718 |     decay_mult: 1
1719 |   }
1720 |   param {
1721 |     lr_mult: 2
1722 |     decay_mult: 0
1723 |   }
1724 |   convolution_param {
1725 |     num_output: 160
1726 |     kernel_size: 1
1727 |     weight_filler {
1728 |       type: "xavier"
1729 |       std: 0.09
1730 |     }
1731 |     bias_filler {
1732 |       type: "constant"
1733 |       value: 0.2
1734 |     }
1735 |   }
1736 | }
1737 | layer {
1738 |   name: "inception_5a/relu_3x3_reduce"
1739 |   type: "ReLU"
1740 |   bottom: "inception_5a/3x3_reduce"
1741 |   top: "inception_5a/3x3_reduce"
1742 | }
1743 | layer {
1744 |   name: "inception_5a/3x3"
1745 |   type: "Convolution"
1746 |   bottom: "inception_5a/3x3_reduce"
1747 |   top: "inception_5a/3x3"
1748 |   param {
1749 |     lr_mult: 1
1750 |     decay_mult: 1
1751 |   }
1752 |   param {
1753 |     lr_mult: 2
1754 |     decay_mult: 0
1755 |   }
1756 |   convolution_param {
1757 |     num_output: 320
1758 |     pad: 1
1759 |     kernel_size: 3
1760 |     weight_filler {
1761 |       type: "xavier"
1762 |       std: 0.03
1763 |     }
1764 |     bias_filler {
1765 |       type: "constant"
1766 |       value: 0.2
1767 |     }
1768 |   }
1769 | }
1770 | layer {
1771 |   name: "inception_5a/relu_3x3"
1772 |   type: "ReLU"
1773 |   bottom: "inception_5a/3x3"
1774 |   top: "inception_5a/3x3"
1775 | }
1776 | layer {
1777 |   name: "inception_5a/5x5_reduce"
1778 |   type: "Convolution"
1779 |   bottom: "pool4/3x3_s2"
1780 |   top: "inception_5a/5x5_reduce"
1781 |   param {
1782 |     lr_mult: 1
1783 |     decay_mult: 1
1784 |   }
1785 |   param {
1786 |     lr_mult: 2
1787 |     decay_mult: 0
1788 |   }
1789 |   convolution_param {
1790 |     num_output: 32
1791 |     kernel_size: 1
1792 |     weight_filler {
1793 |       type: "xavier"
1794 |       std: 0.2
1795 |     }
1796 |     bias_filler {
1797 |       type: "constant"
1798 |       value: 0.2
1799 |     }
1800 |   }
1801 | }
1802 | layer {
1803 |   name: "inception_5a/relu_5x5_reduce"
1804 |   type: "ReLU"
1805 |   bottom: "inception_5a/5x5_reduce"
1806 |   top: "inception_5a/5x5_reduce"
1807 | }
1808 | layer {
1809 |   name: "inception_5a/5x5"
1810 |   type: "Convolution"
1811 |   bottom: "inception_5a/5x5_reduce"
1812 |   top: "inception_5a/5x5"
1813 |   param {
1814 |     lr_mult: 1
1815 |     decay_mult: 1
1816 |   }
1817 |   param {
1818 |     lr_mult: 2
1819 |     decay_mult: 0
1820 |   }
1821 |   convolution_param {
1822 |     num_output: 128
1823 |     pad: 2
1824 |     kernel_size: 5
1825 |     weight_filler {
1826 |       type: "xavier"
1827 |       std: 0.03
1828 |     }
1829 |     bias_filler {
1830 |       type: "constant"
1831 |       value: 0.2
1832 |     }
1833 |   }
1834 | }
1835 | layer {
1836 |   name: "inception_5a/relu_5x5"
1837 |   type: "ReLU"
1838 |   bottom: "inception_5a/5x5"
1839 |   top: "inception_5a/5x5"
1840 | }
1841 | layer {
1842 |   name: "inception_5a/pool"
1843 |   type: "Pooling"
1844 |   bottom: "pool4/3x3_s2"
1845 |   top: "inception_5a/pool"
1846 |   pooling_param {
1847 |     pool: MAX
1848 |     kernel_size: 3
1849 |     stride: 1
1850 |     pad: 1
1851 |   }
1852 | }
1853 | layer {
1854 |   name: "inception_5a/pool_proj"
1855 |   type: "Convolution"
1856 |   bottom: "inception_5a/pool"
1857 |   top: "inception_5a/pool_proj"
1858 |   param {
1859 |     lr_mult: 1
1860 |     decay_mult: 1
1861 |   }
1862 |   param {
1863 |     lr_mult: 2
1864 |     decay_mult: 0
1865 |   }
1866 |   convolution_param {
1867 |     num_output: 128
1868 |     kernel_size: 1
1869 |     weight_filler {
1870 |       type: "xavier"
1871 |       std: 0.1
1872 |     }
1873 |     bias_filler {
1874 |       type: "constant"
1875 |       value: 0.2
1876 |     }
1877 |   }
1878 | }
1879 | layer {
1880 |   name: "inception_5a/relu_pool_proj"
1881 |   type: "ReLU"
1882 |   bottom: "inception_5a/pool_proj"
1883 |   top: "inception_5a/pool_proj"
1884 | }
1885 | layer {
1886 |   name: "inception_5a/output"
1887 |   type: "Concat"
1888 |   bottom: "inception_5a/1x1"
1889 |   bottom: "inception_5a/3x3"
1890 |   bottom: "inception_5a/5x5"
1891 |   bottom: "inception_5a/pool_proj"
1892 |   top: "inception_5a/output"
1893 | }
1894 | layer {
1895 |   name: "inception_5b/1x1"
1896 |   type: "Convolution"
1897 |   bottom: "inception_5a/output"
1898 |   top: "inception_5b/1x1"
1899 |   param {
1900 |     lr_mult: 1
1901 |     decay_mult: 1
1902 |   }
1903 |   param {
1904 |     lr_mult: 2
1905 |     decay_mult: 0
1906 |   }
1907 |   convolution_param {
1908 |     num_output: 384
1909 |     kernel_size: 1
1910 |     weight_filler {
1911 |       type: "xavier"
1912 |       std: 0.03
1913 |     }
1914 |     bias_filler {
1915 |       type: "constant"
1916 |       value: 0.2
1917 |     }
1918 |   }
1919 | }
1920 | layer {
1921 |   name: "inception_5b/relu_1x1"
1922 |   type: "ReLU"
1923 |   bottom: "inception_5b/1x1"
1924 |   top: "inception_5b/1x1"
1925 | }
1926 | layer {
1927 |   name: "inception_5b/3x3_reduce"
1928 |   type: "Convolution"
1929 |   bottom: "inception_5a/output"
1930 |   top: "inception_5b/3x3_reduce"
1931 |   param {
1932 |     lr_mult: 1
1933 |     decay_mult: 1
1934 |   }
1935 |   param {
1936 |     lr_mult: 2
1937 |     decay_mult: 0
1938 |   }
1939 |   convolution_param {
1940 |     num_output: 192
1941 |     kernel_size: 1
1942 |     weight_filler {
1943 |       type: "xavier"
1944 |       std: 0.09
1945 |     }
1946 |     bias_filler {
1947 |       type: "constant"
1948 |       value: 0.2
1949 |     }
1950 |   }
1951 | }
1952 | layer {
1953 |   name: "inception_5b/relu_3x3_reduce"
1954 |   type: "ReLU"
1955 |   bottom: "inception_5b/3x3_reduce"
1956 |   top: "inception_5b/3x3_reduce"
1957 | }
1958 | layer {
1959 |   name: "inception_5b/3x3"
1960 |   type: "Convolution"
1961 |   bottom: "inception_5b/3x3_reduce"
1962 |   top: "inception_5b/3x3"
1963 |   param {
1964 |     lr_mult: 1
1965 |     decay_mult: 1
1966 |   }
1967 |   param {
1968 |     lr_mult: 2
1969 |     decay_mult: 0
1970 |   }
1971 |   convolution_param {
1972 |     num_output: 384
1973 |     pad: 1
1974 |     kernel_size: 3
1975 |     weight_filler {
1976 |       type: "xavier"
1977 |       std: 0.03
1978 |     }
1979 |     bias_filler {
1980 |       type: "constant"
1981 |       value: 0.2
1982 |     }
1983 |   }
1984 | }
1985 | layer {
1986 |   name: "inception_5b/relu_3x3"
1987 |   type: "ReLU"
1988 |   bottom: "inception_5b/3x3"
1989 |   top: "inception_5b/3x3"
1990 | }
1991 | layer {
1992 |   name: "inception_5b/5x5_reduce"
1993 |   type: "Convolution"
1994 |   bottom: "inception_5a/output"
1995 |   top: "inception_5b/5x5_reduce"
1996 |   param {
1997 |     lr_mult: 1
1998 |     decay_mult: 1
1999 |   }
2000 |   param {
2001 |     lr_mult: 2
2002 |     decay_mult: 0
2003 |   }
2004 |   convolution_param {
2005 |     num_output: 48
2006 |     kernel_size: 1
2007 |     weight_filler {
2008 |       type: "xavier"
2009 |       std: 0.2
2010 |     }
2011 |     bias_filler {
2012 |       type: "constant"
2013 |       value: 0.2
2014 |     }
2015 |   }
2016 | }
2017 | layer {
2018 |   name: "inception_5b/relu_5x5_reduce"
2019 |   type: "ReLU"
2020 |   bottom: "inception_5b/5x5_reduce"
2021 |   top: "inception_5b/5x5_reduce"
2022 | }
2023 | layer {
2024 |   name: "inception_5b/5x5"
2025 |   type: "Convolution"
2026 |   bottom: "inception_5b/5x5_reduce"
2027 |   top: "inception_5b/5x5"
2028 |   param {
2029 |     lr_mult: 1
2030 |     decay_mult: 1
2031 |   }
2032 |   param {
2033 |     lr_mult: 2
2034 |     decay_mult: 0
2035 |   }
2036 |   convolution_param {
2037 |     num_output: 128
2038 |     pad: 2
2039 |     kernel_size: 5
2040 |     weight_filler {
2041 |       type: "xavier"
2042 |       std: 0.03
2043 |     }
2044 |     bias_filler {
2045 |       type: "constant"
2046 |       value: 0.2
2047 |     }
2048 |   }
2049 | }
2050 | layer {
2051 |   name: "inception_5b/relu_5x5"
2052 |   type: "ReLU"
2053 |   bottom: "inception_5b/5x5"
2054 |   top: "inception_5b/5x5"
2055 | }
2056 | layer {
2057 |   name: "inception_5b/pool"
2058 |   type: "Pooling"
2059 |   bottom: "inception_5a/output"
2060 |   top: "inception_5b/pool"
2061 |   pooling_param {
2062 |     pool: MAX
2063 |     kernel_size: 3
2064 |     stride: 1
2065 |     pad: 1
2066 |   }
2067 | }
2068 | layer {
2069 |   name: "inception_5b/pool_proj"
2070 |   type: "Convolution"
2071 |   bottom: "inception_5b/pool"
2072 |   top: "inception_5b/pool_proj"
2073 |   param {
2074 |     lr_mult: 1
2075 |     decay_mult: 1
2076 |   }
2077 |   param {
2078 |     lr_mult: 2
2079 |     decay_mult: 0
2080 |   }
2081 |   convolution_param {
2082 |     num_output: 128
2083 |     kernel_size: 1
2084 |     weight_filler {
2085 |       type: "xavier"
2086 |       std: 0.1
2087 |     }
2088 |     bias_filler {
2089 |       type: "constant"
2090 |       value: 0.2
2091 |     }
2092 |   }
2093 | }
2094 | layer {
2095 |   name: "inception_5b/relu_pool_proj"
2096 |   type: "ReLU"
2097 |   bottom: "inception_5b/pool_proj"
2098 |   top: "inception_5b/pool_proj"
2099 | }
2100 | layer {
2101 |   name: "inception_5b/output"
2102 |   type: "Concat"
2103 |   bottom: "inception_5b/1x1"
2104 |   bottom: "inception_5b/3x3"
2105 |   bottom: "inception_5b/5x5"
2106 |   bottom: "inception_5b/pool_proj"
2107 |   top: "inception_5b/output"
2108 | }
2109 | layer {
2110 |   name: "pool5/7x7_s1"
2111 |   type: "Pooling"
2112 |   bottom: "inception_5b/output"
2113 |   top: "pool5/7x7_s1"
2114 |   pooling_param {
2115 |     pool: AVE
2116 |     kernel_size: 7
2117 |     stride: 1
2118 |   }
2119 | }
2120 | layer {
2121 |   name: "pool5/drop_7x7_s1"
2122 |   type: "Dropout"
2123 |   bottom: "pool5/7x7_s1"
2124 |   top: "pool5/7x7_s1"
2125 |   dropout_param {
2126 |     dropout_ratio: 0.4
2127 |   }
2128 | }
2129 | layer {
2130 |   name: "loss3/classifier"
2131 |   type: "InnerProduct"
2132 |   bottom: "pool5/7x7_s1"
2133 |   top: "loss3/classifier"
2134 |   param {
2135 |     lr_mult: 1
2136 |     decay_mult: 1
2137 |   }
2138 |   param {
2139 |     lr_mult: 2
2140 |     decay_mult: 0
2141 |   }
2142 |   inner_product_param {
2143 |     num_output: 1000
2144 |     weight_filler {
2145 |       type: "xavier"
2146 |     }
2147 |     bias_filler {
2148 |       type: "constant"
2149 |       value: 0
2150 |     }
2151 |   }
2152 | }
2153 | #layer {
2154 | #  name: "prob"
2155 | #  type: "Softmax"
2156 | #  bottom: "loss3/classifier"
2157 | #  top: "prob"
2158 | #}
2159 | 


--------------------------------------------------------------------------------
/matlab/data/deploy_vgg_16.prototxt:
--------------------------------------------------------------------------------
  1 | force_backward: true
  2 | name: "VGG_ILSVRC_16_layers"
  3 | input: "data"
  4 | input_dim: 1
  5 | input_dim: 3
  6 | input_dim: 224
  7 | input_dim: 224
  8 | layers {
  9 |   bottom: "data"
 10 |   top: "conv1_1"
 11 |   name: "conv1_1"
 12 |   type: CONVOLUTION
 13 |   convolution_param {
 14 |     num_output: 64
 15 |     pad: 1
 16 |     kernel_size: 3
 17 |   }
 18 | }
 19 | layers {
 20 |   bottom: "conv1_1"
 21 |   top: "conv1_1"
 22 |   name: "relu1_1"
 23 |   type: RELU
 24 | }
 25 | layers {
 26 |   bottom: "conv1_1"
 27 |   top: "conv1_2"
 28 |   name: "conv1_2"
 29 |   type: CONVOLUTION
 30 |   convolution_param {
 31 |     num_output: 64
 32 |     pad: 1
 33 |     kernel_size: 3
 34 |   }
 35 | }
 36 | layers {
 37 |   bottom: "conv1_2"
 38 |   top: "conv1_2"
 39 |   name: "relu1_2"
 40 |   type: RELU
 41 | }
 42 | layers {
 43 |   bottom: "conv1_2"
 44 |   top: "pool1"
 45 |   name: "pool1"
 46 |   type: POOLING
 47 |   pooling_param {
 48 |     pool: MAX
 49 |     kernel_size: 2
 50 |     stride: 2
 51 |   }
 52 | }
 53 | layers {
 54 |   bottom: "pool1"
 55 |   top: "conv2_1"
 56 |   name: "conv2_1"
 57 |   type: CONVOLUTION
 58 |   convolution_param {
 59 |     num_output: 128
 60 |     pad: 1
 61 |     kernel_size: 3
 62 |   }
 63 | }
 64 | layers {
 65 |   bottom: "conv2_1"
 66 |   top: "conv2_1"
 67 |   name: "relu2_1"
 68 |   type: RELU
 69 | }
 70 | layers {
 71 |   bottom: "conv2_1"
 72 |   top: "conv2_2"
 73 |   name: "conv2_2"
 74 |   type: CONVOLUTION
 75 |   convolution_param {
 76 |     num_output: 128
 77 |     pad: 1
 78 |     kernel_size: 3
 79 |   }
 80 | }
 81 | layers {
 82 |   bottom: "conv2_2"
 83 |   top: "conv2_2"
 84 |   name: "relu2_2"
 85 |   type: RELU
 86 | }
 87 | layers {
 88 |   bottom: "conv2_2"
 89 |   top: "pool2"
 90 |   name: "pool2"
 91 |   type: POOLING
 92 |   pooling_param {
 93 |     pool: MAX
 94 |     kernel_size: 2
 95 |     stride: 2
 96 |   }
 97 | }
 98 | layers {
 99 |   bottom: "pool2"
100 |   top: "conv3_1"
101 |   name: "conv3_1"
102 |   type: CONVOLUTION
103 |   convolution_param {
104 |     num_output: 256
105 |     pad: 1
106 |     kernel_size: 3
107 |   }
108 | }
109 | layers {
110 |   bottom: "conv3_1"
111 |   top: "conv3_1"
112 |   name: "relu3_1"
113 |   type: RELU
114 | }
115 | layers {
116 |   bottom: "conv3_1"
117 |   top: "conv3_2"
118 |   name: "conv3_2"
119 |   type: CONVOLUTION
120 |   convolution_param {
121 |     num_output: 256
122 |     pad: 1
123 |     kernel_size: 3
124 |   }
125 | }
126 | layers {
127 |   bottom: "conv3_2"
128 |   top: "conv3_2"
129 |   name: "relu3_2"
130 |   type: RELU
131 | }
132 | layers {
133 |   bottom: "conv3_2"
134 |   top: "conv3_3"
135 |   name: "conv3_3"
136 |   type: CONVOLUTION
137 |   convolution_param {
138 |     num_output: 256
139 |     pad: 1
140 |     kernel_size: 3
141 |   }
142 | }
143 | layers {
144 |   bottom: "conv3_3"
145 |   top: "conv3_3"
146 |   name: "relu3_3"
147 |   type: RELU
148 | }
149 | layers {
150 |   bottom: "conv3_3"
151 |   top: "pool3"
152 |   name: "pool3"
153 |   type: POOLING
154 |   pooling_param {
155 |     pool: MAX
156 |     kernel_size: 2
157 |     stride: 2
158 |   }
159 | }
160 | layers {
161 |   bottom: "pool3"
162 |   top: "conv4_1"
163 |   name: "conv4_1"
164 |   type: CONVOLUTION
165 |   convolution_param {
166 |     num_output: 512
167 |     pad: 1
168 |     kernel_size: 3
169 |   }
170 | }
171 | layers {
172 |   bottom: "conv4_1"
173 |   top: "conv4_1"
174 |   name: "relu4_1"
175 |   type: RELU
176 | }
177 | layers {
178 |   bottom: "conv4_1"
179 |   top: "conv4_2"
180 |   name: "conv4_2"
181 |   type: CONVOLUTION
182 |   convolution_param {
183 |     num_output: 512
184 |     pad: 1
185 |     kernel_size: 3
186 |   }
187 | }
188 | layers {
189 |   bottom: "conv4_2"
190 |   top: "conv4_2"
191 |   name: "relu4_2"
192 |   type: RELU
193 | }
194 | layers {
195 |   bottom: "conv4_2"
196 |   top: "conv4_3"
197 |   name: "conv4_3"
198 |   type: CONVOLUTION
199 |   convolution_param {
200 |     num_output: 512
201 |     pad: 1
202 |     kernel_size: 3
203 |   }
204 | }
205 | layers {
206 |   bottom: "conv4_3"
207 |   top: "conv4_3"
208 |   name: "relu4_3"
209 |   type: RELU
210 | }
211 | layers {
212 |   bottom: "conv4_3"
213 |   top: "pool4"
214 |   name: "pool4"
215 |   type: POOLING
216 |   pooling_param {
217 |     pool: MAX
218 |     kernel_size: 2
219 |     stride: 2
220 |   }
221 | }
222 | layers {
223 |   bottom: "pool4"
224 |   top: "conv5_1"
225 |   name: "conv5_1"
226 |   type: CONVOLUTION
227 |   convolution_param {
228 |     num_output: 512
229 |     pad: 1
230 |     kernel_size: 3
231 |   }
232 | }
233 | layers {
234 |   bottom: "conv5_1"
235 |   top: "conv5_1"
236 |   name: "relu5_1"
237 |   type: RELU
238 | }
239 | layers {
240 |   bottom: "conv5_1"
241 |   top: "conv5_2"
242 |   name: "conv5_2"
243 |   type: CONVOLUTION
244 |   convolution_param {
245 |     num_output: 512
246 |     pad: 1
247 |     kernel_size: 3
248 |   }
249 | }
250 | layers {
251 |   bottom: "conv5_2"
252 |   top: "conv5_2"
253 |   name: "relu5_2"
254 |   type: RELU
255 | }
256 | layers {
257 |   bottom: "conv5_2"
258 |   top: "conv5_3"
259 |   name: "conv5_3"
260 |   type: CONVOLUTION
261 |   convolution_param {
262 |     num_output: 512
263 |     pad: 1
264 |     kernel_size: 3
265 |   }
266 | }
267 | layers {
268 |   bottom: "conv5_3"
269 |   top: "conv5_3"
270 |   name: "relu5_3"
271 |   type: RELU
272 | }
273 | layers {
274 |   bottom: "conv5_3"
275 |   top: "pool5"
276 |   name: "pool5"
277 |   type: POOLING
278 |   pooling_param {
279 |     pool: MAX
280 |     kernel_size: 2
281 |     stride: 2
282 |   }
283 | }
284 | layers {
285 |   bottom: "pool5"
286 |   top: "fc6"
287 |   name: "fc6"
288 |   type: INNER_PRODUCT
289 |   inner_product_param {
290 |     num_output: 4096
291 |   }
292 | }
293 | layers {
294 |   bottom: "fc6"
295 |   top: "fc6"
296 |   name: "relu6"
297 |   type: RELU
298 | }
299 | layers {
300 |   bottom: "fc6"
301 |   top: "fc6"
302 |   name: "drop6"
303 |   type: DROPOUT
304 |   dropout_param {
305 |     dropout_ratio: 0.5
306 |   }
307 | }
308 | layers {
309 |   bottom: "fc6"
310 |   top: "fc7"
311 |   name: "fc7"
312 |   type: INNER_PRODUCT
313 |   inner_product_param {
314 |     num_output: 4096
315 |   }
316 | }
317 | layers {
318 |   bottom: "fc7"
319 |   top: "fc7"
320 |   name: "relu7"
321 |   type: RELU
322 | }
323 | layers {
324 |   bottom: "fc7"
325 |   top: "fc7"
326 |   name: "drop7"
327 |   type: DROPOUT
328 |   dropout_param {
329 |     dropout_ratio: 0.5
330 |   }
331 | }
332 | layers {
333 |   bottom: "fc7"
334 |   top: "fc8"
335 |   name: "fc8"
336 |   type: INNER_PRODUCT
337 |   inner_product_param {
338 |     num_output: 1000
339 |   }
340 | }
341 | #layers {
342 | #  bottom: "fc8"
343 | #  top: "prob"
344 | #  name: "prob"
345 | #  type: SOFTMAX
346 | #}
347 | 


