├── LICENSE
├── MobileNet-v2-deploy.prototxt
├── README.md
├── mobilenetv2-0.35
    ├── test.prototxt
    └── train.prototxt
├── mobilenetv2-0.5
    ├── test.prototxt
    └── train.prototxt
├── mobilenetv2-0.75
    ├── test.prototxt
    └── train.prototxt
└── mobilenetv2-1.0
    ├── test.prototxt
    └── train.prototxt


/LICENSE:
--------------------------------------------------------------------------------
 1 | MIT License
 2 | 
 3 | Copyright (c) 2018 Yubin Wang
 4 | 
 5 | Permission is hereby granted, free of charge, to any person obtaining a copy
 6 | of this software and associated documentation files (the "Software"), to deal
 7 | in the Software without restriction, including without limitation the rights
 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 9 | copies of the Software, and to permit persons to whom the Software is
10 | furnished to do so, subject to the following conditions:
11 | 
12 | The above copyright notice and this permission notice shall be included in all
13 | copies or substantial portions of the Software.
14 | 
15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 | SOFTWARE.
22 | 


--------------------------------------------------------------------------------
/MobileNet-v2-deploy.prototxt:
--------------------------------------------------------------------------------
   1 | name: "MobileNet-v2"
   2 | input: "data"
   3 | input_dim: 1
   4 | input_dim: 3
   5 | input_dim: 224
   6 | input_dim: 224
   7 | layer {
   8 |   name: "conv_1_1"
   9 |   type: "Convolution"
  10 |   bottom: "data"
  11 |   top: "conv_1_1"
  12 |   param {
  13 |     lr_mult: 1.0
  14 |     decay_mult: 1.0
  15 |   }
  16 |   convolution_param {
  17 |     num_output: 32
  18 |     bias_term: false
  19 |     pad: 1
  20 |     kernel_size: 3
  21 |     stride: 2
  22 |     weight_filler {
  23 |       type: "msra"
  24 |     }
  25 |     bias_filler {
  26 |       type: "constant"
  27 |       value: 0.0
  28 |     }
  29 |   }
  30 | }
  31 | layer {
  32 |   name: "bn_1_1"
  33 |   type: "BatchNorm"
  34 |   bottom: "conv_1_1"
  35 |   top: "conv_1_1"
  36 |   param {
  37 |     lr_mult: 0.0
  38 |     decay_mult: 0.0
  39 |   }
  40 |   param {
  41 |     lr_mult: 0.0
  42 |     decay_mult: 0.0
  43 |   }
  44 |   param {
  45 |     lr_mult: 0.0
  46 |     decay_mult: 0.0
  47 |   }
  48 |   batch_norm_param {
  49 |     use_global_stats: true
  50 |   }
  51 | }
  52 | layer {
  53 |   name: "scale_1_1"
  54 |   type: "Scale"
  55 |   bottom: "conv_1_1"
  56 |   top: "conv_1_1"
  57 |   scale_param {
  58 |     bias_term: true
  59 |   }
  60 | }
  61 | layer {
  62 |   name: "relu_1_1"
  63 |   type: "ReLU"
  64 |   bottom: "conv_1_1"
  65 |   top: "conv_1_1"
  66 | }
  67 | layer {
  68 |   name: "conv_2_1_pw"
  69 |   type: "Convolution"
  70 |   bottom: "conv_1_1"
  71 |   top: "conv_2_1_pw"
  72 |   param {
  73 |     lr_mult: 1.0
  74 |     decay_mult: 1.0
  75 |   }
  76 |   convolution_param {
  77 |     num_output: 32
  78 |     bias_term: false
  79 |     pad: 0
  80 |     kernel_size: 1
  81 |     stride: 1
  82 |     weight_filler {
  83 |       type: "msra"
  84 |     }
  85 |     bias_filler {
  86 |       type: "constant"
  87 |       value: 0.0
  88 |     }
  89 |   }
  90 | }
  91 | layer {
  92 |   name: "bn_2_1_pw"
  93 |   type: "BatchNorm"
  94 |   bottom: "conv_2_1_pw"
  95 |   top: "conv_2_1_pw"
  96 |   param {
  97 |     lr_mult: 0.0
  98 |     decay_mult: 0.0
  99 |   }
 100 |   param {
 101 |     lr_mult: 0.0
 102 |     decay_mult: 0.0
 103 |   }
 104 |   param {