--------------------------------------------------------------------------------
/matlab/data/deploy_vgg_19.prototxt:
--------------------------------------------------------------------------------
  1 | force_backward: true
  2 | name: "VGG_ILSVRC_19_layers"
  3 | input: "data"
  4 | input_dim: 1
  5 | input_dim: 3
  6 | input_dim: 224
  7 | input_dim: 224
  8 | layers {
  9 |   bottom: "data"
 10 |   top: "conv1_1"
 11 |   name: "conv1_1"
 12 |   type: CONVOLUTION
 13 |   convolution_param {
 14 |     num_output: 64
 15 |     pad: 1
 16 |     kernel_size: 3
 17 |   }
 18 | }
 19 | layers {
 20 |   bottom: "conv1_1"
 21 |   top: "conv1_1"
 22 |   name: "relu1_1"
 23 |   type: RELU
 24 | }
 25 | layers {
 26 |   bottom: "conv1_1"
 27 |   top: "conv1_2"
 28 |   name: "conv1_2"
 29 |   type: CONVOLUTION
 30 |   convolution_param {
 31 |     num_output: 64
 32 |     pad: 1
 33 |     kernel_size: 3
 34 |   }
 35 | }
 36 | layers {
 37 |   bottom: "conv1_2"
 38 |   top: "conv1_2"
 39 |   name: "relu1_2"
 40 |   type: RELU
 41 | }
 42 | layers {
 43 |   bottom: "conv1_2"
 44 |   top: "pool1"
 45 |   name: "pool1"
 46 |   type: POOLING
 47 |   pooling_param {
 48 |     pool: MAX
 49 |     kernel_size: 2
 50 |     stride: 2
 51 |   }
 52 | }
 53 | layers {
 54 |   bottom: "pool1"
 55 |   top: "conv2_1"
 56 |   name: "conv2_1"
 57 |   type: CONVOLUTION
 58 |   convolution_param {
 59 |     num_output: 128
 60 |     pad: 1
 61 |     kernel_size: 3
 62 |   }
 63 | }
 64 | layers {
 65 |   bottom: "conv2_1"
 66 |   top: "conv2_1"
 67 |   name: "relu2_1"
 68 |   type: RELU
 69 | }
 70 | layers {
 71 |   bottom: "conv2_1"
 72 |   top: "conv2_2"
 73 |   name: "conv2_2"
 74 |   type: CONVOLUTION
 75 |   convolution_param {
 76 |     num_output: 128
 77 |     pad: 1
 78 |     kernel_size: 3
 79 |   }
 80 | }
 81 | layers {
 82 |   bottom: "conv2_2"
 83 |   top: "conv2_2"
 84 |   name: "relu2_2"
 85 |   type: RELU
 86 | }
 87 | layers {
 88 |   bottom: "conv2_2"
 89 |   top: "pool2"
 90 |   name: "pool2"
 91 |   type: POOLING
 92 |   pooling_param {
 93 |     pool: MAX
 94 |     kernel_size: 2
 95 |     stride: 2
 96 |   }
 97 | }
 98 | layers {
 99 |   bottom: "pool2"
100 |   top: "conv3_1"
101 |   name: "conv3_1"
102 |   type: CONVOLUTION
103 |   convolution_param {
104 |     num_output: 256
105 |     pad: 1
106 |     kernel_size: 3
107 |   }
108 | }
109 | layers {
110 |   bottom: "conv3_1"
111 |   top: "conv3_1"
112 |   name: "relu3_1"
113 |   type: RELU
114 | }
115 | layers {
116 |   bottom: "conv3_1"
117 |   top: "conv3_2"
118 |   name: "conv3_2"
119 |   type: CONVOLUTION
120 |   convolution_param {
121 |     num_output: 256
122 |     pad: 1
123 |     kernel_size: 3
124 |   }
125 | }
126 | layers {
127 |   bottom: "conv3_2"
128 |   top: "conv3_2"
129 |   name: "relu3_2"
130 |   type: RELU
131 | }
132 | layers {
133 |   bottom: "conv3_2"
134 |   top: "conv3_3"
135 |   name: "conv3_3"
136 |   type: CONVOLUTION
137 |   convolution_param {
138 |     num_output: 256
139 |     pad: 1
140 |     kernel_size: 3
141 |   }
142 | }
143 | layers {
144 |   bottom: "conv3_3"
145 |   top: "conv3_3"
146 |   name: "relu3_3"
147 |   type: RELU
148 | }
149 | layers {
150 |   bottom: "conv3_3"
151 |   top: "conv3_4"
152 |   name: "conv3_4"
153 |   type: CONVOLUTION
154 |   convolution_param {
155 |     num_output: 256
156 |     pad: 1
157 |     kernel_size: 3
158 |   }
159 | }
160 | layers {
161 |   bottom: "conv3_4"
162 |   top: "conv3_4"
163 |   name: "relu3_4"
164 |   type: RELU
165 | }
166 | layers {
167 |   bottom: "conv3_4"
168 |   top: "pool3"
169 |   name: "pool3"
170 |   type: POOLING
171 |   pooling_param {
172 |     pool: MAX
173 |     kernel_size: 2
174 |     stride: 2
175 |   }
176 | }
177 | layers {
178 |   bottom: "pool3"
179 |   top: "conv4_1"
180 |   name: "conv4_1"
181 |   type: CONVOLUTION
182 |   convolution_param {
183 |     num_output: 512
184 |     pad: 1
185 |     kernel_size: 3
186 |   }
187 | }
188 | layers {
189 |   bottom: "conv4_1"
190 |   top: "conv4_1"
191 |   name: "relu4_1"
192 |   type: RELU
193 | }
194 | layers {
195 |   bottom: "conv4_1"
196 |   top: "conv4_2"
197 |   name: "conv4_2"
198 |   type: CONVOLUTION
199 |   convolution_param {
200 |     num_output: 512
201 |     pad: 1
202 |     kernel_size: 3
203 |   }
204 | }
205 | layers {
206 |   bottom: "conv4_2"
207 |   top: "conv4_2"
208 |   name: "relu4_2"
209 |   type: RELU
210 | }
211 | layers {
212 |   bottom: "conv4_2"
213 |   top: "conv4_3"
214 |   name: "conv4_3"
215 |   type: CONVOLUTION
216 |   convolution_param {
217 |     num_output: 512
218 |     pad: 1
219 |     kernel_size: 3
220 |   }
221 | }
222 | layers {
223 |   bottom: "conv4_3"
224 |   top: "conv4_3"
225 |   name: "relu4_3"
226 |   type: RELU
227 | }
228 | layers {
229 |   bottom: "conv4_3"
230 |   top: "conv4_4"
231 |   name: "conv4_4"
232 |   type: CONVOLUTION
233 |   convolution_param {
234 |     num_output: 512
235 |     pad: 1
236 |     kernel_size: 3
237 |   }
238 | }
239 | layers {
240 |   bottom: "conv4_4"
241 |   top: "conv4_4"
242 |   name: "relu4_4"
243 |   type: RELU
244 | }
245 | layers {
246 |   bottom: "conv4_4"
247 |   top: "pool4"
248 |   name: "pool4"
249 |   type: POOLING
250 |   pooling_param {
251 |     pool: MAX
252 |     kernel_size: 2
253 |     stride: 2
254 |   }
255 | }
256 | layers {
257 |   bottom: "pool4"
258 |   top: "conv5_1"
259 |   name: "conv5_1"
260 |   type: CONVOLUTION
261 |   convolution_param {
262 |     num_output: 512
263 |     pad: 1
264 |     kernel_size: 3
265 |   }
266 | }
267 | layers {
268 |   bottom: "conv5_1"
269 |   top: "conv5_1"
270 |   name: "relu5_1"
271 |   type: RELU
272 | }
273 | layers {
274 |   bottom: "conv5_1"
275 |   top: "conv5_2"
276 |   name: "conv5_2"
277 |   type: CONVOLUTION
278 |   convolution_param {
279 |     num_output: 512
280 |     pad: 1
281 |     kernel_size: 3
282 |   }
283 | }
284 | layers {
285 |   bottom: "conv5_2"
286 |   top: "conv5_2"
287 |   name: "relu5_2"
288 |   type: RELU
289 | }
290 | layers {
291 |   bottom: "conv5_2"
292 |   top: "conv5_3"
293 |   name: "conv5_3"
294 |   type: CONVOLUTION
295 |   convolution_param {
296 |     num_output: 512
297 |     pad: 1
298 |     kernel_size: 3
299 |   }
300 | }
301 | layers {
302 |   bottom: "conv5_3"
303 |   top: "conv5_3"
304 |   name: "relu5_3"
305 |   type: RELU
306 | }
307 | layers {
308 |   bottom: "conv5_3"
309 |   top: "conv5_4"
310 |   name: "conv5_4"
311 |   type: CONVOLUTION
312 |   convolution_param {
313 |     num_output: 512
314 |     pad: 1
315 |     kernel_size: 3
316 |   }
317 | }
318 | layers {
319 |   bottom: "conv5_4"
320 |   top: "conv5_4"
321 |   name: "relu5_4"
322 |   type: RELU
323 | }
324 | layers {
325 |   bottom: "conv5_4"
326 |   top: "pool5"
327 |   name: "pool5"
328 |   type: POOLING
329 |   pooling_param {
330 |     pool: MAX
331 |     kernel_size: 2
332 |     stride: 2
333 |   }
334 | }
335 | layers {
336 |   bottom: "pool5"
337 |   top: "fc6"
338 |   name: "fc6"
339 |   type: INNER_PRODUCT
340 |   inner_product_param {
341 |     num_output: 4096
342 |   }
343 | }
344 | layers {
345 |   bottom: "fc6"
346 |   top: "fc6"
347 |   name: "relu6"
348 |   type: RELU
349 | }
350 | layers {
351 |   bottom: "fc6"
352 |   top: "fc6"
353 |   name: "drop6"
354 |   type: DROPOUT
355 |   dropout_param {
356 |     dropout_ratio: 0.5
357 |   }
358 | }
359 | layers {
360 |   bottom: "fc6"
361 |   top: "fc7"
362 |   name: "fc7"
363 |   type: INNER_PRODUCT
364 |   inner_product_param {
365 |     num_output: 4096
366 |   }
367 | }
368 | layers {
369 |   bottom: "fc7"
370 |   top: "fc7"
371 |   name: "relu7"
372 |   type: RELU
373 | }
374 | layers {
375 |   bottom: "fc7"
376 |   top: "fc7"
377 |   name: "drop7"
378 |   type: DROPOUT
379 |   dropout_param {
380 |     dropout_ratio: 0.5
381 |   }
382 | }
383 | layers {
384 |   bottom: "fc7"
385 |   top: "fc8"
386 |   name: "fc8"
387 |   type: INNER_PRODUCT
388 |   inner_product_param {
389 |     num_output: 1000
390 |   }
391 | }
392 | #layers {
393 | #  bottom: "fc8"
394 | #  top: "prob"
395 | #  name: "prob"
396 | #  type: SOFTMAX
397 | #}
398 | 


--------------------------------------------------------------------------------
/matlab/data/deploy_vgg_f.prototxt:
--------------------------------------------------------------------------------
  1 | force_backward: true
  2 | name: "VGG_CNN_F"
  3 | input: "data"
  4 | input_dim: 1
  5 | input_dim: 3
  6 | input_dim: 224
  7 | input_dim: 224
  8 | layers {
  9 |   bottom: "data"
 10 |   top: "conv1"
 11 |   name: "conv1"
 12 |   type: CONVOLUTION
 13 |   convolution_param {
 14 |     num_output: 64
 15 |     kernel_size: 11
 16 |     stride: 4
 17 |   }
 18 | }
 19 | layers {
 20 |   bottom: "conv1"
 21 |   top: "conv1"
 22 |   name: "relu1"
 23 |   type: RELU
 24 | }
 25 | layers {
 26 |   bottom: "conv1"
 27 |   top: "norm1"
 28 |   name: "norm1"
 29 |   type: LRN
 30 |   lrn_param {
 31 |     local_size: 5
 32 |     alpha: 0.0005
 33 |     beta: 0.75
 34 |     k: 2
 35 |   }
 36 | }
 37 | layers {
 38 |   bottom: "norm1"
 39 |   top: "pool1"
 40 |   name: "pool1"
 41 |   type: POOLING
 42 |   pooling_param {
 43 |     pool: MAX
 44 |     kernel_size: 3
 45 |     stride: 2
 46 |   }
 47 | }
 48 | layers {
 49 |   bottom: "pool1"
 50 |   top: "conv2"
 51 |   name: "conv2"
 52 |   type: CONVOLUTION
 53 |   convolution_param {
 54 |     num_output: 256
 55 |     pad: 2
 56 |     kernel_size: 5
 57 |   }
 58 | }
 59 | layers {
 60 |   bottom: "conv2"
 61 |   top: "conv2"
 62 |   name: "relu2"
 63 |   type: RELU
 64 | }
 65 | layers {
 66 |   bottom: "conv2"
 67 |   top: "norm2"
 68 |   name: "norm2"
 69 |   type: LRN
 70 |   lrn_param {
 71 |     local_size: 5
 72 |     alpha: 0.0005
 73 |     beta: 0.75
 74 |     k: 2
 75 |   }
 76 | }
 77 | layers {
 78 |   bottom: "norm2"
 79 |   top: "pool2"
 80 |   name: "pool2"
 81 |   type: POOLING
 82 |   pooling_param {
 83 |     pool: MAX
 84 |     kernel_size: 3
 85 |     stride: 2
 86 |   }
 87 | }
 88 | layers {
 89 |   bottom: "pool2"
 90 |   top: "conv3"
 91 |   name: "conv3"
 92 |   type: CONVOLUTION
 93 |   convolution_param {
 94 |     num_output: 256
 95 |     pad: 1
 96 |     kernel_size: 3
 97 |   }
 98 | }
 99 | layers {
100 |   bottom: "conv3"
101 |   top: "conv3"
102 |   name: "relu3"
103 |   type: RELU
104 | }
105 | layers {
106 |   bottom: "conv3"
107 |   top: "conv4"
108 |   name: "conv4"
109 |   type: CONVOLUTION
110 |   convolution_param {
111 |     num_output: 256
112 |     pad: 1
113 |     kernel_size: 3
114 |   }
115 | }
116 | layers {
117 |   bottom: "conv4"
118 |   top: "conv4"
119 |   name: "relu4"
120 |   type: RELU
121 | }
122 | layers {
123 |   bottom: "conv4"
124 |   top: "conv5"
125 |   name: "conv5"
126 |   type: CONVOLUTION
127 |   convolution_param {
128 |     num_output: 256
129 |     pad: 1
130 |     kernel_size: 3
131 |   }
132 | }
133 | layers {
134 |   bottom: "conv5"
135 |   top: "conv5"
136 |   name: "relu5"
137 |   type: RELU
138 | }
139 | layers {
140 |   bottom: "conv5"
141 |   top: "pool5"
142 |   name: "pool5"
143 |   type: POOLING
144 |   pooling_param {
145 |     pool: MAX
146 |     kernel_size: 3
147 |     stride: 2
148 |   }
149 | }
150 | layers {
151 |   bottom: "pool5"
152 |   top: "fc6"
153 |   name: "fc6"
154 |   type: INNER_PRODUCT
155 |   inner_product_param {
156 |     num_output: 4096
157 |   }
158 | }
159 | layers {
160 |   bottom: "fc6"
161 |   top: "fc6"
162 |   name: "relu6"
163 |   type: RELU
164 | }
165 | layers {
166 |   bottom: "fc6"
167 |   top: "fc6"
168 |   name: "drop6"
169 |   type: DROPOUT
170 |   dropout_param {
171 |     dropout_ratio: 0.5
172 |   }
173 | }
174 | layers {
175 |   bottom: "fc6"
176 |   top: "fc7"
177 |   name: "fc7"
178 |   type: INNER_PRODUCT
179 |   inner_product_param {
180 |     num_output: 4096
181 |   }
182 | }
183 | layers {
184 |   bottom: "fc7"
185 |   top: "fc7"
186 |   name: "relu7"
187 |   type: RELU
188 | }
189 | layers {
190 |   bottom: "fc7"
191 |   top: "fc7"
192 |   name: "drop7"
193 |   type: DROPOUT
194 |   dropout_param {
195 |     dropout_ratio: 0.5
196 |   }
197 | }
198 | layers {
199 |   bottom: "fc7"
200 |   top: "fc8"
201 |   name: "fc8"
202 |   type: INNER_PRODUCT
203 |   inner_product_param {
204 |     num_output: 1000
205 |   }
206 | }
207 | #layers {
208 | #  bottom: "fc8"
209 | #  top: "prob"
210 | #  name: "prob"
211 | #  type: SOFTMAX
212 | #}
213 | 


--------------------------------------------------------------------------------
/matlab/data/ilsvrc_2012_mean.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/matlab/data/ilsvrc_2012_mean.mat


--------------------------------------------------------------------------------
/matlab/data/synset_words.txt:
--------------------------------------------------------------------------------
   1 | n01440764 tench, Tinca tinca
   2 | n01443537 goldfish, Carassius auratus
   3 | n01484850 great white shark, white shark, man-eater, man-eating shark, Carcharodon carcharias
   4 | n01491361 tiger shark, Galeocerdo cuvieri
   5 | n01494475 hammerhead, hammerhead shark
   6 | n01496331 electric ray, crampfish, numbfish, torpedo
   7 | n01498041 stingray
   8 | n01514668 cock
   9 | n01514859 hen
  10 | n01518878 ostrich, Struthio camelus
  11 | n01530575 brambling, Fringilla montifringilla
  12 | n01531178 goldfinch, Carduelis carduelis
  13 | n01532829 house finch, linnet, Carpodacus mexicanus
  14 | n01534433 junco, snowbird
  15 | n01537544 indigo bunting, indigo finch, indigo bird, Passerina cyanea
  16 | n01558993 robin, American robin, Turdus migratorius
  17 | n01560419 bulbul
  18 | n01580077 jay
  19 | n01582220 magpie
  20 | n01592084 chickadee
  21 | n01601694 water ouzel, dipper
  22 | n01608432 kite
  23 | n01614925 bald eagle, American eagle, Haliaeetus leucocephalus
  24 | n01616318 vulture
  25 | n01622779 great grey owl, great gray owl, Strix nebulosa
  26 | n01629819 European fire salamander, Salamandra salamandra
  27 | n01630670 common newt, Triturus vulgaris
  28 | n01631663 eft
  29 | n01632458 spotted salamander, Ambystoma maculatum
  30 | n01632777 axolotl, mud puppy, Ambystoma mexicanum
  31 | n01641577 bullfrog, Rana catesbeiana
  32 | n01644373 tree frog, tree-frog
  33 | n01644900 tailed frog, bell toad, ribbed toad, tailed toad, Ascaphus trui
  34 | n01664065 loggerhead, loggerhead turtle, Caretta caretta
  35 | n01665541 leatherback turtle, leatherback, leathery turtle, Dermochelys coriacea
  36 | n01667114 mud turtle
  37 | n01667778 terrapin
  38 | n01669191 box turtle, box tortoise
  39 | n01675722 banded gecko
  40 | n01677366 common iguana, iguana, Iguana iguana
  41 | n01682714 American chameleon, anole, Anolis carolinensis
  42 | n01685808 whiptail, whiptail lizard
  43 | n01687978 agama
  44 | n01688243 frilled lizard, Chlamydosaurus kingi
  45 | n01689811 alligator lizard
  46 | n01692333 Gila monster, Heloderma suspectum
  47 | n01693334 green lizard, Lacerta viridis
  48 | n01694178 African chameleon, Chamaeleo chamaeleon
  49 | n01695060 Komodo dragon, Komodo lizard, dragon lizard, giant lizard, Varanus komodoensis
  50 | n01697457 African crocodile, Nile crocodile, Crocodylus niloticus
  51 | n01698640 American alligator, Alligator mississipiensis
  52 | n01704323 triceratops
  53 | n01728572 thunder snake, worm snake, Carphophis amoenus
  54 | n01728920 ringneck snake, ring-necked snake, ring snake
  55 | n01729322 hognose snake, puff adder, sand viper
  56 | n01729977 green snake, grass snake
  57 | n01734418 king snake, kingsnake
  58 | n01735189 garter snake, grass snake
  59 | n01737021 water snake
  60 | n01739381 vine snake
  61 | n01740131 night snake, Hypsiglena torquata
  62 | n01742172 boa constrictor, Constrictor constrictor
  63 | n01744401 rock python, rock snake, Python sebae
  64 | n01748264 Indian cobra, Naja naja
  65 | n01749939 green mamba
  66 | n01751748 sea snake
  67 | n01753488 horned viper, cerastes, sand viper, horned asp, Cerastes cornutus
  68 | n01755581 diamondback, diamondback rattlesnake, Crotalus adamanteus
  69 | n01756291 sidewinder, horned rattlesnake, Crotalus cerastes
  70 | n01768244 trilobite
  71 | n01770081 harvestman, daddy longlegs, Phalangium opilio
  72 | n01770393 scorpion
  73 | n01773157 black and gold garden spider, Argiope aurantia
  74 | n01773549 barn spider, Araneus cavaticus
  75 | n01773797 garden spider, Aranea diademata
  76 | n01774384 black widow, Latrodectus mactans
  77 | n01774750 tarantula
  78 | n01775062 wolf spider, hunting spider
  79 | n01776313 tick
  80 | n01784675 centipede
  81 | n01795545 black grouse
  82 | n01796340 ptarmigan
  83 | n01797886 ruffed grouse, partridge, Bonasa umbellus
  84 | n01798484 prairie chicken, prairie grouse, prairie fowl
  85 | n01806143 peacock
  86 | n01806567 quail
  87 | n01807496 partridge
  88 | n01817953 African grey, African gray, Psittacus erithacus
  89 | n01818515 macaw
  90 | n01819313 sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita
  91 | n01820546 lorikeet
  92 | n01824575 coucal
  93 | n01828970 bee eater
  94 | n01829413 hornbill
  95 | n01833805 hummingbird
  96 | n01843065 jacamar
  97 | n01843383 toucan
  98 | n01847000 drake
  99 | n01855032 red-breasted merganser, Mergus serrator
 100 | n01855672 goose
 101 | n01860187 black swan, Cygnus atratus
 102 | n01871265 tusker
 103 | n01872401 echidna, spiny anteater, anteater
 104 | n01873310 platypus, duckbill, duckbilled platypus, duck-billed platypus, Ornithorhynchus anatinus
 105 | n01877812 wallaby, brush kangaroo
 106 | n01882714 koala, koala bear, kangaroo bear, native bear, Phascolarctos cinereus
 107 | n01883070 wombat
 108 | n01910747 jellyfish
 109 | n01914609 sea anemone, anemone
 110 | n01917289 brain coral
 111 | n01924916 flatworm, platyhelminth
 112 | n01930112 nematode, nematode worm, roundworm
 113 | n01943899 conch
 114 | n01944390 snail
 115 | n01945685 slug
 116 | n01950731 sea slug, nudibranch
 117 | n01955084 chiton, coat-of-mail shell, sea cradle, polyplacophore
 118 | n01968897 chambered nautilus, pearly nautilus, nautilus
 119 | n01978287 Dungeness crab, Cancer magister
 120 | n01978455 rock crab, Cancer irroratus
 121 | n01980166 fiddler crab
 122 | n01981276 king crab, Alaska crab, Alaskan king crab, Alaska king crab, Paralithodes camtschatica
 123 | n01983481 American lobster, Northern lobster, Maine lobster, Homarus americanus
 124 | n01984695 spiny lobster, langouste, rock lobster, crawfish, crayfish, sea crawfish
 125 | n01985128 crayfish, crawfish, crawdad, crawdaddy
 126 | n01986214 hermit crab
 127 | n01990800 isopod
 128 | n02002556 white stork, Ciconia ciconia
 129 | n02002724 black stork, Ciconia nigra
 130 | n02006656 spoonbill
 131 | n02007558 flamingo
 132 | n02009229 little blue heron, Egretta caerulea
 133 | n02009912 American egret, great white heron, Egretta albus
 134 | n02011460 bittern
 135 | n02012849 crane
 136 | n02013706 limpkin, Aramus pictus
 137 | n02017213 European gallinule, Porphyrio porphyrio
 138 | n02018207 American coot, marsh hen, mud hen, water hen, Fulica americana
 139 | n02018795 bustard
 140 | n02025239 ruddy turnstone, Arenaria interpres
 141 | n02027492 red-backed sandpiper, dunlin, Erolia alpina
 142 | n02028035 redshank, Tringa totanus
 143 | n02033041 dowitcher
 144 | n02037110 oystercatcher, oyster catcher
 145 | n02051845 pelican
 146 | n02056570 king penguin, Aptenodytes patagonica
 147 | n02058221 albatross, mollymawk
 148 | n02066245 grey whale, gray whale, devilfish, Eschrichtius gibbosus, Eschrichtius robustus
 149 | n02071294 killer whale, killer, orca, grampus, sea wolf, Orcinus orca
 150 | n02074367 dugong, Dugong dugon
 151 | n02077923 sea lion
 152 | n02085620 Chihuahua
 153 | n02085782 Japanese spaniel
 154 | n02085936 Maltese dog, Maltese terrier, Maltese
 155 | n02086079 Pekinese, Pekingese, Peke
 156 | n02086240 Shih-Tzu
 157 | n02086646 Blenheim spaniel
 158 | n02086910 papillon
 159 | n02087046 toy terrier
 160 | n02087394 Rhodesian ridgeback
 161 | n02088094 Afghan hound, Afghan
 162 | n02088238 basset, basset hound
 163 | n02088364 beagle
 164 | n02088466 bloodhound, sleuthhound
 165 | n02088632 bluetick
 166 | n02089078 black-and-tan coonhound
 167 | n02089867 Walker hound, Walker foxhound
 168 | n02089973 English foxhound
 169 | n02090379 redbone
 170 | n02090622 borzoi, Russian wolfhound
 171 | n02090721 Irish wolfhound
 172 | n02091032 Italian greyhound
 173 | n02091134 whippet
 174 | n02091244 Ibizan hound, Ibizan Podenco
 175 | n02091467 Norwegian elkhound, elkhound
 176 | n02091635 otterhound, otter hound
 177 | n02091831 Saluki, gazelle hound
 178 | n02092002 Scottish deerhound, deerhound
 179 | n02092339 Weimaraner
 180 | n02093256 Staffordshire bullterrier, Staffordshire bull terrier
 181 | n02093428 American Staffordshire terrier, Staffordshire terrier, American pit bull terrier, pit bull terrier
 182 | n02093647 Bedlington terrier
 183 | n02093754 Border terrier
 184 | n02093859 Kerry blue terrier
 185 | n02093991 Irish terrier
 186 | n02094114 Norfolk terrier
 187 | n02094258 Norwich terrier
 188 | n02094433 Yorkshire terrier
 189 | n02095314 wire-haired fox terrier
 190 | n02095570 Lakeland terrier
 191 | n02095889 Sealyham terrier, Sealyham
 192 | n02096051 Airedale, Airedale terrier
 193 | n02096177 cairn, cairn terrier
 194 | n02096294 Australian terrier
 195 | n02096437 Dandie Dinmont, Dandie Dinmont terrier
 196 | n02096585 Boston bull, Boston terrier
 197 | n02097047 miniature schnauzer
 198 | n02097130 giant schnauzer
 199 | n02097209 standard schnauzer
 200 | n02097298 Scotch terrier, Scottish terrier, Scottie
 201 | n02097474 Tibetan terrier, chrysanthemum dog
 202 | n02097658 silky terrier, Sydney silky
 203 | n02098105 soft-coated wheaten terrier
 204 | n02098286 West Highland white terrier
 205 | n02098413 Lhasa, Lhasa apso
 206 | n02099267 flat-coated retriever
 207 | n02099429 curly-coated retriever
 208 | n02099601 golden retriever
 209 | n02099712 Labrador retriever
 210 | n02099849 Chesapeake Bay retriever
 211 | n02100236 German short-haired pointer
 212 | n02100583 vizsla, Hungarian pointer
 213 | n02100735 English setter
 214 | n02100877 Irish setter, red setter
 215 | n02101006 Gordon setter
 216 | n02101388 Brittany spaniel
 217 | n02101556 clumber, clumber spaniel
 218 | n02102040 English springer, English springer spaniel
 219 | n02102177 Welsh springer spaniel
 220 | n02102318 cocker spaniel, English cocker spaniel, cocker
 221 | n02102480 Sussex spaniel
 222 | n02102973 Irish water spaniel
 223 | n02104029 kuvasz
 224 | n02104365 schipperke
 225 | n02105056 groenendael
 226 | n02105162 malinois
 227 | n02105251 briard
 228 | n02105412 kelpie
 229 | n02105505 komondor
 230 | n02105641 Old English sheepdog, bobtail
 231 | n02105855 Shetland sheepdog, Shetland sheep dog, Shetland
 232 | n02106030 collie
 233 | n02106166 Border collie
 234 | n02106382 Bouvier des Flandres, Bouviers des Flandres
 235 | n02106550 Rottweiler
 236 | n02106662 German shepherd, German shepherd dog, German police dog, alsatian
 237 | n02107142 Doberman, Doberman pinscher
 238 | n02107312 miniature pinscher
 239 | n02107574 Greater Swiss Mountain dog
 240 | n02107683 Bernese mountain dog
 241 | n02107908 Appenzeller
 242 | n02108000 EntleBucher
 243 | n02108089 boxer
 244 | n02108422 bull mastiff
 245 | n02108551 Tibetan mastiff
 246 | n02108915 French bulldog
 247 | n02109047 Great Dane
 248 | n02109525 Saint Bernard, St Bernard
 249 | n02109961 Eskimo dog, husky
 250 | n02110063 malamute, malemute, Alaskan malamute
 251 | n02110185 Siberian husky
 252 | n02110341 dalmatian, coach dog, carriage dog
 253 | n02110627 affenpinscher, monkey pinscher, monkey dog
 254 | n02110806 basenji
 255 | n02110958 pug, pug-dog
 256 | n02111129 Leonberg
 257 | n02111277 Newfoundland, Newfoundland dog
 258 | n02111500 Great Pyrenees
 259 | n02111889 Samoyed, Samoyede
 260 | n02112018 Pomeranian
 261 | n02112137 chow, chow chow
 262 | n02112350 keeshond
 263 | n02112706 Brabancon griffon
 264 | n02113023 Pembroke, Pembroke Welsh corgi
 265 | n02113186 Cardigan, Cardigan Welsh corgi
 266 | n02113624 toy poodle
 267 | n02113712 miniature poodle
 268 | n02113799 standard poodle
 269 | n02113978 Mexican hairless
 270 | n02114367 timber wolf, grey wolf, gray wolf, Canis lupus
 271 | n02114548 white wolf, Arctic wolf, Canis lupus tundrarum
 272 | n02114712 red wolf, maned wolf, Canis rufus, Canis niger
 273 | n02114855 coyote, prairie wolf, brush wolf, Canis latrans
 274 | n02115641 dingo, warrigal, warragal, Canis dingo
 275 | n02115913 dhole, Cuon alpinus
 276 | n02116738 African hunting dog, hyena dog, Cape hunting dog, Lycaon pictus
 277 | n02117135 hyena, hyaena
 278 | n02119022 red fox, Vulpes vulpes
 279 | n02119789 kit fox, Vulpes macrotis
 280 | n02120079 Arctic fox, white fox, Alopex lagopus
 281 | n02120505 grey fox, gray fox, Urocyon cinereoargenteus
 282 | n02123045 tabby, tabby cat
 283 | n02123159 tiger cat
 284 | n02123394 Persian cat
 285 | n02123597 Siamese cat, Siamese
 286 | n02124075 Egyptian cat
 287 | n02125311 cougar, puma, catamount, mountain lion, painter, panther, Felis concolor
 288 | n02127052 lynx, catamount
 289 | n02128385 leopard, Panthera pardus
 290 | n02128757 snow leopard, ounce, Panthera uncia
 291 | n02128925 jaguar, panther, Panthera onca, Felis onca
 292 | n02129165 lion, king of beasts, Panthera leo
 293 | n02129604 tiger, Panthera tigris
 294 | n02130308 cheetah, chetah, Acinonyx jubatus
 295 | n02132136 brown bear, bruin, Ursus arctos
 296 | n02133161 American black bear, black bear, Ursus americanus, Euarctos americanus
 297 | n02134084 ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus
 298 | n02134418 sloth bear, Melursus ursinus, Ursus ursinus
 299 | n02137549 mongoose
 300 | n02138441 meerkat, mierkat
 301 | n02165105 tiger beetle
 302 | n02165456 ladybug, ladybeetle, lady beetle, ladybird, ladybird beetle
 303 | n02167151 ground beetle, carabid beetle
 304 | n02168699 long-horned beetle, longicorn, longicorn beetle
 305 | n02169497 leaf beetle, chrysomelid
 306 | n02172182 dung beetle
 307 | n02174001 rhinoceros beetle
 308 | n02177972 weevil
 309 | n02190166 fly
 310 | n02206856 bee
 311 | n02219486 ant, emmet, pismire
 312 | n02226429 grasshopper, hopper
 313 | n02229544 cricket
 314 | n02231487 walking stick, walkingstick, stick insect
 315 | n02233338 cockroach, roach
 316 | n02236044 mantis, mantid
 317 | n02256656 cicada, cicala
 318 | n02259212 leafhopper
 319 | n02264363 lacewing, lacewing fly
 320 | n02268443 dragonfly, darning needle, devil's darning needle, sewing needle, snake feeder, snake doctor, mosquito hawk, skeeter hawk
 321 | n02268853 damselfly
 322 | n02276258 admiral
 323 | n02277742 ringlet, ringlet butterfly
 324 | n02279972 monarch, monarch butterfly, milkweed butterfly, Danaus plexippus
 325 | n02280649 cabbage butterfly
 326 | n02281406 sulphur butterfly, sulfur butterfly
 327 | n02281787 lycaenid, lycaenid butterfly
 328 | n02317335 starfish, sea star
 329 | n02319095 sea urchin
 330 | n02321529 sea cucumber, holothurian
 331 | n02325366 wood rabbit, cottontail, cottontail rabbit
 332 | n02326432 hare
 333 | n02328150 Angora, Angora rabbit
 334 | n02342885 hamster
 335 | n02346627 porcupine, hedgehog
 336 | n02356798 fox squirrel, eastern fox squirrel, Sciurus niger
 337 | n02361337 marmot
 338 | n02363005 beaver
 339 | n02364673 guinea pig, Cavia cobaya
 340 | n02389026 sorrel
 341 | n02391049 zebra
 342 | n02395406 hog, pig, grunter, squealer, Sus scrofa
 343 | n02396427 wild boar, boar, Sus scrofa
 344 | n02397096 warthog
 345 | n02398521 hippopotamus, hippo, river horse, Hippopotamus amphibius
 346 | n02403003 ox
 347 | n02408429 water buffalo, water ox, Asiatic buffalo, Bubalus bubalis
 348 | n02410509 bison
 349 | n02412080 ram, tup
 350 | n02415577 bighorn, bighorn sheep, cimarron, Rocky Mountain bighorn, Rocky Mountain sheep, Ovis canadensis
 351 | n02417914 ibex, Capra ibex
 352 | n02422106 hartebeest
 353 | n02422699 impala, Aepyceros melampus
 354 | n02423022 gazelle
 355 | n02437312 Arabian camel, dromedary, Camelus dromedarius
 356 | n02437616 llama
 357 | n02441942 weasel
 358 | n02442845 mink
 359 | n02443114 polecat, fitch, foulmart, foumart, Mustela putorius
 360 | n02443484 black-footed ferret, ferret, Mustela nigripes
 361 | n02444819 otter
 362 | n02445715 skunk, polecat, wood pussy
 363 | n02447366 badger
 364 | n02454379 armadillo
 365 | n02457408 three-toed sloth, ai, Bradypus tridactylus
 366 | n02480495 orangutan, orang, orangutang, Pongo pygmaeus
 367 | n02480855 gorilla, Gorilla gorilla
 368 | n02481823 chimpanzee, chimp, Pan troglodytes
 369 | n02483362 gibbon, Hylobates lar
 370 | n02483708 siamang, Hylobates syndactylus, Symphalangus syndactylus
 371 | n02484975 guenon, guenon monkey
 372 | n02486261 patas, hussar monkey, Erythrocebus patas
 373 | n02486410 baboon
 374 | n02487347 macaque
 375 | n02488291 langur
 376 | n02488702 colobus, colobus monkey
 377 | n02489166 proboscis monkey, Nasalis larvatus
 378 | n02490219 marmoset
 379 | n02492035 capuchin, ringtail, Cebus capucinus
 380 | n02492660 howler monkey, howler
 381 | n02493509 titi, titi monkey
 382 | n02493793 spider monkey, Ateles geoffroyi
 383 | n02494079 squirrel monkey, Saimiri sciureus
 384 | n02497673 Madagascar cat, ring-tailed lemur, Lemur catta
 385 | n02500267 indri, indris, Indri indri, Indri brevicaudatus
 386 | n02504013 Indian elephant, Elephas maximus
 387 | n02504458 African elephant, Loxodonta africana
 388 | n02509815 lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens
 389 | n02510455 giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca
 390 | n02514041 barracouta, snoek
 391 | n02526121 eel
 392 | n02536864 coho, cohoe, coho salmon, blue jack, silver salmon, Oncorhynchus kisutch
 393 | n02606052 rock beauty, Holocanthus tricolor
 394 | n02607072 anemone fish
 395 | n02640242 sturgeon
 396 | n02641379 gar, garfish, garpike, billfish, Lepisosteus osseus
 397 | n02643566 lionfish
 398 | n02655020 puffer, pufferfish, blowfish, globefish
 399 | n02666196 abacus
 400 | n02667093 abaya
 401 | n02669723 academic gown, academic robe, judge's robe
 402 | n02672831 accordion, piano accordion, squeeze box
 403 | n02676566 acoustic guitar
 404 | n02687172 aircraft carrier, carrier, flattop, attack aircraft carrier
 405 | n02690373 airliner
 406 | n02692877 airship, dirigible
 407 | n02699494 altar
 408 | n02701002 ambulance
 409 | n02704792 amphibian, amphibious vehicle
 410 | n02708093 analog clock
 411 | n02727426 apiary, bee house
 412 | n02730930 apron
 413 | n02747177 ashcan, trash can, garbage can, wastebin, ash bin, ash-bin, ashbin, dustbin, trash barrel, trash bin
 414 | n02749479 assault rifle, assault gun
 415 | n02769748 backpack, back pack, knapsack, packsack, rucksack, haversack
 416 | n02776631 bakery, bakeshop, bakehouse
 417 | n02777292 balance beam, beam
 418 | n02782093 balloon
 419 | n02783161 ballpoint, ballpoint pen, ballpen, Biro
 420 | n02786058 Band Aid
 421 | n02787622 banjo
 422 | n02788148 bannister, banister, balustrade, balusters, handrail
 423 | n02790996 barbell
 424 | n02791124 barber chair
 425 | n02791270 barbershop
 426 | n02793495 barn
 427 | n02794156 barometer
 428 | n02795169 barrel, cask
 429 | n02797295 barrow, garden cart, lawn cart, wheelbarrow
 430 | n02799071 baseball
 431 | n02802426 basketball
 432 | n02804414 bassinet
 433 | n02804610 bassoon
 434 | n02807133 bathing cap, swimming cap
 435 | n02808304 bath towel
 436 | n02808440 bathtub, bathing tub, bath, tub
 437 | n02814533 beach wagon, station wagon, wagon, estate car, beach waggon, station waggon, waggon
 438 | n02814860 beacon, lighthouse, beacon light, pharos
 439 | n02815834 beaker
 440 | n02817516 bearskin, busby, shako
 441 | n02823428 beer bottle
 442 | n02823750 beer glass
 443 | n02825657 bell cote, bell cot
 444 | n02834397 bib
 445 | n02835271 bicycle-built-for-two, tandem bicycle, tandem
 446 | n02837789 bikini, two-piece
 447 | n02840245 binder, ring-binder