 105 |     lr_mult: 0.0
 106 |     decay_mult: 0.0
 107 |   }
 108 |   batch_norm_param {
 109 |     use_global_stats: true
 110 |   }
 111 | }
 112 | layer {
 113 |   name: "scale_2_1_pw"
 114 |   type: "Scale"
 115 |   bottom: "conv_2_1_pw"
 116 |   top: "conv_2_1_pw"
 117 |   scale_param {
 118 |     bias_term: true
 119 |   }
 120 | }
 121 | layer {
 122 |   name: "relu_2_1_pw"
 123 |   type: "ReLU"
 124 |   bottom: "conv_2_1_pw"
 125 |   top: "conv_2_1_pw"
 126 | }
 127 | layer {
 128 |   name: "conv_2_1_dw"
 129 |   type: "Convolution"
 130 |   bottom: "conv_2_1_pw"
 131 |   top: "conv_2_1_dw"
 132 |   param {
 133 |     lr_mult: 1.0
 134 |     decay_mult: 1.0
 135 |   }
 136 |   convolution_param {
 137 |     num_output: 32
 138 |     bias_term: false
 139 |     pad: 1
 140 |     kernel_size: 3
 141 |     group: 32
 142 |     stride: 1
 143 |     weight_filler {
 144 |       type: "msra"
 145 |     }
 146 |     bias_filler {
 147 |       type: "constant"
 148 |       value: 0.0
 149 |     }
 150 |     engine: CAFFE
 151 |   }
 152 | }
 153 | layer {
 154 |   name: "bn_2_1_dw"
 155 |   type: "BatchNorm"
 156 |   bottom: "conv_2_1_dw"
 157 |   top: "conv_2_1_dw"
 158 |   param {
 159 |     lr_mult: 0.0
 160 |     decay_mult: 0.0
 161 |   }
 162 |   param {
 163 |     lr_mult: 0.0
 164 |     decay_mult: 0.0
 165 |   }
 166 |   param {
 167 |     lr_mult: 0.0
 168 |     decay_mult: 0.0
 169 |   }
 170 |   batch_norm_param {
 171 |     use_global_stats: true
 172 |   }
 173 | }
 174 | layer {
 175 |   name: "scale_2_1_dw"
 176 |   type: "Scale"
 177 |   bottom: "conv_2_1_dw"
 178 |   top: "conv_2_1_dw"
 179 |   scale_param {
 180 |     bias_term: true
 181 |   }
 182 | }
 183 | layer {
 184 |   name: "relu_2_1_dw"
 185 |   type: "ReLU"
 186 |   bottom: "conv_2_1_dw"
 187 |   top: "conv_2_1_dw"
 188 | }
 189 | layer {
 190 |   name: "conv_2_1_linear"
 191 |   type: "Convolution"
 192 |   bottom: "conv_2_1_dw"
 193 |   top: "conv_2_1_linear"
 194 |   param {
 195 |     lr_mult: 1.0
 196 |     decay_mult: 1.0
 197 |   }
 198 |   convolution_param {
 199 |     num_output: 16
 200 |     bias_term: false
 201 |     pad: 0
 202 |     kernel_size: 1
 203 |     stride: 1
 204 |     weight_filler {
 205 |       type: "msra"
 206 |     }
 207 |     bias_filler {
 208 |       type: "constant"
 209 |       value: 0.0
 210 |     }
 211 |   }
 212 | }
 213 | layer {
 214 |   name: "bn_2_1_linear"
 215 |   type: "BatchNorm"
 216 |   bottom: "conv_2_1_linear"
 217 |   top: "conv_2_1_linear"
 218 |   param {
 219 |     lr_mult: 0.0
 220 |     decay_mult: 0.0
 221 |   }
 222 |   param {
 223 |     lr_mult: 0.0
 224 |     decay_mult: 0.0
 225 |   }
 226 |   param {
 227 |     lr_mult: 0.0
 228 |     decay_mult: 0.0
 229 |   }
 230 |   batch_norm_param {
 231 |     use_global_stats: true
 232 |   }
 233 | }
 234 | layer {
 235 |   name: "scale_2_1_linear"
 236 |   type: "Scale"
 237 |   bottom: "conv_2_1_linear"
 238 |   top: "conv_2_1_linear"
 239 |   scale_param {
 240 |     bias_term: true
 241 |   }
 242 | }
 243 | layer {
 244 |   name: "conv_3_1_pw"
 245 |   type: "Convolution"
 246 |   bottom: "conv_2_1_linear"
 247 |   top: "conv_3_1_pw"
 248 |   param {
 249 |     lr_mult: 1.0
 250 |     decay_mult: 1.0
 251 |   }
 252 |   convolution_param {
 253 |     num_output: 96
 254 |     bias_term: false
 255 |     pad: 0
 256 |     kernel_size: 1
 257 |     stride: 1
 258 |     weight_filler {
 259 |       type: "msra"
 260 |     }
 261 |     bias_filler {
 262 |       type: "constant"
 263 |       value: 0.0
 264 |     }
 265 |   }
 266 | }