 448 | n02841315 binoculars, field glasses, opera glasses
 449 | n02843684 birdhouse
 450 | n02859443 boathouse
 451 | n02860847 bobsled, bobsleigh, bob
 452 | n02865351 bolo tie, bolo, bola tie, bola
 453 | n02869837 bonnet, poke bonnet
 454 | n02870880 bookcase
 455 | n02871525 bookshop, bookstore, bookstall
 456 | n02877765 bottlecap
 457 | n02879718 bow
 458 | n02883205 bow tie, bow-tie, bowtie
 459 | n02892201 brass, memorial tablet, plaque
 460 | n02892767 brassiere, bra, bandeau
 461 | n02894605 breakwater, groin, groyne, mole, bulwark, seawall, jetty
 462 | n02895154 breastplate, aegis, egis
 463 | n02906734 broom
 464 | n02909870 bucket, pail
 465 | n02910353 buckle
 466 | n02916936 bulletproof vest
 467 | n02917067 bullet train, bullet
 468 | n02927161 butcher shop, meat market
 469 | n02930766 cab, hack, taxi, taxicab
 470 | n02939185 caldron, cauldron
 471 | n02948072 candle, taper, wax light
 472 | n02950826 cannon
 473 | n02951358 canoe
 474 | n02951585 can opener, tin opener
 475 | n02963159 cardigan
 476 | n02965783 car mirror
 477 | n02966193 carousel, carrousel, merry-go-round, roundabout, whirligig
 478 | n02966687 carpenter's kit, tool kit
 479 | n02971356 carton
 480 | n02974003 car wheel
 481 | n02977058 cash machine, cash dispenser, automated teller machine, automatic teller machine, automated teller, automatic teller, ATM
 482 | n02978881 cassette
 483 | n02979186 cassette player
 484 | n02980441 castle
 485 | n02981792 catamaran
 486 | n02988304 CD player
 487 | n02992211 cello, violoncello
 488 | n02992529 cellular telephone, cellular phone, cellphone, cell, mobile phone
 489 | n02999410 chain
 490 | n03000134 chainlink fence
 491 | n03000247 chain mail, ring mail, mail, chain armor, chain armour, ring armor, ring armour
 492 | n03000684 chain saw, chainsaw
 493 | n03014705 chest
 494 | n03016953 chiffonier, commode
 495 | n03017168 chime, bell, gong
 496 | n03018349 china cabinet, china closet
 497 | n03026506 Christmas stocking
 498 | n03028079 church, church building
 499 | n03032252 cinema, movie theater, movie theatre, movie house, picture palace
 500 | n03041632 cleaver, meat cleaver, chopper
 501 | n03042490 cliff dwelling
 502 | n03045698 cloak
 503 | n03047690 clog, geta, patten, sabot
 504 | n03062245 cocktail shaker
 505 | n03063599 coffee mug
 506 | n03063689 coffeepot
 507 | n03065424 coil, spiral, volute, whorl, helix
 508 | n03075370 combination lock
 509 | n03085013 computer keyboard, keypad
 510 | n03089624 confectionery, confectionary, candy store
 511 | n03095699 container ship, containership, container vessel
 512 | n03100240 convertible
 513 | n03109150 corkscrew, bottle screw
 514 | n03110669 cornet, horn, trumpet, trump
 515 | n03124043 cowboy boot
 516 | n03124170 cowboy hat, ten-gallon hat
 517 | n03125729 cradle
 518 | n03126707 crane
 519 | n03127747 crash helmet
 520 | n03127925 crate
 521 | n03131574 crib, cot
 522 | n03133878 Crock Pot
 523 | n03134739 croquet ball
 524 | n03141823 crutch
 525 | n03146219 cuirass
 526 | n03160309 dam, dike, dyke
 527 | n03179701 desk
 528 | n03180011 desktop computer
 529 | n03187595 dial telephone, dial phone
 530 | n03188531 diaper, nappy, napkin
 531 | n03196217 digital clock
 532 | n03197337 digital watch
 533 | n03201208 dining table, board
 534 | n03207743 dishrag, dishcloth
 535 | n03207941 dishwasher, dish washer, dishwashing machine
 536 | n03208938 disk brake, disc brake
 537 | n03216828 dock, dockage, docking facility
 538 | n03218198 dogsled, dog sled, dog sleigh
 539 | n03220513 dome
 540 | n03223299 doormat, welcome mat
 541 | n03240683 drilling platform, offshore rig
 542 | n03249569 drum, membranophone, tympan
 543 | n03250847 drumstick
 544 | n03255030 dumbbell
 545 | n03259280 Dutch oven
 546 | n03271574 electric fan, blower
 547 | n03272010 electric guitar
 548 | n03272562 electric locomotive
 549 | n03290653 entertainment center
 550 | n03291819 envelope
 551 | n03297495 espresso maker
 552 | n03314780 face powder
 553 | n03325584 feather boa, boa
 554 | n03337140 file, file cabinet, filing cabinet
 555 | n03344393 fireboat
 556 | n03345487 fire engine, fire truck
 557 | n03347037 fire screen, fireguard
 558 | n03355925 flagpole, flagstaff
 559 | n03372029 flute, transverse flute
 560 | n03376595 folding chair
 561 | n03379051 football helmet
 562 | n03384352 forklift
 563 | n03388043 fountain
 564 | n03388183 fountain pen
 565 | n03388549 four-poster
 566 | n03393912 freight car
 567 | n03394916 French horn, horn
 568 | n03400231 frying pan, frypan, skillet
 569 | n03404251 fur coat
 570 | n03417042 garbage truck, dustcart
 571 | n03424325 gasmask, respirator, gas helmet
 572 | n03425413 gas pump, gasoline pump, petrol pump, island dispenser
 573 | n03443371 goblet
 574 | n03444034 go-kart
 575 | n03445777 golf ball
 576 | n03445924 golfcart, golf cart
 577 | n03447447 gondola
 578 | n03447721 gong, tam-tam
 579 | n03450230 gown
 580 | n03452741 grand piano, grand
 581 | n03457902 greenhouse, nursery, glasshouse
 582 | n03459775 grille, radiator grille
 583 | n03461385 grocery store, grocery, food market, market
 584 | n03467068 guillotine
 585 | n03476684 hair slide
 586 | n03476991 hair spray
 587 | n03478589 half track
 588 | n03481172 hammer
 589 | n03482405 hamper
 590 | n03483316 hand blower, blow dryer, blow drier, hair dryer, hair drier
 591 | n03485407 hand-held computer, hand-held microcomputer
 592 | n03485794 handkerchief, hankie, hanky, hankey
 593 | n03492542 hard disc, hard disk, fixed disk
 594 | n03494278 harmonica, mouth organ, harp, mouth harp
 595 | n03495258 harp
 596 | n03496892 harvester, reaper
 597 | n03498962 hatchet
 598 | n03527444 holster
 599 | n03529860 home theater, home theatre
 600 | n03530642 honeycomb
 601 | n03532672 hook, claw
 602 | n03534580 hoopskirt, crinoline
 603 | n03535780 horizontal bar, high bar
 604 | n03538406 horse cart, horse-cart
 605 | n03544143 hourglass
 606 | n03584254 iPod
 607 | n03584829 iron, smoothing iron
 608 | n03590841 jack-o'-lantern
 609 | n03594734 jean, blue jean, denim
 610 | n03594945 jeep, landrover
 611 | n03595614 jersey, T-shirt, tee shirt
 612 | n03598930 jigsaw puzzle
 613 | n03599486 jinrikisha, ricksha, rickshaw
 614 | n03602883 joystick
 615 | n03617480 kimono
 616 | n03623198 knee pad
 617 | n03627232 knot
 618 | n03630383 lab coat, laboratory coat
 619 | n03633091 ladle
 620 | n03637318 lampshade, lamp shade
 621 | n03642806 laptop, laptop computer
 622 | n03649909 lawn mower, mower
 623 | n03657121 lens cap, lens cover
 624 | n03658185 letter opener, paper knife, paperknife
 625 | n03661043 library
 626 | n03662601 lifeboat
 627 | n03666591 lighter, light, igniter, ignitor
 628 | n03670208 limousine, limo
 629 | n03673027 liner, ocean liner
 630 | n03676483 lipstick, lip rouge
 631 | n03680355 Loafer
 632 | n03690938 lotion
 633 | n03691459 loudspeaker, speaker, speaker unit, loudspeaker system, speaker system
 634 | n03692522 loupe, jeweler's loupe
 635 | n03697007 lumbermill, sawmill
 636 | n03706229 magnetic compass
 637 | n03709823 mailbag, postbag
 638 | n03710193 mailbox, letter box
 639 | n03710637 maillot
 640 | n03710721 maillot, tank suit
 641 | n03717622 manhole cover
 642 | n03720891 maraca
 643 | n03721384 marimba, xylophone
 644 | n03724870 mask
 645 | n03729826 matchstick
 646 | n03733131 maypole
 647 | n03733281 maze, labyrinth
 648 | n03733805 measuring cup
 649 | n03742115 medicine chest, medicine cabinet
 650 | n03743016 megalith, megalithic structure
 651 | n03759954 microphone, mike
 652 | n03761084 microwave, microwave oven
 653 | n03763968 military uniform
 654 | n03764736 milk can
 655 | n03769881 minibus
 656 | n03770439 miniskirt, mini
 657 | n03770679 minivan
 658 | n03773504 missile
 659 | n03775071 mitten
 660 | n03775546 mixing bowl
 661 | n03776460 mobile home, manufactured home
 662 | n03777568 Model T
 663 | n03777754 modem
 664 | n03781244 monastery
 665 | n03782006 monitor
 666 | n03785016 moped
 667 | n03786901 mortar
 668 | n03787032 mortarboard
 669 | n03788195 mosque
 670 | n03788365 mosquito net
 671 | n03791053 motor scooter, scooter
 672 | n03792782 mountain bike, all-terrain bike, off-roader
 673 | n03792972 mountain tent
 674 | n03793489 mouse, computer mouse
 675 | n03794056 mousetrap
 676 | n03796401 moving van
 677 | n03803284 muzzle
 678 | n03804744 nail
 679 | n03814639 neck brace
 680 | n03814906 necklace
 681 | n03825788 nipple
 682 | n03832673 notebook, notebook computer
 683 | n03837869 obelisk
 684 | n03838899 oboe, hautboy, hautbois
 685 | n03840681 ocarina, sweet potato
 686 | n03841143 odometer, hodometer, mileometer, milometer
 687 | n03843555 oil filter
 688 | n03854065 organ, pipe organ
 689 | n03857828 oscilloscope, scope, cathode-ray oscilloscope, CRO
 690 | n03866082 overskirt
 691 | n03868242 oxcart
 692 | n03868863 oxygen mask
 693 | n03871628 packet
 694 | n03873416 paddle, boat paddle
 695 | n03874293 paddlewheel, paddle wheel
 696 | n03874599 padlock
 697 | n03876231 paintbrush
 698 | n03877472 pajama, pyjama, pj's, jammies
 699 | n03877845 palace
 700 | n03884397 panpipe, pandean pipe, syrinx
 701 | n03887697 paper towel
 702 | n03888257 parachute, chute
 703 | n03888605 parallel bars, bars
 704 | n03891251 park bench
 705 | n03891332 parking meter
 706 | n03895866 passenger car, coach, carriage
 707 | n03899768 patio, terrace
 708 | n03902125 pay-phone, pay-station
 709 | n03903868 pedestal, plinth, footstall
 710 | n03908618 pencil box, pencil case
 711 | n03908714 pencil sharpener
 712 | n03916031 perfume, essence
 713 | n03920288 Petri dish
 714 | n03924679 photocopier
 715 | n03929660 pick, plectrum, plectron
 716 | n03929855 pickelhaube
 717 | n03930313 picket fence, paling
 718 | n03930630 pickup, pickup truck
 719 | n03933933 pier
 720 | n03935335 piggy bank, penny bank
 721 | n03937543 pill bottle
 722 | n03938244 pillow
 723 | n03942813 ping-pong ball
 724 | n03944341 pinwheel
 725 | n03947888 pirate, pirate ship
 726 | n03950228 pitcher, ewer
 727 | n03954731 plane, carpenter's plane, woodworking plane
 728 | n03956157 planetarium
 729 | n03958227 plastic bag
 730 | n03961711 plate rack
 731 | n03967562 plow, plough
 732 | n03970156 plunger, plumber's helper
 733 | n03976467 Polaroid camera, Polaroid Land camera
 734 | n03976657 pole
 735 | n03977966 police van, police wagon, paddy wagon, patrol wagon, wagon, black Maria
 736 | n03980874 poncho
 737 | n03982430 pool table, billiard table, snooker table
 738 | n03983396 pop bottle, soda bottle
 739 | n03991062 pot, flowerpot
 740 | n03992509 potter's wheel
 741 | n03995372 power drill
 742 | n03998194 prayer rug, prayer mat
 743 | n04004767 printer
 744 | n04005630 prison, prison house
 745 | n04008634 projectile, missile
 746 | n04009552 projector
 747 | n04019541 puck, hockey puck
 748 | n04023962 punching bag, punch bag, punching ball, punchball
 749 | n04026417 purse
 750 | n04033901 quill, quill pen
 751 | n04033995 quilt, comforter, comfort, puff
 752 | n04037443 racer, race car, racing car
 753 | n04039381 racket, racquet
 754 | n04040759 radiator
 755 | n04041544 radio, wireless
 756 | n04044716 radio telescope, radio reflector
 757 | n04049303 rain barrel
 758 | n04065272 recreational vehicle, RV, R.V.
 759 | n04067472 reel
 760 | n04069434 reflex camera
 761 | n04070727 refrigerator, icebox
 762 | n04074963 remote control, remote
 763 | n04081281 restaurant, eating house, eating place, eatery
 764 | n04086273 revolver, six-gun, six-shooter
 765 | n04090263 rifle
 766 | n04099969 rocking chair, rocker
 767 | n04111531 rotisserie
 768 | n04116512 rubber eraser, rubber, pencil eraser
 769 | n04118538 rugby ball
 770 | n04118776 rule, ruler
 771 | n04120489 running shoe
 772 | n04125021 safe
 773 | n04127249 safety pin
 774 | n04131690 saltshaker, salt shaker
 775 | n04133789 sandal
 776 | n04136333 sarong
 777 | n04141076 sax, saxophone
 778 | n04141327 scabbard
 779 | n04141975 scale, weighing machine
 780 | n04146614 school bus
 781 | n04147183 schooner
 782 | n04149813 scoreboard
 783 | n04152593 screen, CRT screen
 784 | n04153751 screw
 785 | n04154565 screwdriver
 786 | n04162706 seat belt, seatbelt
 787 | n04179913 sewing machine
 788 | n04192698 shield, buckler
 789 | n04200800 shoe shop, shoe-shop, shoe store
 790 | n04201297 shoji
 791 | n04204238 shopping basket
 792 | n04204347 shopping cart
 793 | n04208210 shovel
 794 | n04209133 shower cap
 795 | n04209239 shower curtain
 796 | n04228054 ski
 797 | n04229816 ski mask
 798 | n04235860 sleeping bag
 799 | n04238763 slide rule, slipstick
 800 | n04239074 sliding door
 801 | n04243546 slot, one-armed bandit
 802 | n04251144 snorkel
 803 | n04252077 snowmobile
 804 | n04252225 snowplow, snowplough
 805 | n04254120 soap dispenser
 806 | n04254680 soccer ball
 807 | n04254777 sock
 808 | n04258138 solar dish, solar collector, solar furnace
 809 | n04259630 sombrero
 810 | n04263257 soup bowl
 811 | n04264628 space bar
 812 | n04265275 space heater
 813 | n04266014 space shuttle
 814 | n04270147 spatula
 815 | n04273569 speedboat
 816 | n04275548 spider web, spider's web
 817 | n04277352 spindle
 818 | n04285008 sports car, sport car
 819 | n04286575 spotlight, spot
 820 | n04296562 stage
 821 | n04310018 steam locomotive
 822 | n04311004 steel arch bridge
 823 | n04311174 steel drum
 824 | n04317175 stethoscope
 825 | n04325704 stole
 826 | n04326547 stone wall
 827 | n04328186 stopwatch, stop watch
 828 | n04330267 stove
 829 | n04332243 strainer
 830 | n04335435 streetcar, tram, tramcar, trolley, trolley car
 831 | n04336792 stretcher
 832 | n04344873 studio couch, day bed
 833 | n04346328 stupa, tope
 834 | n04347754 submarine, pigboat, sub, U-boat
 835 | n04350905 suit, suit of clothes
 836 | n04355338 sundial
 837 | n04355933 sunglass
 838 | n04356056 sunglasses, dark glasses, shades
 839 | n04357314 sunscreen, sunblock, sun blocker
 840 | n04366367 suspension bridge
 841 | n04367480 swab, swob, mop
 842 | n04370456 sweatshirt
 843 | n04371430 swimming trunks, bathing trunks
 844 | n04371774 swing
 845 | n04372370 switch, electric switch, electrical switch
 846 | n04376876 syringe
 847 | n04380533 table lamp
 848 | n04389033 tank, army tank, armored combat vehicle, armoured combat vehicle
 849 | n04392985 tape player
 850 | n04398044 teapot
 851 | n04399382 teddy, teddy bear
 852 | n04404412 television, television system
 853 | n04409515 tennis ball
 854 | n04417672 thatch, thatched roof
 855 | n04418357 theater curtain, theatre curtain
 856 | n04423845 thimble
 857 | n04428191 thresher, thrasher, threshing machine
 858 | n04429376 throne
 859 | n04435653 tile roof
 860 | n04442312 toaster
 861 | n04443257 tobacco shop, tobacconist shop, tobacconist
 862 | n04447861 toilet seat
 863 | n04456115 torch
 864 | n04458633 totem pole
 865 | n04461696 tow truck, tow car, wrecker
 866 | n04462240 toyshop
 867 | n04465501 tractor
 868 | n04467665 trailer truck, tractor trailer, trucking rig, rig, articulated lorry, semi
 869 | n04476259 tray
 870 | n04479046 trench coat
 871 | n04482393 tricycle, trike, velocipede
 872 | n04483307 trimaran
 873 | n04485082 tripod
 874 | n04486054 triumphal arch
 875 | n04487081 trolleybus, trolley coach, trackless trolley
 876 | n04487394 trombone
 877 | n04493381 tub, vat
 878 | n04501370 turnstile
 879 | n04505470 typewriter keyboard
 880 | n04507155 umbrella
 881 | n04509417 unicycle, monocycle
 882 | n04515003 upright, upright piano
 883 | n04517823 vacuum, vacuum cleaner
 884 | n04522168 vase
 885 | n04523525 vault
 886 | n04525038 velvet
 887 | n04525305 vending machine
 888 | n04532106 vestment
 889 | n04532670 viaduct
 890 | n04536866 violin, fiddle
 891 | n04540053 volleyball
 892 | n04542943 waffle iron
 893 | n04548280 wall clock
 894 | n04548362 wallet, billfold, notecase, pocketbook
 895 | n04550184 wardrobe, closet, press
 896 | n04552348 warplane, military plane
 897 | n04553703 washbasin, handbasin, washbowl, lavabo, wash-hand basin
 898 | n04554684 washer, automatic washer, washing machine
 899 | n04557648 water bottle
 900 | n04560804 water jug
 901 | n04562935 water tower
 902 | n04579145 whiskey jug
 903 | n04579432 whistle
 904 | n04584207 wig
 905 | n04589890 window screen
 906 | n04590129 window shade
 907 | n04591157 Windsor tie
 908 | n04591713 wine bottle
 909 | n04592741 wing
 910 | n04596742 wok
 911 | n04597913 wooden spoon
 912 | n04599235 wool, woolen, woollen
 913 | n04604644 worm fence, snake fence, snake-rail fence, Virginia fence
 914 | n04606251 wreck
 915 | n04612504 yawl
 916 | n04613696 yurt
 917 | n06359193 web site, website, internet site, site
 918 | n06596364 comic book
 919 | n06785654 crossword puzzle, crossword
 920 | n06794110 street sign
 921 | n06874185 traffic light, traffic signal, stoplight
 922 | n07248320 book jacket, dust cover, dust jacket, dust wrapper
 923 | n07565083 menu
 924 | n07579787 plate
 925 | n07583066 guacamole
 926 | n07584110 consomme
 927 | n07590611 hot pot, hotpot
 928 | n07613480 trifle
 929 | n07614500 ice cream, icecream
 930 | n07615774 ice lolly, lolly, lollipop, popsicle
 931 | n07684084 French loaf
 932 | n07693725 bagel, beigel
 933 | n07695742 pretzel
 934 | n07697313 cheeseburger
 935 | n07697537 hotdog, hot dog, red hot
 936 | n07711569 mashed potato
 937 | n07714571 head cabbage
 938 | n07714990 broccoli
 939 | n07715103 cauliflower
 940 | n07716358 zucchini, courgette
 941 | n07716906 spaghetti squash
 942 | n07717410 acorn squash
 943 | n07717556 butternut squash
 944 | n07718472 cucumber, cuke
 945 | n07718747 artichoke, globe artichoke
 946 | n07720875 bell pepper
 947 | n07730033 cardoon
 948 | n07734744 mushroom
 949 | n07742313 Granny Smith
 950 | n07745940 strawberry
 951 | n07747607 orange
 952 | n07749582 lemon
 953 | n07753113 fig
 954 | n07753275 pineapple, ananas
 955 | n07753592 banana
 956 | n07754684 jackfruit, jak, jack
 957 | n07760859 custard apple
 958 | n07768694 pomegranate
 959 | n07802026 hay
 960 | n07831146 carbonara
 961 | n07836838 chocolate sauce, chocolate syrup
 962 | n07860988 dough
 963 | n07871810 meat loaf, meatloaf
 964 | n07873807 pizza, pizza pie
 965 | n07875152 potpie
 966 | n07880968 burrito
 967 | n07892512 red wine
 968 | n07920052 espresso
 969 | n07930864 cup
 970 | n07932039 eggnog
 971 | n09193705 alp
 972 | n09229709 bubble
 973 | n09246464 cliff, drop, drop-off
 974 | n09256479 coral reef
 975 | n09288635 geyser
 976 | n09332890 lakeside, lakeshore
 977 | n09399592 promontory, headland, head, foreland
 978 | n09421951 sandbar, sand bar
 979 | n09428293 seashore, coast, seacoast, sea-coast
 980 | n09468604 valley, vale
 981 | n09472597 volcano
 982 | n09835506 ballplayer, baseball player
 983 | n10148035 groom, bridegroom
 984 | n10565667 scuba diver
 985 | n11879895 rapeseed
 986 | n11939491 daisy
 987 | n12057211 yellow lady's slipper, yellow lady-slipper, Cypripedium calceolus, Cypripedium parviflorum
 988 | n12144580 corn
 989 | n12267677 acorn
 990 | n12620546 hip, rose hip, rosehip
 991 | n12768682 buckeye, horse chestnut, conker
 992 | n12985857 coral fungus
 993 | n12998815 agaric
 994 | n13037406 gyromitra
 995 | n13040303 stinkhorn, carrion fungus
 996 | n13044778 earthstar
 997 | n13052670 hen-of-the-woods, hen of the woods, Polyporus frondosus, Grifola frondosa
 998 | n13054560 bolete
 999 | n13133613 ear, spike, capitulum
1000 | n15075141 toilet tissue, toilet paper, bathroom tissue
1001 | 


--------------------------------------------------------------------------------
/matlab/data/test_img.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/matlab/data/test_img.png


--------------------------------------------------------------------------------
/matlab/demo_caffe.m:
--------------------------------------------------------------------------------
  1 | clear; clc; close all;
  2 | 
  3 | DEVICE_ID = 0; %set gpu id starting from 0
  4 | 
  5 | % set paths to Caffe and DeepFool
  6 | PATH_CAFFE = '/path/to/caffe/matlab';
  7 | PATH_DEEPFOOL = '/path/to/DeepFool';
  8 | PATH_IMAGENET_TRAIN = '/path/to/ILSVRC2012/train';
  9 | 
 10 | addpath(PATH_CAFFE);
 11 | addpath(PATH_DEEPFOOL);
 12 | 
 13 | caffe.set_mode_gpu();
 14 | caffe.set_device(DEVICE_ID);
 15 | 
 16 | model = 'caffenet';
 17 | 
 18 | if (strcmp(model, 'vgg_16'))
 19 |     fprintf('Loading VGG 16\n');
 20 |     net_model_path = fullfile('data', 'deploy_vgg_16.prototxt');
 21 |     net_weights_path = fullfile('data', 'vgg_16.caffemodel');
 22 |     net_url = 'http://www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_16_layers.caffemodel';
 23 | elseif (strcmp(model, 'vgg_19'))
 24 |     fprintf('Loading VGG 19\n');
 25 |     net_model_path = fullfile('data', 'deploy_vgg_19.prototxt');
 26 |     net_weights_path = fullfile('data', 'vgg_19.caffemodel');
 27 |     net_url = 'http://www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_19_layers.caffemodel';
 28 | elseif (strcmp(model, 'googlenet'))
 29 |     fprintf('Loading GoogLeNet\n');
 30 |     net_model_path = fullfile('data', 'deploy_googlenet.prototxt');
 31 |     net_weights_path = fullfile('data', 'googlenet.caffemodel');
 32 |     net_url = 'http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel';
 33 | elseif (strcmp(model, 'vgg_f'))
 34 |     fprintf('Loading VGG-F\n');
 35 |     net_model_path = fullfile('data', 'deploy_vgg_f.prototxt');
 36 |     net_weights_path = fullfile('data', 'googlenet.caffemodel');
 37 |     net_url = 'http://dl.caffe.berkeleyvision.org/bvlc_vgg_f.caffemodel';
 38 | elseif (strcmp(model, 'caffenet'))
 39 |     fprintf('Loading CaffeNet\n');
 40 |     net_model_path = fullfile('data', 'deploy_caffenet.prototxt');
 41 |     net_weights_path = fullfile('data', 'caffenet.caffemodel');
 42 |     net_url = 'http://dl.caffe.berkeleyvision.org/bvlc_reference_caffenet.caffemodel';
 43 | elseif (strcmp(model, 'resnet-152'))
 44 |     fprintf('Loading ResNet-152\n');
 45 |     net_model_path = fullfile('data', 'deploy_resnet.prototxt');
 46 |     net_weights_path = fullfile('data', 'caffenet.caffemodel');
 47 |     net_url = 'https://deepdetect.com/models/resnet/ResNet-152-model.caffemodel';
 48 | else
 49 |     error('Model is not recognized!');
 50 | end
 51 | 
 52 | if (~exist(net_weights_path, 'file'))
 53 |     fprintf('Downloading model...\n');
 54 |     websave(net_weights_path, net_url);
 55 | end
 56 |     
 57 | 
 58 | net = caffe.Net(net_model_path, net_weights_path, 'test'); % run with phase test (so that dropout isn't applied)
 59 | fprintf('Network is loaded\n');
 60 | 
 61 | % Loading the data
 62 | fprintf('Loading the data...\n');
 63 | im_array = makeImagenetData(PATH_IMAGENET_TRAIN);
 64 | fprintf('Data is loaded\n');
 65 | 
 66 | % set options
 67 | opts.library = 'caffe';
 68 | opts.net = net;
 69 | 
 70 | v = universal_perturbation(im_array, opts);
 71 | 
 72 | % Test perturbation on a sample image
 73 | 
 74 | d = load(fullfile('data', 'ilsvrc_2012_mean.mat'));
 75 | mean_data = d.mean_data;
 76 | 
 77 | labels_file = fullfile('data', 'synset_words.txt');
 78 | fileID = fopen(labels_file);
 79 | C = textscan(fileID, '%c %d %s', 'Delimiter', '\n');
 80 | synset_names = C{3};
 81 | synset_names_short = synset_names;
 82 | for i = 1:1000
 83 |     synset_names_short(i) = strtok(synset_names(i), ',');
 84 | end
 85 | 
 86 | im_test = imread(fullfile('data', 'test_img.png'));
 87 | im_test = preprocess_img(im_test, mean_data);
 88 | 
 89 | original_label = predict_caffe(im_test, net, 1);
 90 | perturbed_label = predict_caffe(im_test+v, net, 1);
 91 | 
 92 | % Show original and perturbed images
 93 | im_ = im_test + mean_data(1:224, 1:224, :);
 94 | im_ = permute(im_,[2,1,3]);
 95 | im_ = im_(:,:,[3,2,1],:);
 96 | 
 97 | figure; imshow(uint8(im_));
 98 | xlabel(synset_names_short(original_label));
 99 | set(gca,'xtick',[],'ytick',[])
100 | set(gcf, 'Color', 'white');
101 | set(gca, 'FontSize', 15);
102 | title('Original image');
103 | 
104 | v_ = permute(v,[2,1,3]);
105 | v_ = v_(:,:,[3,2,1],:);
106 | figure; imshow(uint8(im_ + v_));
107 | xlabel(synset_names_short(perturbed_label));
108 | set(gca,'xtick',[],'ytick',[])
109 | set(gcf, 'Color', 'white');
110 | set(gca, 'FontSize', 15);
111 | title('Perturbed image');
112 | 


--------------------------------------------------------------------------------
/matlab/makeImagenetData.m:
--------------------------------------------------------------------------------
 1 | function im_array = makeImagenetData(imagenet_path)
 2 |     
 3 |     if (exist(fullfile('data', 'ImageNet.h5'), 'file'))
 4 |         im_array = h5read(fullfile('data', 'ImageNet.h5'), ['/batch',num2str(1)]);
 5 |         return;
 6 |     end
 7 |     
 8 |     d = load(fullfile('data', 'ilsvrc_2012_mean.mat'));
 9 |     mean_data = d.mean_data;
10 |     
11 |     num_class = 1000;
12 |     num_images = 10000;
13 |     num_per_class = num_images/num_class;
14 | 
15 |     allLabels = dir( [imagenet_path] );
16 |     class_name = {allLabels(3:end).name};
17 | 
18 |     im_array = zeros(224,224,3,num_images,'single');
19 | 
20 |     for j=1:num_class
21 |         fprintf('Processing data %d\n', j);
22 |         ff = fullfile(imagenet_path, class_name{j});
23 |         allFiles = dir( ff );
24 |         fullfile(imagenet_path, class_name{j})
25 |         allNames = { allFiles.name };
26 |         for i=1:num_per_class
27 |             im = imread(char(fullfile(ff, allNames(i+3))));
28 |           if size(im,3)==1
29 |             im = repmat(im,1,1,3);
30 |           end
31 |           im_array_par(:,:,:,i) = preprocess_img(im, mean_data);
32 |         end
33 |         im_array(:,:,:,(j-1)*num_per_class+(1:num_per_class)) = im_array_par;
34 |     end
35 |     
36 |     % Save data
37 |     % This can take a significant amount of space. Comment if needed.
38 |     fprintf('Saving the data...\n');
39 |     h5create(fullfile('data', 'ImageNet.h5'),['/batch',num2str(1)],[224 224 3 10000],'DataType','single');
40 |     h5write(fullfile('data', 'ImageNet.h5'), ['/batch',num2str(1)], im_array);


--------------------------------------------------------------------------------
/matlab/predict_caffe.m:
--------------------------------------------------------------------------------
 1 | function [l,out] = predict_caffe(im,net,BATCH_SIZE)
 2 | % Computes the labels of im predicted by net
 3 | 
 4 | blob_shape = net.blobs('data').shape(); % store the original input size
 5 | n = size(im,4);
 6 | 
 7 | change_input_blob(net,BATCH_SIZE); % rehsape the input batch size
 8 | 
 9 | K = floor(n/BATCH_SIZE);
10 | R = rem(n,BATCH_SIZE);
11 | 
12 | for k=1:K
13 |     res = net.forward({im(:,:,:,(k-1)*BATCH_SIZE+(1:BATCH_SIZE))});
14 |     res2(:,((k-1)*BATCH_SIZE+1):k*BATCH_SIZE)=res{1};
15 | end
16 | 
17 | if(R~=0)
18 |     change_input_blob(net,R);
19 |     res = net.forward({im(:,:,:,K*BATCH_SIZE+(1:R))});
20 |     res2(:,(K*BATCH_SIZE+1):K*BATCH_SIZE+R)=res{1};
21 | end
22 | [~,l] = max(res2,[],1);
23 | out = res2;
24 | 
25 | net.blobs('data').reshape(blob_shape); % reshape to the original input size
26 | net.reshape();
27 | end
28 | 


--------------------------------------------------------------------------------
/matlab/predict_matconvnet.m:
--------------------------------------------------------------------------------
 1 | function [ys, scores] = predict_matconvnet(ims, net, batchsize)
 2 | 
 3 | ys = zeros(size(ims, 4), 1);
 4 | scores = zeros(1000, size(ims, 4)); % Number of classes
 5 | for b = 1:batchsize:size(ims, 4)
 6 |     res = vl_simplenn(net, gpuArray(ims(:, :, :, b:(b+batchsize-1))), [], [], 'Mode', 'test');
 7 |     
 8 |     cc = squeeze(gather(res(end).x));
 9 |     scores(:, b:(b+batchsize-1)) = cc;
10 |     [~, ys(b:(b+batchsize-1))] = max(squeeze(gather(res(end).x)));
11 | end
12 | 
13 | end


--------------------------------------------------------------------------------
/matlab/preprocess_img.m:
--------------------------------------------------------------------------------
 1 | function imgs_out = preprocess_img(imgs, mean_data)
 2 | 
 3 |     IMAGE_DIM = 256;
 4 | 
 5 |     if size(imgs,3)==1
 6 |         imgs = repmat(imgs,1,1,3);
 7 |     end
 8 |     im_ = imgs(:, :, [3, 2, 1]);  % permute channels from RGB to BGR
 9 |     im_ = permute(im_, [2, 1, 3]);  % flip width and height
10 |     im_ = single(im_);  % convert from uint8 to single
11 |     im_ = imresize(im_, [IMAGE_DIM IMAGE_DIM], 'bilinear');  % resize im_data
12 |     im_ = im_ - mean_data;  % subtract mean_data (already in W x H x C, BGR)
13 |     imgs_out = im_(1:224,1:224,:);
14 |     
15 | end


--------------------------------------------------------------------------------
/matlab/readme.md:
--------------------------------------------------------------------------------
 1 | #MATLAB
 2 | 
 3 | ###universal_perturbation.m
 4 | 
 5 | This function implements the algorithm proposed in [[1]](http://arxiv.org/pdf/1610.08401) to find a universal perturbation.
 6 | 
 7 | ###Inputs
 8 | - `dataset`: images in `W*H*C*N` format, where `W`:width, `H`:height, `C`:channels (RGB), and `N`:number of images.
 9 | - `opts`: A structure containing the following optionals and non-optionals parameters:
10 |   - `library`: framework to be used: Caffe or MatConvNet. __(Obligatory)__
11 |   - `net`: corresponding network structure. __(Obligatory)__
12 |   - `delta`: controls the desired fooling rate (default = 80% fooling rate).
13 |   - `MAX_ITER_UNIV`: termination criterion (maximum number of iterations, default = Inf).
14 |   - `xi`: controls the l_p magnitude of the perturbation (default = 10).
15 |   - `p`: norm to be used (FOR NOW, ONLY p = 2, and p = Inf ARE ACCEPTED!) (default = Inf).
16 |   - `df`: DeepFool's parameters (see [DeepFool](http://github.com/lts4/deepfool) for more information).
17 |   
18 | ###Output
19 | - `v`: a `W*H*C` perturbation image.
20 | 
21 | ##Usage
22 | 
23 | A demo using Caffe is provided in `demo_caffe.m`.
24 | 
25 | First of all, you need to setup the paths to [DeepFool](http://github.com/lts4/deepfool) and [Caffe](http://caffe.berkeleyvision.org) (or [MatConvNet](http://www.vlfeat.org/matconvnet/)), e.g.:
26 | ```
27 | addpath('./DeepFool');
28 | addpath('./caffe/matlab');
29 | ```
30 | Next, you have to load your pre-trained model. __Note that the last layer (usually softmax layer) should have been removed (see [DeepFool](http://github.com/lts4/deepfool) for more information).__ We provide in the `data` folder the prototxt files for the common networks, without the softmax layer. The pre-trained model can be loaded as follows, using Caffe:
31 | ```
32 | caffe.set_mode_gpu(); % initialize Caffe in gpu mode
33 | caffe.set_device(0); % choose the first gpu
34 | 
35 | net_model = 'deploy_googlenet.prototxt'; % GoogLeNet without softmax layer
36 | net_weights = 'googlenet.caffemodel'; % weights
37 | net = caffe.Net(net_model, net_weights, 'test'); % run with phase test
38 | ```
39 | After loading your pre-trained model using your preferred framework, you have to load the set of images to compute the universal perturbation. __Be careful that the images should be pre-processed in the same way that the training images are pre-processed.__ We provide an example script in `makeImagenetData.m`, where 10,000 training images are used. Once you have your data, you can load it and compute a universal perturbation as follows:
40 | 
41 | ```
42 | % Load data
43 | dataset = h5read(fullfile('data', 'ImageNet.h5'), ['/batch1']);
44 | % Set parameters
45 | opts.library = 'caffe' % or 'matconvnet';
46 | opts.net = net;
47 | % Compute universal perturbation
48 | v = universal_perturbation(dataset, opts);
49 | ```
50 | 
51 | ##Reference
52 | [1] S. Moosavi-Dezfooli\*, A. Fawzi\*, O. Fawzi, P. Frossard:
53 | [*Universal adversarial perturbations*](http://arxiv.org/pdf/1610.08401), CVPR 2017.
54 | 


--------------------------------------------------------------------------------
/matlab/universal_perturbation.m:
--------------------------------------------------------------------------------
 1 | %%
 2 | %   MATLAB code for Universal Perturbation
 3 | %
 4 | %   universal_perturbation(dataset, varargin):
 5 | %   computes a universal perturbation for a Caffe's or MatConvNet's model
 6 | %
 7 | %   INPUTS 
 8 | %   dataset: images in W*H*C*N format  
 9 | %   opts: A struct containing the following parameters:
10 | %     delta: controls the desired fooling rate (default = 80% fooling rate)
11 | %     MAX_ITER_UNIV: optional other termination criterion (maximum number of iteration, default = Inf)
12 | %     xi: controls the l_p magnitude of the perturbation (default = 10)
13 | %     p: norm to be used (FOR NOW, ONLY p = 2, and p = Inf ARE ACCEPTED!) (default = Inf)
14 | %     df.labels_limit: DeepFool's labels_limit (limits the number of classes to test against, by default = 10)
15 | %     df.overshoot: DeepFool's overshoot (used as a termination criterion to prevent vanishing updates, by default = 0.02)
16 | %     df.MAX_ITER: DeepFool's MAX_ITER (maximum number of iterations for DeepFool , by default = 10)
17 | %
18 | %   OUTPUT
19 | %   v: universal perturbation
20 | %
21 | %   S. Moosavi-Dezfooli, A. Fawzi, O.Fawzi, P. Frossard: Universal Adversarial Perturbations, arXiv:1610.08401.
22 | %%
23 | function v = universal_perturbation(dataset, varargin)
24 | 
25 | %Options
26 | opts.delta=0.2;
27 | opts.MAX_ITER_UNIV = 10;
28 | opts.xi = 10;
29 | opts.p = Inf;
30 | opts.BATCH_SIZE = 100;
31 | opts.df.labels_limit=10;
32 | opts.df.overshoot=0.02;
33 | opts.df.MAX_ITER=10;
34 | opts.library = [];
35 | opts.net = [];
36 | 
37 | opts = vl_argparse(opts, varargin);
38 | 
39 | v = 0;
40 | fooling_rate = 0;
41 | num_images = size(dataset,4);
42 | 
43 | if isempty(opts.net) % check the network
44 |     fprintf('Please specify the network!\nopts.net = net');
45 |     return;
46 | end
47 | 
48 | if(strcmp(opts.library,'caffe')) % check the framework: only Caffe and MatConvNet are supported.
49 |     adversarial_DF = @(x) adversarial_DeepFool_caffe(x,opts.net,opts.df);
50 |     predict = @(x)predict_caffe(x,opts.net,opts.BATCH_SIZE);
51 | elseif(strcmp(opts.library,'matconvnet'))
52 |     adversarial_DF = @(x) adversarial_DeepFool_matconvnet(x,opts.net,opts.df);
53 |     predict = @(x)predict_matconvnet(x,opts.net,opts.BATCH_SIZE);
54 | else
55 |     fprintf('Library is not supported!\n');
56 |     return;
57 | end
58 | 
59 | itr = 0;
60 | while(fooling_rate<(1-opts.delta) && itr<opts.MAX_ITER_UNIV)
61 |     itr = itr + 1;    
62 |     dataset = dataset(:,:,:,randperm(num_images)); % shuffle the dataset
63 |     disp('Data randomly permuted');
64 |     for cnt = 1:num_images % go through the data set and compute the perturbation increments sequentially
65 |         if (predict(dataset(:,:,:,cnt))==predict(dataset(:,:,:,cnt)+v))
66 |             [dv,~,~,iter] = adversarial_DF(dataset(:,:,:,cnt)+v); % compute adversarial perturbation
67 |             if(iter<opts.df.MAX_ITER-1) % check for convergence
68 |                 v = v + dv; % update v
69 |                 v = proj_lp(v, opts.xi, opts.p); % project v on l_p ball
70 |             end
71 |         end
72 |         clc;
73 |         fprintf('Fooling rate for pass %d = %.3f\n', itr-1, fooling_rate);
74 |         fprintf('Iteration %d/%d\n', cnt, num_images);
75 |         
76 |     end
77 |     fprintf('Computing the new fooling rate...\n');
78 |     est_labels_orig = predict(dataset(:,:,:,1:num_images)); % compute the original labels
79 |     est_labels_pert = predict(bsxfun(@plus,dataset(:,:,:,1:num_images),v)); % compute the perturbed labels
80 |     fooling_rate = sum(est_labels_orig~=est_labels_pert)/num_images; % compute the fooling rate
81 |     fprintf('New fooling rate = %f\n', fooling_rate);
82 | end
83 | 
84 | function v_projected = proj_lp(v, xi, p)
85 | % project on the l_p ball centered at 0 and of radius xi
86 | % SUPPORTS only p = 2 and p = Inf for now
87 | 
88 | if p==Inf
89 |     v_projected = sign(v).*min(abs(v),xi);
90 | elseif p==2
91 |     v_projected = v * min(1, xi/norm(v(:)));
92 | end
93 | 