 267 | layer {
 268 |   name: "bn_3_1_pw"
 269 |   type: "BatchNorm"
 270 |   bottom: "conv_3_1_pw"
 271 |   top: "conv_3_1_pw"
 272 |   param {
 273 |     lr_mult: 0.0
 274 |     decay_mult: 0.0
 275 |   }
 276 |   param {
 277 |     lr_mult: 0.0
 278 |     decay_mult: 0.0
 279 |   }
 280 |   param {
 281 |     lr_mult: 0.0
 282 |     decay_mult: 0.0
 283 |   }
 284 |   batch_norm_param {
 285 |     use_global_stats: true
 286 |   }
 287 | }
 288 | layer {
 289 |   name: "scale_3_1_pw"
 290 |   type: "Scale"
 291 |   bottom: "conv_3_1_pw"
 292 |   top: "conv_3_1_pw"
 293 |   scale_param {
 294 |     bias_term: true
 295 |   }
 296 | }
 297 | layer {
 298 |   name: "relu_3_1_pw"
 299 |   type: "ReLU"
 300 |   bottom: "conv_3_1_pw"
 301 |   top: "conv_3_1_pw"
 302 | }
 303 | layer {
 304 |   name: "conv_3_1_dw"
 305 |   type: "Convolution"
 306 |   bottom: "conv_3_1_pw"
 307 |   top: "conv_3_1_dw"
 308 |   param {
 309 |     lr_mult: 1.0
 310 |     decay_mult: 1.0
 311 |   }
 312 |   convolution_param {
 313 |     num_output: 96
 314 |     bias_term: false
 315 |     pad: 1
 316 |     kernel_size: 3
 317 |     group: 96
 318 |     stride: 2
 319 |     weight_filler {
 320 |       type: "msra"
 321 |     }
 322 |     bias_filler {
 323 |       type: "constant"
 324 |       value: 0.0
 325 |     }
 326 |     engine: CAFFE
 327 |   }
 328 | }
 329 | layer {
 330 |   name: "bn_3_1_dw"
 331 |   type: "BatchNorm"
 332 |   bottom: "conv_3_1_dw"
 333 |   top: "conv_3_1_dw"
 334 |   param {
 335 |     lr_mult: 0.0
 336 |     decay_mult: 0.0
 337 |   }
 338 |   param {
 339 |     lr_mult: 0.0
 340 |     decay_mult: 0.0
 341 |   }
 342 |   param {
 343 |     lr_mult: 0.0
 344 |     decay_mult: 0.0
 345 |   }
 346 |   batch_norm_param {
 347 |     use_global_stats: true
 348 |   }
 349 | }
 350 | layer {
 351 |   name: "scale_3_1_dw"
 352 |   type: "Scale"
 353 |   bottom: "conv_3_1_dw"
 354 |   top: "conv_3_1_dw"
 355 |   scale_param {
 356 |     bias_term: true
 357 |   }
 358 | }
 359 | layer {
 360 |   name: "relu_3_1_dw"
 361 |   type: "ReLU"
 362 |   bottom: "conv_3_1_dw"
 363 |   top: "conv_3_1_dw"
 364 | }
 365 | layer {
 366 |   name: "conv_3_1_linear"
 367 |   type: "Convolution"
 368 |   bottom: "conv_3_1_dw"
 369 |   top: "conv_3_1_linear"
 370 |   param {
 371 |     lr_mult: 1.0
 372 |     decay_mult: 1.0
 373 |   }
 374 |   convolution_param {
 375 |     num_output: 24
 376 |     bias_term: false
 377 |     pad: 0
 378 |     kernel_size: 1
 379 |     stride: 1
 380 |     weight_filler {
 381 |       type: "msra"
 382 |     }
 383 |     bias_filler {
 384 |       type: "constant"
 385 |       value: 0.0
 386 |     }
 387 |   }
 388 | }
 389 | layer {
 390 |   name: "bn_3_1_linear"
 391 |   type: "BatchNorm"
 392 |   bottom: "conv_3_1_linear"
 393 |   top: "conv_3_1_linear"
 394 |   param {
 395 |     lr_mult: 0.0
 396 |     decay_mult: 0.0
 397 |   }
 398 |   param {
 399 |     lr_mult: 0.0
 400 |     decay_mult: 0.0
 401 |   }
 402 |   param {
 403 |     lr_mult: 0.0
 404 |     decay_mult: 0.0
 405 |   }
 406 |   batch_norm_param {
 407 |     use_global_stats: true
 408 |   }
 409 | }
 410 | layer {
 411 |   name: "scale_3_1_linear"
 412 |   type: "Scale"
 413 |   bottom: "conv_3_1_linear"
 414 |   top: "conv_3_1_linear"
 415 |   scale_param {
 416 |     bias_term: true
 417 |   }
 418 | }
 419 | layer {
 420 |   name: "conv_3_2_pw"
 421 |   type: "Convolution"
 422 |   bottom: "conv_3_1_linear"
 423 |   top: "conv_3_2_pw"
 424 |   param {
 425 |     lr_mult: 1.0
 426 |     decay_mult: 1.0
 427 |   }
 428 |   convolution_param {