--------------------------------------------------------------------------------
/matlab/vl_argparse.m:
--------------------------------------------------------------------------------
  1 | function [opts, args] = vl_argparse(opts, args, varargin)
  2 | %VL_ARGPARSE Parse list of parameter-value pairs.
  3 | %   OPTS = VL_ARGPARSE(OPTS, ARGS) updates the structure OPTS based on
  4 | %   the specified parameter-value pairs ARGS={PAR1, VAL1, ... PARN,
  5 | %   VALN}. If a parameter PAR cannot be matched to any of the fields
  6 | %   in OPTS, the function generates an error.
  7 | %
  8 | %   Parameters that have a struct value in OPTS are processed
  9 | %   recursively, updating the individual subfields.  This behaviour
 10 | %   can be suppressed by using VL_ARGPARSE(OPTS, ARGS, 'nonrecursive'),
 11 | %   in which case the struct value is copied directly (hence deleting any
 12 | %   existing subfield existing in OPTS). A direct copy occurrs also if the
 13 | %   struct value in OPTS is a structure with no fields. The nonrecursive
 14 | %   mode is especially useful when a processing time is a concern.
 15 | %
 16 | %   One or more of the (PAR, VAL) pairs in the argument list can be
 17 | %   replaced by a structure; in this case, the fields of the structure
 18 | %   are used as paramater names and the field values as parameter
 19 | %   values. This behaviour, while orthogonal to structure-valued parameters,
 20 | %   is also disabled in the 'nonrecursive' mode.
 21 | %
 22 | %   [OPTS, ARGS] = VL_ARGPARSE(OPTS, ARGS) copies any parameter in
 23 | %   ARGS that does not match OPTS back to ARGS instead of producing an
 24 | %   error. Options specified as structures are passed back as a list
 25 | %   of (PAR, VAL) pairs.
 26 | %
 27 | %   Example::
 28 | %     The function can be used to parse a list of arguments
 29 | %     passed to a MATLAB functions:
 30 | %
 31 | %        function myFunction(x,y,z,varargin)
 32 | %        opts.parameterName = defaultValue ;
 33 | %        opts = vl_argparse(opts, varargin)
 34 | %
 35 | %     If only a subset of the options should be parsed, for example
 36 | %     because the other options are interpreted by a subroutine, then
 37 | %     use the form
 38 | %
 39 | %        [opts, varargin] = vl_argparse(opts, varargin)
 40 | %
 41 | %     that copies back to VARARGIN any unknown parameter.
 42 | %
 43 | %   See also: VL_HELP().
 44 | 
 45 | % Copyright (C) 2015-16 Andrea Vedaldi and Karel Lenc.
 46 | % Copyright (C) 2007-12 Andrea Vedaldi and Brian Fulkerson.
 47 | % All rights reserved.
 48 | %
 49 | % Tishis file is part of the VLFeat library and is made available under
 50 | % the terms of the BSD license (see the COPYING file).
 51 | 
 52 | if ~isstruct(opts) && ~isobject(opts), error('OPTS must be a structure') ; end
 53 | if ~iscell(args), args = {args} ; end
 54 | 
 55 | recursive = true ;
 56 | if numel(varargin) == 1
 57 |   if strcmp(lower(varargin{1}), 'nonrecursive') ;
 58 |     recursive = false ;
 59 |   else
 60 |     error('Unknown option specified.') ;
 61 |   end
 62 | end
 63 | if numel(varargin) > 1
 64 |   error('There can be at most one option.') ;
 65 | end
 66 | 
 67 | optNames = fieldnames(opts)' ;
 68 | 
 69 | % convert ARGS into a structure
 70 | ai = 1 ;
 71 | keep = false(size(args)) ;
 72 | while ai <= numel(args)
 73 | 
 74 |   % Check whether the argument is a (param,value) pair or a structure.
 75 |   if recursive && isstruct(args{ai})
 76 |     params = fieldnames(args{ai})' ;
 77 |     values = struct2cell(args{ai})' ;
 78 |     if nargout == 1
 79 |       opts = vl_argparse(opts, vertcat(params,values)) ;
 80 |     else
 81 |       [opts, rest] = vl_argparse(opts, reshape(vertcat(params,values), 1, [])) ;
 82 |       args{ai} = cell2struct(rest(2:2:end), rest(1:2:end), 2) ;
 83 |       keep(ai) = true ;
 84 |     end
 85 |     ai = ai + 1 ;
 86 |     continue ;
 87 |   end
 88 | 
 89 |   if ~isstr(args{ai})
 90 |     error('Expected either a param-value pair or a structure.') ;
 91 |   end
 92 | 
 93 |   param = args{ai} ;
 94 |   value = args{ai+1} ;
 95 | 
 96 |   p = find(strcmpi(param, optNames)) ;
 97 |   if numel(p) ~= 1
 98 |     if nargout == 1
 99 |       error('Unknown parameter ''%s''', param) ;
100 |     else
101 |       keep([ai,ai+1]) = true ;
102 |       ai = ai + 2 ;
103 |       continue ;
104 |     end
105 |   end
106 |   field = optNames{p} ;
107 | 
108 |   if ~recursive
109 |     opts.(field) = value ;
110 |   else
111 |     if isstruct(opts.(field)) && numel(fieldnames(opts.(field))) > 0
112 |       % The parameter has a  non-empty struct value in OPTS:
113 |       % process recursively.
114 |       if ~isstruct(value)
115 |         error('Cannot assign a non-struct value to the struct parameter ''%s''.', ...
116 |           field) ;
117 |       end
118 |       if nargout > 1
119 |         [opts.(field), args{ai+1}] = vl_argparse(opts.(field), value) ;
120 |       else
121 |         opts.(field) = vl_argparse(opts.(field), value) ;
122 |       end
123 |     else
124 |       % The parameter does not have a struct value in OPTS: copy as is.
125 |       opts.(field) = value ;
126 |     end
127 |   end
128 | 
129 |   ai = ai + 2 ;
130 | end
131 | 
132 | args = args(keep) ;
133 | 


--------------------------------------------------------------------------------
/precomputed/CaffeNet.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/precomputed/CaffeNet.mat


--------------------------------------------------------------------------------
/precomputed/GoogLeNet.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/precomputed/GoogLeNet.mat


--------------------------------------------------------------------------------
/precomputed/README.md:
--------------------------------------------------------------------------------
1 | #Precomputed perturbations
2 | 
3 | Precomputed universal perturbations for different classification models. 
4 | 
5 | Perturbations have a format `WxHxC`, where the channels are arranged in RGB format.
6 | ##Reference
7 | [1] S. Moosavi-Dezfooli\*, A. Fawzi\*, O. Fawzi, P. Frossard:
8 | [*Universal adversarial perturbations*](http://arxiv.org/pdf/1610.08401), CVPR 2017.
9 | 


--------------------------------------------------------------------------------
/precomputed/ResNet-152.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/precomputed/ResNet-152.mat


--------------------------------------------------------------------------------
/precomputed/VGG-16.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/precomputed/VGG-16.mat


--------------------------------------------------------------------------------
/precomputed/VGG-19.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/precomputed/VGG-19.mat


--------------------------------------------------------------------------------
/precomputed/VGG-F.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/precomputed/VGG-F.mat


--------------------------------------------------------------------------------
/python/README.md:
--------------------------------------------------------------------------------
 1 | # Python
 2 | 
 3 | Python code to find a universal perturbation [[1]](http://arxiv.org/pdf/1610.08401), using the [TensorFlow](https://www.tensorflow.org/) library.
 4 | 
 5 | ## Usage
 6 | 
 7 | ### Get started
 8 | 
 9 | To get started, you can run the demo code to apply a pre-computed universal perturbation for Inception on the image of your choice
10 | ```
11 | python demo_inception.py -i data/test_img.png	
12 | ```
13 | This will download the pre-trained model, and show the image without and with universal perturbation with the estimated labels.
14 | In this example, the pre-computed universal perturbation in `data/universal.npy` is used.
15 | 
16 | ### Computing a universal perturbation for your model
17 | 
18 | To compute a universal perturbation for your model, please follow the same struture as in `demo_inception.py`.
19 | In particular, you should use the `universal_perturbation` function (see `universal_pert.py` for details), with the set of training images 
20 | used to compute the perturbation, as well as the feedforward and gradient functions.
21 | 
22 | ## Reference
23 | [1] S. Moosavi-Dezfooli\*, A. Fawzi\*, O. Fawzi, P. Frossard:
24 | [*Universal adversarial perturbations*](http://arxiv.org/pdf/1610.08401), CVPR 2017.
25 | 


--------------------------------------------------------------------------------
/python/data/labels.txt:
--------------------------------------------------------------------------------
   1 | kit fox, Vulpes macrotis
   2 | English setter
   3 | Siberian husky
   4 | Australian terrier
   5 | English springer, English springer spaniel
   6 | grey whale, gray whale, devilfish, Eschrichtius gibbosus, Eschrichtius robustus
   7 | lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens
   8 | Egyptian cat
   9 | ibex, Capra ibex
  10 | Persian cat
  11 | cougar, puma, catamount, mountain lion, painter, panther, Felis concolor
  12 | gazelle
  13 | porcupine, hedgehog
  14 | sea lion
  15 | malamute, malemute, Alaskan malamute
  16 | badger
  17 | Great Dane
  18 | Walker hound, Walker foxhound
  19 | Welsh springer spaniel
  20 | whippet
  21 | Scottish deerhound, deerhound
  22 | killer whale, killer, orca, grampus, sea wolf, Orcinus orca
  23 | mink
  24 | African elephant, Loxodonta africana
  25 | Weimaraner
  26 | soft-coated wheaten terrier
  27 | Dandie Dinmont, Dandie Dinmont terrier
  28 | red wolf, maned wolf, Canis rufus, Canis niger
  29 | Old English sheepdog, bobtail
  30 | jaguar, panther, Panthera onca, Felis onca
  31 | otterhound, otter hound
  32 | bloodhound, sleuthhound
  33 | Airedale, Airedale terrier
  34 | hyena, hyaena
  35 | meerkat, mierkat
  36 | giant schnauzer
  37 | titi, titi monkey
  38 | three-toed sloth, ai, Bradypus tridactylus
  39 | sorrel
  40 | black-footed ferret, ferret, Mustela nigripes
  41 | dalmatian, coach dog, carriage dog
  42 | black-and-tan coonhound
  43 | papillon
  44 | skunk, polecat, wood pussy
  45 | Staffordshire bullterrier, Staffordshire bull terrier
  46 | Mexican hairless
  47 | Bouvier des Flandres, Bouviers des Flandres
  48 | weasel
  49 | miniature poodle
  50 | Cardigan, Cardigan Welsh corgi
  51 | malinois
  52 | bighorn, bighorn sheep, cimarron, Rocky Mountain bighorn, Rocky Mountain sheep, Ovis canadensis
  53 | fox squirrel, eastern fox squirrel, Sciurus niger
  54 | colobus, colobus monkey
  55 | tiger cat
  56 | Lhasa, Lhasa apso
  57 | impala, Aepyceros melampus
  58 | coyote, prairie wolf, brush wolf, Canis latrans
  59 | Yorkshire terrier
  60 | Newfoundland, Newfoundland dog
  61 | brown bear, bruin, Ursus arctos
  62 | red fox, Vulpes vulpes
  63 | Norwegian elkhound, elkhound
  64 | Rottweiler
  65 | hartebeest
  66 | Saluki, gazelle hound
  67 | grey fox, gray fox, Urocyon cinereoargenteus
  68 | schipperke
  69 | Pekinese, Pekingese, Peke
  70 | Brabancon griffon
  71 | West Highland white terrier
  72 | Sealyham terrier, Sealyham
  73 | guenon, guenon monkey
  74 | mongoose
  75 | indri, indris, Indri indri, Indri brevicaudatus
  76 | tiger, Panthera tigris
  77 | Irish wolfhound
  78 | wild boar, boar, Sus scrofa
  79 | EntleBucher
  80 | zebra
  81 | ram, tup
  82 | French bulldog
  83 | orangutan, orang, orangutang, Pongo pygmaeus
  84 | basenji
  85 | leopard, Panthera pardus
  86 | Bernese mountain dog
  87 | Maltese dog, Maltese terrier, Maltese
  88 | Norfolk terrier
  89 | toy terrier
  90 | vizsla, Hungarian pointer
  91 | cairn, cairn terrier
  92 | squirrel monkey, Saimiri sciureus
  93 | groenendael
  94 | clumber, clumber spaniel
  95 | Siamese cat, Siamese
  96 | chimpanzee, chimp, Pan troglodytes
  97 | komondor
  98 | Afghan hound, Afghan
  99 | Japanese spaniel
 100 | proboscis monkey, Nasalis larvatus
 101 | guinea pig, Cavia cobaya
 102 | white wolf, Arctic wolf, Canis lupus tundrarum
 103 | ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus
 104 | gorilla, Gorilla gorilla
 105 | borzoi, Russian wolfhound
 106 | toy poodle
 107 | Kerry blue terrier
 108 | ox
 109 | Scotch terrier, Scottish terrier, Scottie
 110 | Tibetan mastiff
 111 | spider monkey, Ateles geoffroyi
 112 | Doberman, Doberman pinscher
 113 | Boston bull, Boston terrier
 114 | Greater Swiss Mountain dog
 115 | Appenzeller
 116 | Shih-Tzu
 117 | Irish water spaniel
 118 | Pomeranian
 119 | Bedlington terrier
 120 | warthog
 121 | Arabian camel, dromedary, Camelus dromedarius
 122 | siamang, Hylobates syndactylus, Symphalangus syndactylus
 123 | miniature schnauzer
 124 | collie
 125 | golden retriever
 126 | Irish terrier
 127 | affenpinscher, monkey pinscher, monkey dog
 128 | Border collie
 129 | hare
 130 | boxer
 131 | silky terrier, Sydney silky
 132 | beagle
 133 | Leonberg
 134 | German short-haired pointer
 135 | patas, hussar monkey, Erythrocebus patas
 136 | dhole, Cuon alpinus
 137 | baboon
 138 | macaque
 139 | Chesapeake Bay retriever
 140 | bull mastiff
 141 | kuvasz
 142 | capuchin, ringtail, Cebus capucinus
 143 | pug, pug-dog
 144 | curly-coated retriever
 145 | Norwich terrier
 146 | flat-coated retriever
 147 | hog, pig, grunter, squealer, Sus scrofa
 148 | keeshond
 149 | Eskimo dog, husky
 150 | Brittany spaniel
 151 | standard poodle
 152 | Lakeland terrier
 153 | snow leopard, ounce, Panthera uncia
 154 | Gordon setter
 155 | dingo, warrigal, warragal, Canis dingo
 156 | standard schnauzer
 157 | hamster
 158 | Tibetan terrier, chrysanthemum dog
 159 | Arctic fox, white fox, Alopex lagopus
 160 | wire-haired fox terrier
 161 | basset, basset hound
 162 | water buffalo, water ox, Asiatic buffalo, Bubalus bubalis
 163 | American black bear, black bear, Ursus americanus, Euarctos americanus
 164 | Angora, Angora rabbit
 165 | bison
 166 | howler monkey, howler
 167 | hippopotamus, hippo, river horse, Hippopotamus amphibius
 168 | chow, chow chow
 169 | giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca
 170 | American Staffordshire terrier, Staffordshire terrier, American pit bull terrier, pit bull terrier
 171 | Shetland sheepdog, Shetland sheep dog, Shetland
 172 | Great Pyrenees
 173 | Chihuahua
 174 | tabby, tabby cat
 175 | marmoset
 176 | Labrador retriever
 177 | Saint Bernard, St Bernard
 178 | armadillo
 179 | Samoyed, Samoyede
 180 | bluetick
 181 | redbone
 182 | polecat, fitch, foulmart, foumart, Mustela putorius
 183 | marmot
 184 | kelpie
 185 | gibbon, Hylobates lar
 186 | llama
 187 | miniature pinscher
 188 | wood rabbit, cottontail, cottontail rabbit
 189 | Italian greyhound
 190 | lion, king of beasts, Panthera leo
 191 | cocker spaniel, English cocker spaniel, cocker
 192 | Irish setter, red setter
 193 | dugong, Dugong dugon
 194 | Indian elephant, Elephas maximus
 195 | beaver
 196 | Sussex spaniel
 197 | Pembroke, Pembroke Welsh corgi
 198 | Blenheim spaniel
 199 | Madagascar cat, ring-tailed lemur, Lemur catta
 200 | Rhodesian ridgeback
 201 | lynx, catamount
 202 | African hunting dog, hyena dog, Cape hunting dog, Lycaon pictus
 203 | langur
 204 | Ibizan hound, Ibizan Podenco
 205 | timber wolf, grey wolf, gray wolf, Canis lupus
 206 | cheetah, chetah, Acinonyx jubatus
 207 | English foxhound
 208 | briard
 209 | sloth bear, Melursus ursinus, Ursus ursinus
 210 | Border terrier
 211 | German shepherd, German shepherd dog, German police dog, alsatian
 212 | otter
 213 | koala, koala bear, kangaroo bear, native bear, Phascolarctos cinereus
 214 | tusker
 215 | echidna, spiny anteater, anteater
 216 | wallaby, brush kangaroo
 217 | platypus, duckbill, duckbilled platypus, duck-billed platypus, Ornithorhynchus anatinus
 218 | wombat
 219 | revolver, six-gun, six-shooter
 220 | umbrella
 221 | schooner
 222 | soccer ball
 223 | accordion, piano accordion, squeeze box
 224 | ant, emmet, pismire
 225 | starfish, sea star
 226 | chambered nautilus, pearly nautilus, nautilus
 227 | grand piano, grand
 228 | laptop, laptop computer
 229 | strawberry
 230 | airliner
 231 | warplane, military plane
 232 | airship, dirigible
 233 | balloon
 234 | space shuttle
 235 | fireboat
 236 | gondola
 237 | speedboat
 238 | lifeboat
 239 | canoe
 240 | yawl
 241 | catamaran
 242 | trimaran
 243 | container ship, containership, container vessel
 244 | liner, ocean liner
 245 | pirate, pirate ship
 246 | aircraft carrier, carrier, flattop, attack aircraft carrier
 247 | submarine, pigboat, sub, U-boat
 248 | wreck
 249 | half track
 250 | tank, army tank, armored combat vehicle, armoured combat vehicle
 251 | missile
 252 | bobsled, bobsleigh, bob
 253 | dogsled, dog sled, dog sleigh
 254 | bicycle-built-for-two, tandem bicycle, tandem
 255 | mountain bike, all-terrain bike, off-roader
 256 | freight car
 257 | passenger car, coach, carriage
 258 | barrow, garden cart, lawn cart, wheelbarrow
 259 | shopping cart
 260 | motor scooter, scooter
 261 | forklift
 262 | electric locomotive
 263 | steam locomotive
 264 | amphibian, amphibious vehicle
 265 | ambulance
 266 | beach wagon, station wagon, wagon, estate car, beach waggon, station waggon, waggon
 267 | cab, hack, taxi, taxicab
 268 | convertible
 269 | jeep, landrover
 270 | limousine, limo
 271 | minivan
 272 | Model T
 273 | racer, race car, racing car
 274 | sports car, sport car
 275 | go-kart
 276 | golfcart, golf cart
 277 | moped
 278 | snowplow, snowplough
 279 | fire engine, fire truck
 280 | garbage truck, dustcart
 281 | pickup, pickup truck
 282 | tow truck, tow car, wrecker
 283 | trailer truck, tractor trailer, trucking rig, rig, articulated lorry, semi
 284 | moving van
 285 | police van, police wagon, paddy wagon, patrol wagon, wagon, black Maria
 286 | recreational vehicle, RV, R.V.
 287 | streetcar, tram, tramcar, trolley, trolley car
 288 | snowmobile
 289 | tractor
 290 | mobile home, manufactured home
 291 | tricycle, trike, velocipede
 292 | unicycle, monocycle
 293 | horse cart, horse-cart
 294 | jinrikisha, ricksha, rickshaw
 295 | oxcart
 296 | bassinet
 297 | cradle
 298 | crib, cot
 299 | four-poster
 300 | bookcase
 301 | china cabinet, china closet
 302 | medicine chest, medicine cabinet
 303 | chiffonier, commode
 304 | table lamp
 305 | file, file cabinet, filing cabinet
 306 | park bench
 307 | barber chair
 308 | throne
 309 | folding chair
 310 | rocking chair, rocker
 311 | studio couch, day bed
 312 | toilet seat
 313 | desk
 314 | pool table, billiard table, snooker table
 315 | dining table, board
 316 | entertainment center
 317 | wardrobe, closet, press
 318 | Granny Smith
 319 | orange
 320 | lemon
 321 | fig
 322 | pineapple, ananas
 323 | banana
 324 | jackfruit, jak, jack
 325 | custard apple
 326 | pomegranate
 327 | acorn
 328 | hip, rose hip, rosehip
 329 | ear, spike, capitulum
 330 | rapeseed
 331 | corn
 332 | buckeye, horse chestnut, conker
 333 | organ, pipe organ
 334 | upright, upright piano
 335 | chime, bell, gong
 336 | drum, membranophone, tympan
 337 | gong, tam-tam
 338 | maraca
 339 | marimba, xylophone
 340 | steel drum
 341 | banjo
 342 | cello, violoncello
 343 | violin, fiddle
 344 | harp
 345 | acoustic guitar
 346 | electric guitar
 347 | cornet, horn, trumpet, trump
 348 | French horn, horn
 349 | trombone
 350 | harmonica, mouth organ, harp, mouth harp
 351 | ocarina, sweet potato
 352 | panpipe, pandean pipe, syrinx
 353 | bassoon
 354 | oboe, hautboy, hautbois
 355 | sax, saxophone
 356 | flute, transverse flute
 357 | daisy
 358 | yellow lady's slipper, yellow lady-slipper, Cypripedium calceolus, Cypripedium parviflorum
 359 | cliff, drop, drop-off
 360 | valley, vale
 361 | alp
 362 | volcano
 363 | promontory, headland, head, foreland
 364 | sandbar, sand bar
 365 | coral reef
 366 | lakeside, lakeshore
 367 | seashore, coast, seacoast, sea-coast
 368 | geyser
 369 | hatchet
 370 | cleaver, meat cleaver, chopper
 371 | letter opener, paper knife, paperknife
 372 | plane, carpenter's plane, woodworking plane
 373 | power drill
 374 | lawn mower, mower
 375 | hammer
 376 | corkscrew, bottle screw
 377 | can opener, tin opener
 378 | plunger, plumber's helper
 379 | screwdriver
 380 | shovel
 381 | plow, plough
 382 | chain saw, chainsaw
 383 | cock
 384 | hen
 385 | ostrich, Struthio camelus
 386 | brambling, Fringilla montifringilla
 387 | goldfinch, Carduelis carduelis
 388 | house finch, linnet, Carpodacus mexicanus
 389 | junco, snowbird
 390 | indigo bunting, indigo finch, indigo bird, Passerina cyanea
 391 | robin, American robin, Turdus migratorius
 392 | bulbul
 393 | jay
 394 | magpie
 395 | chickadee
 396 | water ouzel, dipper
 397 | kite
 398 | bald eagle, American eagle, Haliaeetus leucocephalus
 399 | vulture
 400 | great grey owl, great gray owl, Strix nebulosa
 401 | black grouse
 402 | ptarmigan
 403 | ruffed grouse, partridge, Bonasa umbellus
 404 | prairie chicken, prairie grouse, prairie fowl
 405 | peacock
 406 | quail
 407 | partridge
 408 | African grey, African gray, Psittacus erithacus
 409 | macaw
 410 | sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita
 411 | lorikeet
 412 | coucal
 413 | bee eater
 414 | hornbill
 415 | hummingbird
 416 | jacamar
 417 | toucan
 418 | drake
 419 | red-breasted merganser, Mergus serrator
 420 | goose
 421 | black swan, Cygnus atratus
 422 | white stork, Ciconia ciconia
 423 | black stork, Ciconia nigra
 424 | spoonbill
 425 | flamingo
 426 | American egret, great white heron, Egretta albus
 427 | little blue heron, Egretta caerulea
 428 | bittern
 429 | crane
 430 | limpkin, Aramus pictus
 431 | American coot, marsh hen, mud hen, water hen, Fulica americana
 432 | bustard
 433 | ruddy turnstone, Arenaria interpres
 434 | red-backed sandpiper, dunlin, Erolia alpina
 435 | redshank, Tringa totanus
 436 | dowitcher
 437 | oystercatcher, oyster catcher
 438 | European gallinule, Porphyrio porphyrio
 439 | pelican
 440 | king penguin, Aptenodytes patagonica
 441 | albatross, mollymawk
 442 | great white shark, white shark, man-eater, man-eating shark, Carcharodon carcharias
 443 | tiger shark, Galeocerdo cuvieri
 444 | hammerhead, hammerhead shark
 445 | electric ray, crampfish, numbfish, torpedo
 446 | stingray
 447 | barracouta, snoek
 448 | coho, cohoe, coho salmon, blue jack, silver salmon, Oncorhynchus kisutch
 449 | tench, Tinca tinca
 450 | goldfish, Carassius auratus
 451 | eel
 452 | rock beauty, Holocanthus tricolor
 453 | anemone fish
 454 | lionfish
 455 | puffer, pufferfish, blowfish, globefish
 456 | sturgeon
 457 | gar, garfish, garpike, billfish, Lepisosteus osseus
 458 | loggerhead, loggerhead turtle, Caretta caretta
 459 | leatherback turtle, leatherback, leathery turtle, Dermochelys coriacea
 460 | mud turtle
 461 | terrapin
 462 | box turtle, box tortoise
 463 | banded gecko
 464 | common iguana, iguana, Iguana iguana
 465 | American chameleon, anole, Anolis carolinensis
 466 | whiptail, whiptail lizard
 467 | agama
 468 | frilled lizard, Chlamydosaurus kingi
 469 | alligator lizard
 470 | Gila monster, Heloderma suspectum
 471 | green lizard, Lacerta viridis
 472 | African chameleon, Chamaeleo chamaeleon
 473 | Komodo dragon, Komodo lizard, dragon lizard, giant lizard, Varanus komodoensis
 474 | triceratops
 475 | African crocodile, Nile crocodile, Crocodylus niloticus
 476 | American alligator, Alligator mississipiensis
 477 | thunder snake, worm snake, Carphophis amoenus
 478 | ringneck snake, ring-necked snake, ring snake
 479 | hognose snake, puff adder, sand viper
 480 | green snake, grass snake
 481 | king snake, kingsnake
 482 | garter snake, grass snake
 483 | water snake
 484 | vine snake
 485 | night snake, Hypsiglena torquata
 486 | boa constrictor, Constrictor constrictor
 487 | rock python, rock snake, Python sebae
 488 | Indian cobra, Naja naja
 489 | green mamba
 490 | sea snake
 491 | horned viper, cerastes, sand viper, horned asp, Cerastes cornutus
 492 | diamondback, diamondback rattlesnake, Crotalus adamanteus
 493 | sidewinder, horned rattlesnake, Crotalus cerastes
 494 | European fire salamander, Salamandra salamandra
 495 | common newt, Triturus vulgaris
 496 | eft
 497 | spotted salamander, Ambystoma maculatum
 498 | axolotl, mud puppy, Ambystoma mexicanum
 499 | bullfrog, Rana catesbeiana
 500 | tree frog, tree-frog
 501 | tailed frog, bell toad, ribbed toad, tailed toad, Ascaphus trui
 502 | whistle
 503 | wing
 504 | paintbrush
 505 | hand blower, blow dryer, blow drier, hair dryer, hair drier
 506 | oxygen mask
 507 | snorkel
 508 | loudspeaker, speaker, speaker unit, loudspeaker system, speaker system
 509 | microphone, mike
 510 | screen, CRT screen
 511 | mouse, computer mouse
 512 | electric fan, blower
 513 | oil filter
 514 | strainer
 515 | space heater
 516 | stove
 517 | guillotine
 518 | barometer
 519 | rule, ruler
 520 | odometer, hodometer, mileometer, milometer
 521 | scale, weighing machine
 522 | analog clock
 523 | digital clock
 524 | wall clock
 525 | hourglass
 526 | sundial
 527 | parking meter
 528 | stopwatch, stop watch
 529 | digital watch
 530 | stethoscope
 531 | syringe
 532 | magnetic compass
 533 | binoculars, field glasses, opera glasses
 534 | projector
 535 | sunglasses, dark glasses, shades
 536 | loupe, jeweler's loupe
 537 | radio telescope, radio reflector
 538 | bow
 539 | cannon
 540 | assault rifle, assault gun
 541 | rifle
 542 | projectile, missile
 543 | computer keyboard, keypad
 544 | typewriter keyboard
 545 | crane
 546 | lighter, light, igniter, ignitor
 547 | abacus
 548 | cash machine, cash dispenser, automated teller machine, automatic teller machine, automated teller, automatic teller, ATM
 549 | slide rule, slipstick
 550 | desktop computer
 551 | hand-held computer, hand-held microcomputer
 552 | notebook, notebook computer
 553 | web site, website, internet site, site
 554 | harvester, reaper
 555 | thresher, thrasher, threshing machine
 556 | printer
 557 | slot, one-armed bandit
 558 | vending machine
 559 | sewing machine
 560 | joystick
 561 | switch, electric switch, electrical switch
 562 | hook, claw
 563 | car wheel
 564 | paddlewheel, paddle wheel
 565 | pinwheel
 566 | potter's wheel
 567 | gas pump, gasoline pump, petrol pump, island dispenser
 568 | carousel, carrousel, merry-go-round, roundabout, whirligig
 569 | swing
 570 | reel
 571 | radiator
 572 | puck, hockey puck
 573 | hard disc, hard disk, fixed disk
 574 | sunglass
 575 | pick, plectrum, plectron
 576 | car mirror
 577 | solar dish, solar collector, solar furnace
 578 | remote control, remote
 579 | disk brake, disc brake
 580 | buckle
 581 | hair slide
 582 | knot
 583 | combination lock
 584 | padlock
 585 | nail
 586 | safety pin
 587 | screw
 588 | muzzle
 589 | seat belt, seatbelt
 590 | ski
 591 | candle, taper, wax light
 592 | jack-o'-lantern
 593 | spotlight, spot
 594 | torch
 595 | neck brace
 596 | pier
 597 | tripod
 598 | maypole
 599 | mousetrap
 600 | spider web, spider's web
 601 | trilobite
 602 | harvestman, daddy longlegs, Phalangium opilio
 603 | scorpion
 604 | black and gold garden spider, Argiope aurantia
 605 | barn spider, Araneus cavaticus
 606 | garden spider, Aranea diademata
 607 | black widow, Latrodectus mactans
 608 | tarantula
 609 | wolf spider, hunting spider
 610 | tick
 611 | centipede
 612 | isopod
 613 | Dungeness crab, Cancer magister
 614 | rock crab, Cancer irroratus
 615 | fiddler crab
 616 | king crab, Alaska crab, Alaskan king crab, Alaska king crab, Paralithodes camtschatica
 617 | American lobster, Northern lobster, Maine lobster, Homarus americanus
 618 | spiny lobster, langouste, rock lobster, crawfish, crayfish, sea crawfish
 619 | crayfish, crawfish, crawdad, crawdaddy
 620 | hermit crab
 621 | tiger beetle
 622 | ladybug, ladybeetle, lady beetle, ladybird, ladybird beetle
 623 | ground beetle, carabid beetle
 624 | long-horned beetle, longicorn, longicorn beetle
 625 | leaf beetle, chrysomelid
 626 | dung beetle
 627 | rhinoceros beetle
 628 | weevil
 629 | fly
 630 | bee
 631 | grasshopper, hopper
 632 | cricket
 633 | walking stick, walkingstick, stick insect
 634 | cockroach, roach
 635 | mantis, mantid
 636 | cicada, cicala
 637 | leafhopper
 638 | lacewing, lacewing fly
 639 | dragonfly, darning needle, devil's darning needle, sewing needle, snake feeder, snake doctor, mosquito hawk, skeeter hawk
 640 | damselfly
 641 | admiral
 642 | ringlet, ringlet butterfly
 643 | monarch, monarch butterfly, milkweed butterfly, Danaus plexippus
 644 | cabbage butterfly
 645 | sulphur butterfly, sulfur butterfly
 646 | lycaenid, lycaenid butterfly
 647 | jellyfish
 648 | sea anemone, anemone
 649 | brain coral
 650 | flatworm, platyhelminth
 651 | nematode, nematode worm, roundworm
 652 | conch
 653 | snail
 654 | slug
 655 | sea slug, nudibranch
 656 | chiton, coat-of-mail shell, sea cradle, polyplacophore
 657 | sea urchin
 658 | sea cucumber, holothurian
 659 | iron, smoothing iron
 660 | espresso maker
 661 | microwave, microwave oven
 662 | Dutch oven
 663 | rotisserie
 664 | toaster
 665 | waffle iron
 666 | vacuum, vacuum cleaner
 667 | dishwasher, dish washer, dishwashing machine
 668 | refrigerator, icebox
 669 | washer, automatic washer, washing machine
 670 | Crock Pot
 671 | frying pan, frypan, skillet
 672 | wok
 673 | caldron, cauldron
 674 | coffeepot
 675 | teapot
 676 | spatula
 677 | altar
 678 | triumphal arch
 679 | patio, terrace
 680 | steel arch bridge
 681 | suspension bridge
 682 | viaduct
 683 | barn
 684 | greenhouse, nursery, glasshouse
 685 | palace
 686 | monastery
 687 | library
 688 | apiary, bee house
 689 | boathouse
 690 | church, church building
 691 | mosque
 692 | stupa, tope
 693 | planetarium
 694 | restaurant, eating house, eating place, eatery
 695 | cinema, movie theater, movie theatre, movie house, picture palace
 696 | home theater, home theatre
 697 | lumbermill, sawmill
 698 | coil, spiral, volute, whorl, helix
 699 | obelisk
 700 | totem pole
 701 | castle
 702 | prison, prison house
 703 | grocery store, grocery, food market, market
 704 | bakery, bakeshop, bakehouse
 705 | barbershop
 706 | bookshop, bookstore, bookstall
 707 | butcher shop, meat market
 708 | confectionery, confectionary, candy store
 709 | shoe shop, shoe-shop, shoe store
 710 | tobacco shop, tobacconist shop, tobacconist
 711 | toyshop
 712 | fountain
 713 | cliff dwelling
 714 | yurt
 715 | dock, dockage, docking facility
 716 | brass, memorial tablet, plaque
 717 | megalith, megalithic structure
 718 | bannister, banister, balustrade, balusters, handrail
 719 | breakwater, groin, groyne, mole, bulwark, seawall, jetty
 720 | dam, dike, dyke
 721 | chainlink fence
 722 | picket fence, paling
 723 | worm fence, snake fence, snake-rail fence, Virginia fence
 724 | stone wall
 725 | grille, radiator grille
 726 | sliding door
 727 | turnstile
 728 | mountain tent
 729 | scoreboard
 730 | honeycomb
 731 | plate rack
 732 | pedestal, plinth, footstall
 733 | beacon, lighthouse, beacon light, pharos
 734 | mashed potato
 735 | bell pepper
 736 | head cabbage
 737 | broccoli
 738 | cauliflower
 739 | zucchini, courgette
 740 | spaghetti squash
 741 | acorn squash
 742 | butternut squash
 743 | cucumber, cuke
 744 | artichoke, globe artichoke
 745 | cardoon
 746 | mushroom
 747 | shower curtain
 748 | jean, blue jean, denim
 749 | carton
 750 | handkerchief, hankie, hanky, hankey
 751 | sandal
 752 | ashcan, trash can, garbage can, wastebin, ash bin, ash-bin, ashbin, dustbin, trash barrel, trash bin
 753 | safe
 754 | plate
 755 | necklace
 756 | croquet ball
 757 | fur coat
 758 | thimble
 759 | pajama, pyjama, pj's, jammies
 760 | running shoe
 761 | cocktail shaker
 762 | chest
 763 | manhole cover
 764 | modem
 765 | tub, vat
 766 | tray
 767 | balance beam, beam
 768 | bagel, beigel
 769 | prayer rug, prayer mat
 770 | kimono
 771 | hot pot, hotpot
 772 | whiskey jug
 773 | knee pad
 774 | book jacket, dust cover, dust jacket, dust wrapper
 775 | spindle
 776 | ski mask
 777 | beer bottle
 778 | crash helmet
 779 | bottlecap
 780 | tile roof
 781 | mask
 782 | maillot
 783 | Petri dish
 784 | football helmet
 785 | bathing cap, swimming cap
 786 | teddy, teddy bear
 787 | holster
 788 | pop bottle, soda bottle
 789 | photocopier
 790 | vestment
 791 | crossword puzzle, crossword
 792 | golf ball
 793 | trifle
 794 | suit, suit of clothes
 795 | water tower
 796 | feather boa, boa
 797 | cloak
 798 | red wine
 799 | drumstick
 800 | shield, buckler
 801 | Christmas stocking
 802 | hoopskirt, crinoline
 803 | menu
 804 | stage
 805 | bonnet, poke bonnet
 806 | meat loaf, meatloaf
 807 | baseball
 808 | face powder
 809 | scabbard
 810 | sunscreen, sunblock, sun blocker
 811 | beer glass
 812 | hen-of-the-woods, hen of the woods, Polyporus frondosus, Grifola frondosa
 813 | guacamole
 814 | lampshade, lamp shade
 815 | wool, woolen, woollen
 816 | hay
 817 | bow tie, bow-tie, bowtie
 818 | mailbag, postbag
 819 | water jug
 820 | bucket, pail
 821 | dishrag, dishcloth
 822 | soup bowl
 823 | eggnog
 824 | mortar
 825 | trench coat
 826 | paddle, boat paddle
 827 | chain
 828 | swab, swob, mop
 829 | mixing bowl
 830 | potpie
 831 | wine bottle
 832 | shoji
 833 | bulletproof vest
 834 | drilling platform, offshore rig
 835 | binder, ring-binder
 836 | cardigan
 837 | sweatshirt
 838 | pot, flowerpot
 839 | birdhouse
 840 | hamper
 841 | ping-pong ball
 842 | pencil box, pencil case
 843 | pay-phone, pay-station
 844 | consomme
 845 | apron
 846 | punching bag, punch bag, punching ball, punchball
 847 | backpack, back pack, knapsack, packsack, rucksack, haversack
 848 | groom, bridegroom
 849 | bearskin, busby, shako
 850 | pencil sharpener
 851 | broom
 852 | mosquito net
 853 | abaya
 854 | mortarboard
 855 | poncho
 856 | crutch
 857 | Polaroid camera, Polaroid Land camera
 858 | space bar
 859 | cup
 860 | racket, racquet
 861 | traffic light, traffic signal, stoplight
 862 | quill, quill pen
 863 | radio, wireless
 864 | dough
 865 | cuirass
 866 | military uniform
 867 | lipstick, lip rouge
 868 | shower cap
 869 | monitor
 870 | oscilloscope, scope, cathode-ray oscilloscope, CRO
 871 | mitten
 872 | brassiere, bra, bandeau
 873 | French loaf
 874 | vase
 875 | milk can
 876 | rugby ball
 877 | paper towel
 878 | earthstar
 879 | envelope
 880 | miniskirt, mini
 881 | cowboy hat, ten-gallon hat
 882 | trolleybus, trolley coach, trackless trolley
 883 | perfume, essence
 884 | bathtub, bathing tub, bath, tub
 885 | hotdog, hot dog, red hot
 886 | coral fungus
 887 | bullet train, bullet
 888 | pillow
 889 | toilet tissue, toilet paper, bathroom tissue
 890 | cassette
 891 | carpenter's kit, tool kit
 892 | ladle
 893 | stinkhorn, carrion fungus
 894 | lotion
 895 | hair spray
 896 | academic gown, academic robe, judge's robe
 897 | dome
 898 | crate
 899 | wig
 900 | burrito
 901 | pill bottle
 902 | chain mail, ring mail, mail, chain armor, chain armour, ring armor, ring armour
 903 | theater curtain, theatre curtain
 904 | window shade
 905 | barrel, cask
 906 | washbasin, handbasin, washbowl, lavabo, wash-hand basin
 907 | ballpoint, ballpoint pen, ballpen, Biro
 908 | basketball
 909 | bath towel
 910 | cowboy boot
 911 | gown
 912 | window screen
 913 | agaric
 914 | cellular telephone, cellular phone, cellphone, cell, mobile phone
 915 | nipple
 916 | barbell
 917 | mailbox, letter box
 918 | lab coat, laboratory coat
 919 | fire screen, fireguard
 920 | minibus
 921 | packet
 922 | maze, labyrinth
 923 | pole
 924 | horizontal bar, high bar
 925 | sombrero
 926 | pickelhaube
 927 | rain barrel
 928 | wallet, billfold, notecase, pocketbook
 929 | cassette player
 930 | comic book
 931 | piggy bank, penny bank
 932 | street sign
 933 | bell cote, bell cot
 934 | fountain pen
 935 | Windsor tie
 936 | volleyball
 937 | overskirt
 938 | sarong
 939 | purse
 940 | bolo tie, bolo, bola tie, bola
 941 | bib
 942 | parachute, chute
 943 | sleeping bag
 944 | television, television system
 945 | swimming trunks, bathing trunks
 946 | measuring cup
 947 | espresso
 948 | pizza, pizza pie
 949 | breastplate, aegis, egis
 950 | shopping basket
 951 | wooden spoon
 952 | saltshaker, salt shaker
 953 | chocolate sauce, chocolate syrup
 954 | ballplayer, baseball player
 955 | goblet
 956 | gyromitra
 957 | stretcher
 958 | water bottle
 959 | dial telephone, dial phone
 960 | soap dispenser
 961 | jersey, T-shirt, tee shirt
 962 | school bus
 963 | jigsaw puzzle
 964 | plastic bag
 965 | reflex camera
 966 | diaper, nappy, napkin
 967 | Band Aid
 968 | ice lolly, lolly, lollipop, popsicle
 969 | velvet
 970 | tennis ball
 971 | gasmask, respirator, gas helmet
 972 | doormat, welcome mat
 973 | Loafer
 974 | ice cream, icecream
 975 | pretzel
 976 | quilt, comforter, comfort, puff
 977 | maillot, tank suit
 978 | tape player
 979 | clog, geta, patten, sabot
 980 | iPod
 981 | bolete
 982 | scuba diver
 983 | pitcher, ewer
 984 | matchstick
 985 | bikini, two-piece
 986 | sock
 987 | CD player
 988 | lens cap, lens cover
 989 | thatch, thatched roof
 990 | vault
 991 | beaker
 992 | bubble
 993 | cheeseburger
 994 | parallel bars, bars
 995 | flagpole, flagstaff
 996 | coffee mug
 997 | rubber eraser, rubber, pencil eraser
 998 | stole
 999 | carbonara
1000 | dumbbell
1001 | 