 429 |     num_output: 144
 430 |     bias_term: false
 431 |     pad: 0
 432 |     kernel_size: 1
 433 |     stride: 1
 434 |     weight_filler {
 435 |       type: "msra"
 436 |     }
 437 |     bias_filler {
 438 |       type: "constant"
 439 |       value: 0.0
 440 |     }
 441 |   }
 442 | }
 443 | layer {
 444 |   name: "bn_3_2_pw"
 445 |   type: "BatchNorm"
 446 |   bottom: "conv_3_2_pw"
 447 |   top: "conv_3_2_pw"
 448 |   param {
 449 |     lr_mult: 0.0
 450 |     decay_mult: 0.0
 451 |   }
 452 |   param {
 453 |     lr_mult: 0.0
 454 |     decay_mult: 0.0
 455 |   }
 456 |   param {
 457 |     lr_mult: 0.0
 458 |     decay_mult: 0.0
 459 |   }
 460 |   batch_norm_param {
 461 |     use_global_stats: true
 462 |   }
 463 | }
 464 | layer {
 465 |   name: "scale_3_2_pw"
 466 |   type: "Scale"
 467 |   bottom: "conv_3_2_pw"
 468 |   top: "conv_3_2_pw"
 469 |   scale_param {
 470 |     bias_term: true
 471 |   }
 472 | }
 473 | layer {
 474 |   name: "relu_3_2_pw"
 475 |   type: "ReLU"
 476 |   bottom: "conv_3_2_pw"
 477 |   top: "conv_3_2_pw"
 478 | }
 479 | layer {
 480 |   name: "conv_3_2_dw"
 481 |   type: "Convolution"
 482 |   bottom: "conv_3_2_pw"
 483 |   top: "conv_3_2_dw"
 484 |   param {
 485 |     lr_mult: 1.0
 486 |     decay_mult: 1.0
 487 |   }
 488 |   convolution_param {
 489 |     num_output: 144
 490 |     bias_term: false
 491 |     pad: 1
 492 |     kernel_size: 3
 493 |     group: 144
 494 |     stride: 1
 495 |     weight_filler {
 496 |       type: "msra"
 497 |     }
 498 |     bias_filler {
 499 |       type: "constant"
 500 |       value: 0.0
 501 |     }
 502 |     engine: CAFFE
 503 |   }
 504 | }
 505 | layer {
 506 |   name: "bn_3_2_dw"
 507 |   type: "BatchNorm"
 508 |   bottom: "conv_3_2_dw"
 509 |   top: "conv_3_2_dw"
 510 |   param {
 511 |     lr_mult: 0.0
 512 |     decay_mult: 0.0
 513 |   }
 514 |   param {
 515 |     lr_mult: 0.0
 516 |     decay_mult: 0.0
 517 |   }
 518 |   param {
 519 |     lr_mult: 0.0
 520 |     decay_mult: 0.0
 521 |   }
 522 |   batch_norm_param {
 523 |     use_global_stats: true
 524 |   }
 525 | }
 526 | layer {
 527 |   name: "scale_3_2_dw"
 528 |   type: "Scale"
 529 |   bottom: "conv_3_2_dw"
 530 |   top: "conv_3_2_dw"
 531 |   scale_param {
 532 |     bias_term: true
 533 |   }
 534 | }
 535 | layer {
 536 |   name: "relu_3_2_dw"
 537 |   type: "ReLU"
 538 |   bottom: "conv_3_2_dw"
 539 |   top: "conv_3_2_dw"
 540 | }
 541 | layer {
 542 |   name: "conv_3_2_linear"
 543 |   type: "Convolution"
 544 |   bottom: "conv_3_2_dw"
 545 |   top: "conv_3_2_linear"
 546 |   param {
 547 |     lr_mult: 1.0
 548 |     decay_mult: 1.0
 549 |   }
 550 |   convolution_param {
 551 |     num_output: 24
 552 |     bias_term: false
 553 |     pad: 0
 554 |     kernel_size: 1
 555 |     stride: 1
 556 |     weight_filler {
 557 |       type: "msra"
 558 |     }
 559 |     bias_filler {
 560 |       type: "constant"
 561 |       value: 0.0
 562 |     }
 563 |   }
 564 | }
 565 | layer {
 566 |   name: "bn_3_2_linear"
 567 |   type: "BatchNorm"
 568 |   bottom: "conv_3_2_linear"
 569 |   top: "conv_3_2_linear"
 570 |   param {
 571 |     lr_mult: 0.0
 572 |     decay_mult: 0.0
 573 |   }
 574 |   param {
 575 |     lr_mult: 0.0
 576 |     decay_mult: 0.0
 577 |   }
 578 |   param {
 579 |     lr_mult: 0.0
 580 |     decay_mult: 0.0
 581 |   }
 582 |   batch_norm_param {
 583 |     use_global_stats: true
 584 |   }
 585 | }
 586 | layer {
 587 |   name: "scale_3_2_linear"
 588 |   type: "Scale"
 589 |   bottom: "conv_3_2_linear"
 590 |   top: "conv_3_2_linear"
 591 |   scale_param {
 592 |     bias_term: true
 593 |   }
 594 | }
 595 | layer {
 596 |   name: "add_3_2"
 597 |   type: "Eltwise"
 598 |   bottom: "conv_3_1_linear"
 599 |   bottom: "conv_3_2_linear"
 600 |   top: "add_3_2"
 601 | }
 602 | layer {
 603 |   name: "conv_4_1_pw"
 604 |   type: "Convolution"
 605 |   bottom: "add_3_2"
 606 |   top: "conv_4_1_pw"
 607 |   param {
 608 |     lr_mult: 1.0
 609 |     decay_mult: 1.0
 610 |   }
 611 |   convolution_param {
 612 |     num_output: 144
 613 |     bias_term: false
 614 |     pad: 0
 615 |     kernel_size: 1
 616 |     stride: 1
 617 |     weight_filler {
 618 |       type: "msra"
 619 |     }
 620 |     bias_filler {
 621 |       type: "constant"
 622 |       value: 0.0
 623 |     }
 624 |   }
 625 | }
 626 | layer {
 627 |   name: "bn_4_1_pw"
 628 |   type: "BatchNorm"
 629 |   bottom: "conv_4_1_pw"
 630 |   top: "conv_4_1_pw"
 631 |   param {
 632 |     lr_mult: 0.0
 633 |     decay_mult: 0.0
 634 |   }
 635 |   param {
 636 |     lr_mult: 0.0
 637 |     decay_mult: 0.0
 638 |   }
 639 |   param {
 640 |     lr_mult: 0.0
 641 |     decay_mult: 0.0
 642 |   }
 643 |   batch_norm_param {
 644 |     use_global_stats: true
 645 |   }
 646 | }
 647 | layer {
 648 |   name: "scale_4_1_pw"
 649 |   type: "Scale"
 650 |   bottom: "conv_4_1_pw"
 651 |   top: "conv_4_1_pw"
 652 |   scale_param {
 653 |     bias_term: true
 654 |   }
 655 | }
 656 | layer {
 657 |   name: "relu_4_1_pw"
 658 |   type: "ReLU"
 659 |   bottom: "conv_4_1_pw"
 660 |   top: "conv_4_1_pw"
 661 | }
 662 | layer {
 663 |   name: "conv_4_1_dw"
 664 |   type: "Convolution"
 665 |   bottom: "conv_4_1_pw"
 666 |   top: "conv_4_1_dw"
 667 |   param {
 668 |     lr_mult: 1.0
 669 |     decay_mult: 1.0
 670 |   }
 671 |   convolution_param {
 672 |     num_output: 144
 673 |     bias_term: false
 674 |     pad: 1
 675 |     kernel_size: 3
 676 |     group: 144
 677 |     stride: 2
 678 |     weight_filler {
 679 |       type: "msra"
 680 |     }
 681 |     bias_filler {
 682 |       type: "constant"
 683 |       value: 0.0
 684 |     }
 685 |     engine: CAFFE
 686 |   }
 687 | }
 688 | layer {
 689 |   name: "bn_4_1_dw"
 690 |   type: "BatchNorm"
 691 |   bottom: "conv_4_1_dw"
 692 |   top: "conv_4_1_dw"
 693 |   param {
 694 |     lr_mult: 0.0
 695 |     decay_mult: 0.0
 696 |   }
 697 |   param {
 698 |     lr_mult: 0.0
 699 |     decay_mult: 0.0
 700 |   }
 701 |   param {
 702 |     lr_mult: 0.0
 703 |     decay_mult: 0.0
 704 |   }
 705 |   batch_norm_param {
 706 |     use_global_stats: true
 707 |   }
 708 | }
 709 | layer {
 710 |   name: "scale_4_1_dw"
 711 |   type: "Scale"
 712 |   bottom: "conv_4_1_dw"
 713 |   top: "conv_4_1_dw"
 714 |   scale_param {
 715 |     bias_term: true
 716 |   }
 717 | }
 718 | layer {
 719 |   name: "relu_4_1_dw"
 720 |   type: "ReLU"
 721 |   bottom: "conv_4_1_dw"
 722 |   top: "conv_4_1_dw"
 723 | }
 724 | layer {
 725 |   name: "conv_4_1_linear"
 726 |   type: "Convolution"
 727 |   bottom: "conv_4_1_dw"
 728 |   top: "conv_4_1_linear"
 729 |   param {
 730 |     lr_mult: 1.0
 731 |     decay_mult: 1.0
 732 |   }
 733 |   convolution_param {
 734 |     num_output: 32
 735 |     bias_term: false
 736 |     pad: 0
 737 |     kernel_size: 1
 738 |     stride: 1
 739 |     weight_filler {
 740 |       type: "msra"
 741 |     }
 742 |     bias_filler {
 743 |       type: "constant"
 744 |       value: 0.0
 745 |     }
 746 |   }
 747 | }
 748 | layer {
 749 |   name: "bn_4_1_linear"
 750 |   type: "BatchNorm"
 751 |   bottom: "conv_4_1_linear"