--------------------------------------------------------------------------------
/python/data/test_img.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/python/data/test_img.png


--------------------------------------------------------------------------------
/python/data/universal.npy:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/LTS4/universal/2e5ab1544a5c0ea53f07b6c21dbecb6053bb1ca1/python/data/universal.npy


--------------------------------------------------------------------------------
/python/deepfool.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | 
 3 | def deepfool(image, f, grads, num_classes=10, overshoot=0.02, max_iter=50):
 4 | 
 5 |     """
 6 |        :param image: Image of size HxWx3
 7 |        :param f: feedforward function (input: images, output: values of activation BEFORE softmax).
 8 |        :param grads: gradient functions with respect to input (as many gradients as classes).
 9 |        :param num_classes: num_classes (limits the number of classes to test against, by default = 10)
10 |        :param overshoot: used as a termination criterion to prevent vanishing updates (default = 0.02).
11 |        :param max_iter: maximum number of iterations for deepfool (default = 10)
12 |        :return: minimal perturbation that fools the classifier, number of iterations that it required, new estimated_label and perturbed image
13 |     """
14 | 
15 |     f_image = np.array(f(image)).flatten()
16 |     I = (np.array(f_image)).flatten().argsort()[::-1]
17 | 
18 |     I = I[0:num_classes]
19 |     label = I[0]
20 | 
21 |     input_shape = image.shape
22 |     pert_image = image
23 | 
24 |     f_i = np.array(f(pert_image)).flatten()
25 |     k_i = int(np.argmax(f_i))
26 | 
27 |     w = np.zeros(input_shape)
28 |     r_tot = np.zeros(input_shape)
29 | 
30 |     loop_i = 0
31 | 
32 |     while k_i == label and loop_i < max_iter:
33 | 
34 |         pert = np.inf
35 |         gradients = np.asarray(grads(pert_image,I))
36 | 
37 |         for k in range(1, num_classes):
38 | 
39 |             # set new w_k and new f_k
40 |             w_k = gradients[k, :, :, :, :] - gradients[0, :, :, :, :]
41 |             f_k = f_i[I[k]] - f_i[I[0]]
42 |             pert_k = abs(f_k)/np.linalg.norm(w_k.flatten())
43 | 
44 |             # determine which w_k to use
45 |             if pert_k < pert:
46 |                 pert = pert_k
47 |                 w = w_k
48 | 
49 |         # compute r_i and r_tot
50 |         r_i =  pert * w / np.linalg.norm(w)
51 |         r_tot = r_tot + r_i
52 | 
53 |         # compute new perturbed image
54 |         pert_image = image + (1+overshoot)*r_tot
55 |         loop_i += 1
56 | 
57 |         # compute new label
58 |         f_i = np.array(f(pert_image)).flatten()
59 |         k_i = int(np.argmax(f_i))
60 | 
61 |     r_tot = (1+overshoot)*r_tot
62 | 
63 |     return r_tot, loop_i, k_i, pert_image
64 | 


--------------------------------------------------------------------------------
/python/demo_inception.py:
--------------------------------------------------------------------------------
  1 | import tensorflow as tf
  2 | import numpy as np
  3 | from tensorflow.python.platform import gfile
  4 | import os.path
  5 | from prepare_imagenet_data import preprocess_image_batch, create_imagenet_npy, undo_image_avg
  6 | import matplotlib.pyplot as plt
  7 | import sys, getopt
  8 | import zipfile
  9 | from timeit import time
 10 | 
 11 | if sys.version_info[0] >= 3:
 12 |     from urllib.request import urlretrieve
 13 | else:
 14 |     from urllib import urlretrieve
 15 | 
 16 | 
 17 | from universal_pert import universal_perturbation
 18 | device = '/gpu:0'
 19 | num_classes = 10
 20 | 
 21 | def jacobian(y_flat, x, inds):
 22 |     n = num_classes # Not really necessary, just a quick fix.
 23 |     loop_vars = [
 24 |          tf.constant(0, tf.int32),
 25 |          tf.TensorArray(tf.float32, size=n),
 26 |     ]
 27 |     _, jacobian = tf.while_loop(
 28 |         lambda j,_: j < n,
 29 |         lambda j,result: (j+1, result.write(j, tf.gradients(y_flat[inds[j]], x))),
 30 |         loop_vars)
 31 |     return jacobian.stack()
 32 | 
 33 | if __name__ == '__main__':
 34 | 
 35 |     # Parse arguments
 36 |     argv = sys.argv[1:]
 37 | 
 38 |     # Default values
 39 |     path_train_imagenet = '/datasets2/ILSVRC2012/train'
 40 |     path_test_image = 'data/test_img.png'
 41 |     
 42 |     try:
 43 |         opts, args = getopt.getopt(argv,"i:t:",["test_image=","training_path="])
 44 |     except getopt.GetoptError:
 45 |         print ('python ' + sys.argv[0] + ' -i <test image> -t <imagenet training path>')
 46 |         sys.exit(2)
 47 | 
 48 |     for opt, arg in opts:
 49 |         if opt == '-t':
 50 |             path_train_imagenet = arg
 51 |         if opt == '-i':
 52 |             path_test_image = arg
 53 | 
 54 |     with tf.device(device):
 55 |         persisted_sess = tf.Session()
 56 |         inception_model_path = os.path.join('data', 'tensorflow_inception_graph.pb')
 57 | 
 58 |         if os.path.isfile(inception_model_path) == 0:
 59 |             print('Downloading Inception model...')
 60 |             urlretrieve ("https://storage.googleapis.com/download.tensorflow.org/models/inception5h.zip", os.path.join('data', 'inception5h.zip'))
 61 |             # Unzipping the file
 62 |             zip_ref = zipfile.ZipFile(os.path.join('data', 'inception5h.zip'), 'r')
 63 |             zip_ref.extract('tensorflow_inception_graph.pb', 'data')
 64 |             zip_ref.close()
 65 | 
 66 |         model = os.path.join(inception_model_path)
 67 | 
 68 |         # Load the Inception model
 69 |         with gfile.FastGFile(model, 'rb') as f:
 70 |             graph_def = tf.GraphDef()
 71 |             graph_def.ParseFromString(f.read())
 72 |             persisted_sess.graph.as_default()
 73 |             tf.import_graph_def(graph_def, name='')
 74 | 
 75 |         persisted_sess.graph.get_operations()
 76 | 
 77 |         persisted_input = persisted_sess.graph.get_tensor_by_name("input:0")
 78 |         persisted_output = persisted_sess.graph.get_tensor_by_name("softmax2_pre_activation:0")
 79 | 
 80 |         print('>> Computing feedforward function...')
 81 |         def f(image_inp): return persisted_sess.run(persisted_output, feed_dict={persisted_input: np.reshape(image_inp, (-1, 224, 224, 3))})
 82 | 
 83 |         file_perturbation = os.path.join('data', 'universal.npy')
 84 | 
 85 |         if os.path.isfile(file_perturbation) == 0:
 86 | 
 87 |             # TODO: Optimize this construction part!
 88 |             print('>> Compiling the gradient tensorflow functions. This might take some time...')
 89 |             y_flat = tf.reshape(persisted_output, (-1,))
 90 |             inds = tf.placeholder(tf.int32, shape=(num_classes,))
 91 |             dydx = jacobian(y_flat,persisted_input,inds)
 92 | 
 93 |             print('>> Computing gradient function...')
 94 |             def grad_fs(image_inp, indices): return persisted_sess.run(dydx, feed_dict={persisted_input: image_inp, inds: indices}).squeeze(axis=1)
 95 | 
 96 |             # Load/Create data
 97 |             datafile = os.path.join('data', 'imagenet_data.npy')
 98 |             if os.path.isfile(datafile) == 0:
 99 |                 print('>> Creating pre-processed imagenet data...')
100 |                 X = create_imagenet_npy(path_train_imagenet)
101 | 
102 |                 print('>> Saving the pre-processed imagenet data')
103 |                 if not os.path.exists('data'):
104 |                     os.makedirs('data')
105 | 
106 |                 # Save the pre-processed images
107 |                 # Caution: This can take take a lot of space. Comment this part to discard saving.
108 |                 np.save(os.path.join('data', 'imagenet_data.npy'), X)
109 | 
110 |             else:
111 |                 print('>> Pre-processed imagenet data detected')
112 |                 X = np.load(datafile)
113 | 
114 |             # Running universal perturbation
115 |             v = universal_perturbation(X, f, grad_fs, delta=0.2,num_classes=num_classes)
116 | 
117 |             # Saving the universal perturbation
118 |             np.save(os.path.join(file_perturbation), v)
119 | 
120 |         else:
121 |             print('>> Found a pre-computed universal perturbation! Retrieving it from ", file_perturbation')
122 |             v = np.load(file_perturbation)
123 | 
124 |         print('>> Testing the universal perturbation on an image')
125 | 
126 |         # Test the perturbation on the image
127 |         labels = open(os.path.join('data', 'labels.txt'), 'r').read().split('\n')
128 | 
129 |         image_original = preprocess_image_batch([path_test_image], img_size=(256, 256), crop_size=(224, 224), color_mode="rgb")
130 |         label_original = np.argmax(f(image_original), axis=1).flatten()
131 |         str_label_original = labels[np.int(label_original)-1].split(',')[0]
132 | 
133 |         # Clip the perturbation to make sure images fit in uint8
134 |         clipped_v = np.clip(undo_image_avg(image_original[0,:,:,:]+v[0,:,:,:]), 0, 255) - np.clip(undo_image_avg(image_original[0,:,:,:]), 0, 255)
135 | 
136 |         image_perturbed = image_original + clipped_v[None, :, :, :]
137 |         label_perturbed = np.argmax(f(image_perturbed), axis=1).flatten()
138 |         str_label_perturbed = labels[np.int(label_perturbed)-1].split(',')[0]
139 | 
140 |         # Show original and perturbed image
141 |         plt.figure()
142 |         plt.subplot(1, 2, 1)
143 |         plt.imshow(undo_image_avg(image_original[0, :, :, :]).astype(dtype='uint8'), interpolation=None)
144 |         plt.title(str_label_original)
145 | 
146 |         plt.subplot(1, 2, 2)
147 |         plt.imshow(undo_image_avg(image_perturbed[0, :, :, :]).astype(dtype='uint8'), interpolation=None)
148 |         plt.title(str_label_perturbed)
149 | 
150 |         plt.show()
151 | 


--------------------------------------------------------------------------------
/python/prepare_imagenet_data.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import os
 3 | from scipy.misc import imread, imresize
 4 | 
 5 | CLASS_INDEX = None
 6 | CLASS_INDEX_PATH = 'https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json'
 7 | 
 8 | 
 9 | def preprocess_image_batch(image_paths, img_size=None, crop_size=None, color_mode="rgb", out=None):
10 |     img_list = []
11 | 
12 |     for im_path in image_paths:
13 |         img = imread(im_path, mode='RGB')
14 |         if img_size:
15 |             img = imresize(img,img_size)
16 | 
17 |         img = img.astype('float32')
18 |         # We normalize the colors (in RGB space) with the empirical means on the training set
19 |         img[:, :, 0] -= 123.68
20 |         img[:, :, 1] -= 116.779
21 |         img[:, :, 2] -= 103.939
22 |         # We permute the colors to get them in the BGR order
23 |         # if color_mode=="bgr":
24 |         #    img[:,:,[0,1,2]] = img[:,:,[2,1,0]]
25 | 
26 |         if crop_size:
27 |             img = img[(img_size[0] - crop_size[0]) // 2:(img_size[0] + crop_size[0]) // 2, (img_size[1]-crop_size[1])//2:(img_size[1]+crop_size[1])//2, :];
28 | 
29 |         img_list.append(img)
30 | 
31 |     try:
32 |         img_batch = np.stack(img_list, axis=0)
33 |     except:
34 |         raise ValueError('when img_size and crop_size are None, images'
35 |                 ' in image_paths must have the same shapes.')
36 | 
37 |     if out is not None and hasattr(out, 'append'):
38 |         out.append(img_batch)
39 |     else:
40 |         return img_batch
41 | 
42 | def undo_image_avg(img):
43 |     img_copy = np.copy(img)
44 |     img_copy[:, :, 0] = img_copy[:, :, 0] + 123.68
45 |     img_copy[:, :, 1] = img_copy[:, :, 1] + 116.779
46 |     img_copy[:, :, 2] = img_copy[:, :, 2] + 103.939
47 |     return img_copy
48 | 
49 | def create_imagenet_npy(path_train_imagenet, len_batch=10000):
50 | 
51 |     # path_train_imagenet = '/datasets2/ILSVRC2012/train';
52 | 
53 |     sz_img = [224, 224]
54 |     num_channels = 3
55 |     num_classes = 1000
56 | 
57 |     im_array = np.zeros([len_batch] + sz_img + [num_channels], dtype=np.float32)
58 |     num_imgs_per_batch = int(len_batch / num_classes)
59 | 
60 |     dirs = [x[0] for x in os.walk(path_train_imagenet)]
61 |     dirs = dirs[1:]
62 | 
63 |     # Sort the directory in alphabetical order (same as synset_words.txt)
64 |     dirs = sorted(dirs)
65 | 
66 |     it = 0
67 |     Matrix = [0 for x in range(1000)]
68 | 
69 |     for d in dirs:
70 |         for _, _, filename in os.walk(os.path.join(path_train_imagenet, d)):
71 |             Matrix[it] = filename
72 |         it = it+1
73 | 
74 | 
75 |     it = 0
76 |     # Load images, pre-process, and save
77 |     for k in range(num_classes):
78 |         for u in range(num_imgs_per_batch):
79 |             print('Processing image number ', it)
80 |             path_img = os.path.join(dirs[k], Matrix[k][u])
81 |             image = preprocess_image_batch([path_img],img_size=(256,256), crop_size=(224,224), color_mode="rgb")
82 |             im_array[it:(it+1), :, :, :] = image
83 |             it = it + 1
84 | 
85 |     return im_array
86 | 


--------------------------------------------------------------------------------
/python/universal_pert.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | from deepfool import deepfool
 3 | 
 4 | def proj_lp(v, xi, p):
 5 | 
 6 |     # Project on the lp ball centered at 0 and of radius xi
 7 | 
 8 |     # SUPPORTS only p = 2 and p = Inf for now
 9 |     if p == 2:
10 |         v = v * min(1, xi/np.linalg.norm(v.flatten(1)))
11 |         # v = v / np.linalg.norm(v.flatten(1)) * xi
12 |     elif p == np.inf:
13 |         v = np.sign(v) * np.minimum(abs(v), xi)
14 |     else:
15 |          raise ValueError('Values of p different from 2 and Inf are currently not supported...')
16 | 
17 |     return v
18 | 
19 | def universal_perturbation(dataset, f, grads, delta=0.2, max_iter_uni = np.inf, xi=10, p=np.inf, num_classes=10, overshoot=0.02, max_iter_df=10):
20 |     """
21 |     :param dataset: Images of size MxHxWxC (M: number of images)
22 | 
23 |     :param f: feedforward function (input: images, output: values of activation BEFORE softmax).
24 | 
25 |     :param grads: gradient functions with respect to input (as many gradients as classes).
26 | 
27 |     :param delta: controls the desired fooling rate (default = 80% fooling rate)
28 | 
29 |     :param max_iter_uni: optional other termination criterion (maximum number of iteration, default = np.inf)
30 | 
31 |     :param xi: controls the l_p magnitude of the perturbation (default = 10)
32 | 
33 |     :param p: norm to be used (FOR NOW, ONLY p = 2, and p = np.inf ARE ACCEPTED!) (default = np.inf)
34 | 
35 |     :param num_classes: num_classes (limits the number of classes to test against, by default = 10)
36 | 
37 |     :param overshoot: used as a termination criterion to prevent vanishing updates (default = 0.02).
38 | 
39 |     :param max_iter_df: maximum number of iterations for deepfool (default = 10)
40 | 
41 |     :return: the universal perturbation.
42 |     """
43 | 
44 |     v = 0
45 |     fooling_rate = 0.0
46 |     num_images =  np.shape(dataset)[0] # The images should be stacked ALONG FIRST DIMENSION
47 | 
48 |     itr = 0
49 |     while fooling_rate < 1-delta and itr < max_iter_uni:
50 |         # Shuffle the dataset
51 |         np.random.shuffle(dataset)
52 | 
53 |         print ('Starting pass number ', itr)
54 | 
55 |         # Go through the data set and compute the perturbation increments sequentially
56 |         for k in range(0, num_images):
57 |             cur_img = dataset[k:(k+1), :, :, :]
58 | 
59 |             if int(np.argmax(np.array(f(cur_img)).flatten())) == int(np.argmax(np.array(f(cur_img+v)).flatten())):
60 |                 print('>> k = ', k, ', pass #', itr)
61 | 
62 |                 # Compute adversarial perturbation
63 |                 dr,iter,_,_ = deepfool(cur_img + v, f, grads, num_classes=num_classes, overshoot=overshoot, max_iter=max_iter_df)
64 | 
65 |                 # Make sure it converged...
66 |                 if iter < max_iter_df-1:
67 |                     v = v + dr
68 | 
69 |                     # Project on l_p ball
70 |                     v = proj_lp(v, xi, p)
71 | 
72 |         itr = itr + 1
73 | 
74 |         # Perturb the dataset with computed perturbation
75 |         dataset_perturbed = dataset + v
76 | 
77 |         est_labels_orig = np.zeros((num_images))
78 |         est_labels_pert = np.zeros((num_images))
79 | 
80 |         batch_size = 100
81 |         num_batches = np.int(np.ceil(np.float(num_images) / np.float(batch_size)))
82 | 
83 |         # Compute the estimated labels in batches
84 |         for ii in range(0, num_batches):
85 |             m = (ii * batch_size)
86 |             M = min((ii+1)*batch_size, num_images)
87 |             est_labels_orig[m:M] = np.argmax(f(dataset[m:M, :, :, :]), axis=1).flatten()
88 |             est_labels_pert[m:M] = np.argmax(f(dataset_perturbed[m:M, :, :, :]), axis=1).flatten()
89 | 
90 |         # Compute the fooling rate
91 |         fooling_rate = float(np.sum(est_labels_pert != est_labels_orig) / float(num_images))
92 |         print('FOOLING RATE = ', fooling_rate)
93 | 
94 |     return v
95 | 


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