 752 |   top: "conv_4_1_linear"
 753 |   param {
 754 |     lr_mult: 0.0
 755 |     decay_mult: 0.0
 756 |   }
 757 |   param {
 758 |     lr_mult: 0.0
 759 |     decay_mult: 0.0
 760 |   }
 761 |   param {
 762 |     lr_mult: 0.0
 763 |     decay_mult: 0.0
 764 |   }
 765 |   batch_norm_param {
 766 |     use_global_stats: true
 767 |   }
 768 | }
 769 | layer {
 770 |   name: "scale_4_1_linear"
 771 |   type: "Scale"
 772 |   bottom: "conv_4_1_linear"
 773 |   top: "conv_4_1_linear"
 774 |   scale_param {
 775 |     bias_term: true
 776 |   }
 777 | }
 778 | layer {
 779 |   name: "conv_4_2_pw"
 780 |   type: "Convolution"
 781 |   bottom: "conv_4_1_linear"
 782 |   top: "conv_4_2_pw"
 783 |   param {
 784 |     lr_mult: 1.0
 785 |     decay_mult: 1.0
 786 |   }
 787 |   convolution_param {
 788 |     num_output: 192
 789 |     bias_term: false
 790 |     pad: 0
 791 |     kernel_size: 1
 792 |     stride: 1
 793 |     weight_filler {
 794 |       type: "msra"
 795 |     }
 796 |     bias_filler {
 797 |       type: "constant"
 798 |       value: 0.0
 799 |     }
 800 |   }
 801 | }
 802 | layer {
 803 |   name: "bn_4_2_pw"
 804 |   type: "BatchNorm"
 805 |   bottom: "conv_4_2_pw"
 806 |   top: "conv_4_2_pw"
 807 |   param {
 808 |     lr_mult: 0.0
 809 |     decay_mult: 0.0
 810 |   }
 811 |   param {
 812 |     lr_mult: 0.0
 813 |     decay_mult: 0.0
 814 |   }
 815 |   param {
 816 |     lr_mult: 0.0
 817 |     decay_mult: 0.0
 818 |   }
 819 |   batch_norm_param {
 820 |     use_global_stats: true
 821 |   }
 822 | }
 823 | layer {
 824 |   name: "scale_4_2_pw"
 825 |   type: "Scale"
 826 |   bottom: "conv_4_2_pw"
 827 |   top: "conv_4_2_pw"
 828 |   scale_param {
 829 |     bias_term: true
 830 |   }
 831 | }
 832 | layer {
 833 |   name: "relu_4_2_pw"
 834 |   type: "ReLU"
 835 |   bottom: "conv_4_2_pw"
 836 |   top: "conv_4_2_pw"
 837 | }
 838 | layer {
 839 |   name: "conv_4_2_dw"
 840 |   type: "Convolution"
 841 |   bottom: "conv_4_2_pw"
 842 |   top: "conv_4_2_dw"
 843 |   param {
 844 |     lr_mult: 1.0
 845 |     decay_mult: 1.0
 846 |   }
 847 |   convolution_param {
 848 |     num_output: 192
 849 |     bias_term: false
 850 |     pad: 1
 851 |     kernel_size: 3
 852 |     group: 192
 853 |     stride: 1
 854 |     weight_filler {
 855 |       type: "msra"
 856 |     }
 857 |     bias_filler {
 858 |       type: "constant"
 859 |       value: 0.0
 860 |     }
 861 |     engine: CAFFE
 862 |   }
 863 | }
 864 | layer {
 865 |   name: "bn_4_2_dw"
 866 |   type: "BatchNorm"
 867 |   bottom: "conv_4_2_dw"
 868 |   top: "conv_4_2_dw"
 869 |   param {
 870 |     lr_mult: 0.0
 871 |     decay_mult: 0.0
 872 |   }
 873 |   param {
 874 |     lr_mult: 0.0
 875 |     decay_mult: 0.0
 876 |   }
 877 |   param {
 878 |     lr_mult: 0.0
 879 |     decay_mult: 0.0
 880 |   }
 881 |   batch_norm_param {
 882 |     use_global_stats: true
 883 |   }
 884 | }
 885 | layer {
 886 |   name: "scale_4_2_dw"
 887 |   type: "Scale"
 888 |   bottom: "conv_4_2_dw"
 889 |   top: "conv_4_2_dw"
 890 |   scale_param {
 891 |     bias_term: true
 892 |   }
 893 | }
 894 | layer {
 895 |   name: "relu_4_2_dw"
 896 |   type: "ReLU"
 897 |   bottom: "conv_4_2_dw"
 898 |   top: "conv_4_2_dw"
 899 | }
 900 | layer {
 901 |   name: "conv_4_2_linear"
 902 |   type: "Convolution"
 903 |   bottom: "conv_4_2_dw"
 904 |   top: "conv_4_2_linear"
 905 |   param {
 906 |     lr_mult: 1.0
 907 |     decay_mult: 1.0
 908 |   }
 909 |   convolution_param {
 910 |     num_output: 32
 911 |     bias_term: false
 912 |     pad: 0
 913 |     kernel_size: 1
 914 |     stride: 1
 915 |     weight_filler {
 916 |       type: "msra"
 917 |     }
 918 |     bias_filler {
 919 |       type: "constant"
 920 |       value: 0.0
 921 |     }
 922 |   }
 923 | }
 924 | layer {
 925 |   name: "bn_4_2_linear"
 926 |   type: "BatchNorm"
 927 |   bottom: "conv_4_2_linear"
 928 |   top: "conv_4_2_linear"
 929 |   param {
 930 |     lr_mult: 0.0
 931 |     decay_mult: 0.0
 932 |   }
 933 |   param {
 934 |     lr_mult: 0.0
 935 |     decay_mult: 0.0
 936 |   }
 937 |   param {
 938 |     lr_mult: 0.0
 939 |     decay_mult: 0.0
 940 |   }
 941 |   batch_norm_param {
 942 |     use_global_stats: true
 943 |   }
 944 | }
 945 | layer {
 946 |   name: "scale_4_2_linear"
 947 |   type: "Scale"
 948 |   bottom: "conv_4_2_linear"
 949 |   top: "conv_4_2_linear"
 950 |   scale_param {
 951 |     bias_term: true
 952 |   }
 953 | }
 954 | layer {
 955 |   name: "add_4_2"
 956 |   type: "Eltwise"
 957 |   bottom: "conv_4_1_linear"
 958 |   bottom: "conv_4_2_linear"
 959 |   top: "add_4_2"
 960 | }
 961 | layer {
 962 |   name: "conv_4_3_pw"
 963 |   type: "Convolution"
 964 |   bottom: "add_4_2"
 965 |   top: "conv_4_3_pw"
 966 |   param {
 967 |     lr_mult: 1.0
 968 |     decay_mult: 1.0
 969 |   }
 970 |   convolution_param {
 971 |     num_output: 192
 972 |     bias_term: false
 973 |     pad: 0
 974 |     kernel_size: 1
 975 |     stride: 1
 976 |     weight_filler {
 977 |       type: "msra"
 978 |     }
 979 |     bias_filler {
 980 |       type: "constant"
 981 |       value: 0.0
 982 |     }
 983 |   }
 984 | }
 985 | layer {
 986 |   name: "bn_4_3_pw"
 987 |   type: "BatchNorm"
 988 |   bottom: "conv_4_3_pw"
 989 |   top: "conv_4_3_pw"
 990 |   param {
 991 |     lr_mult: 0.0
 992 |     decay_mult: 0.0
 993 |   }
 994 |   param {
 995 |     lr_mult: 0.0
 996 |     decay_mult: 0.0
 997 |   }
 998 |   param {
 999 |     lr_mult: 0.0
1000 |     decay_mult: 0.0
1001 |   }
1002 |   batch_norm_param {
1003 |     use_global_stats: true
1004 |   }
1005 | }
1006 | layer {
1007 |   name: "scale_4_3_pw"
1008 |   type: "Scale"
1009 |   bottom: "conv_4_3_pw"
1010 |   top: "conv_4_3_pw"
1011 |   scale_param {
1012 |     bias_term: true
1013 |   }
1014 | }
1015 | layer {
1016 |   name: "relu_4_3_pw"
1017 |   type: "ReLU"
1018 |   bottom: "conv_4_3_pw"
1019 |   top: "conv_4_3_pw"
1020 | }
1021 | layer {
1022 |   name: "conv_4_3_dw"
1023 |   type: "Convolution"
1024 |   bottom: "conv_4_3_pw"
1025 |   top: "conv_4_3_dw"
1026 |   param {
1027 |     lr_mult: 1.0
1028 |     decay_mult: 1.0
1029 |   }
1030 |   convolution_param {
1031 |     num_output: 192
1032 |     bias_term: false
1033 |     pad: 1
1034 |     kernel_size: 3
1035 |     group: 192
1036 |     stride: 1
1037 |     weight_filler {
1038 |       type: "msra"
1039 |     }
1040 |     bias_filler {
1041 |       type: "constant"
1042 |       value: 0.0
1043 |     }
1044 |     engine: CAFFE
1045 |   }
1046 | }
1047 | layer {
1048 |   name: "bn_4_3_dw"
1049 |   type: "BatchNorm"
1050 |   bottom: "conv_4_3_dw"
1051 |   top: "conv_4_3_dw"
1052 |   param {
1053 |     lr_mult: 0.0
1054 |     decay_mult: 0.0
1055 |   }
1056 |   param {
1057 |     lr_mult: 0.0
1058 |     decay_mult: 0.0
1059 |   }
1060 |   param {
1061 |     lr_mult: 0.0
1062 |     decay_mult: 0.0
1063 |   }
1064 |   batch_norm_param {
1065 |     use_global_stats: true
1066 |   }
1067 | }
1068 | layer {
1069 |   name: "scale_4_3_dw"
1070 |   type: "Scale"
1071 |   bottom: "conv_4_3_dw"
1072 |   top: "conv_4_3_dw"
1073 |   scale_param {
1074 |     bias_term: true
1075 |   }
1076 | }
1077 | layer {
1078 |   name: "relu_4_3_dw"
1079 |   type: "ReLU"
1080 |   bottom: "conv_4_3_dw"
1081 |   top: "conv_4_3_dw"
1082 | }
1083 | layer {
1084 |   name: "conv_4_3_linear"
1085 |   type: "Convolution"
1086 |   bottom: "conv_4_3_dw"
1087 |   top: "conv_4_3_linear"
1088 |   param {
1089 |     lr_mult: 1.0
1090 |     decay_mult: 1.0
1091 |   }
1092 |   convolution_param {
1093 |     num_output: 32
1094 |     bias_term: false
1095 |     pad: 0
1096 |     kernel_size: 1
1097 |     stride: 1
1098 |     weight_filler {
1099 |       type: "msra"
1100 |     }
1101 |     bias_filler {
1102 |       type: "constant"
1103 |       value: 0.0
1104 |     }
1105 |   }
1106 | }
1107 | layer {
1108 |   name: "bn_4_3_linear"
1109 |   type: "BatchNorm"
1110 |   bottom: "conv_4_3_linear"
1111 |   top: "conv_4_3_linear"
1112 |   param {
1113 |     lr_mult: 0.0
1114 |     decay_mult: 0.0
1115 |   }
1116 |   param {
1117 |     lr_mult: 0.0
1118 |     decay_mult: 0.0
1119 |   }
1120 |   param {
1121 |     lr_mult: 0.0
1122 |     decay_mult: 0.0
1123 |   }
1124 |   batch_norm_param {
1125 |     use_global_stats: true
1126 |   }
1127 | }
1128 | layer {
1129 |   name: "scale_4_3_linear"
1130 |   type: "Scale"
1131 |   bottom: "conv_4_3_linear"
1132 |   top: "conv_4_3_linear"
1133 |   scale_param {
1134 |     bias_term: true
1135 |   }
1136 | }
1137 | layer {
1138 |   name: "add_4_3"
1139 |   type: "Eltwise"
1140 |   bottom: "add_4_2"
1141 |   bottom: "conv_4_3_linear"
1142 |   top: "add_4_3"
1143 | }
1144 | layer {
1145 |   name: "conv_5_1_pw"
1146 |   type: "Convolution"
1147 |   bottom: "add_4_3"
1148 |   top: "conv_5_1_pw"
1149 |   param {
1150 |     lr_mult: 1.0
1151 |     decay_mult: 1.0
1152 |   }
1153 |   convolution_param {
1154 |     num_output: 192
1155 |     bias_term: false
1156 |     pad: 0
1157 |     kernel_size: 1
1158 |     stride: 1
1159 |     weight_filler {
1160 |       type: "msra"
1161 |     }
1162 |     bias_filler {
1163 |       type: "constant"
1164 |       value: 0.0
1165 |     }
1166 |   }
1167 | }
1168 | layer {
1169 |   name: "bn_5_1_pw"
1170 |   type: "BatchNorm"
1171 |   bottom: "conv_5_1_pw"
1172 |   top: "conv_5_1_pw"
1173 |   param {
1174 |     lr_mult: 0.0
1175 |     decay_mult: 0.0
1176 |   }
1177 |   param {
1178 |     lr_mult: 0.0
1179 |     decay_mult: 0.0
1180 |   }
1181 |   param {
1182 |     lr_mult: 0.0
1183 |     decay_mult: 0.0
1184 |   }
1185 |   batch_norm_param {
1186 |     use_global_stats: true
1187 |   }
1188 | }
1189 | layer {
1190 |   name: "scale_5_1_pw"
1191 |   type: "Scale"
1192 |   bottom: "conv_5_1_pw"
1193 |   top: "conv_5_1_pw"
1194 |   scale_param {
1195 |     bias_term: true
1196 |   }
1197 | }
1198 | layer {
1199 |   name: "relu_5_1_pw"
1200 |   type: "ReLU"
1201 |   bottom: "conv_5_1_pw"
1202 |   top: "conv_5_1_pw"
1203 | }
1204 | layer {
1205 |   name: "conv_5_1_dw"
1206 |   type: "Convolution"
1207 |   bottom: "conv_5_1_pw"
1208 |   top: "conv_5_1_dw"
1209 |   param {
1210 |     lr_mult: 1.0
1211 |     decay_mult: 1.0
1212 |   }
1213 |   convolution_param {
1214 |     num_output: 192
1215 |     bias_term: false
1216 |     pad: 1
1217 |     kernel_size: 3
1218 |     group: 192
1219 |     stride: 2
1220 |     weight_filler {
1221 |       type: "msra"
1222 |     }
1223 |     bias_filler {
1224 |       type: "constant"
1225 |       value: 0.0
1226 |     }
1227 |     engine: CAFFE
1228 |   }
1229 | }
1230 | layer {
1231 |   name: "bn_5_1_dw"
1232 |   type: "BatchNorm"
1233 |   bottom: "conv_5_1_dw"
1234 |   top: "conv_5_1_dw"
1235 |   param {
1236 |     lr_mult: 0.0
1237 |     decay_mult: 0.0
1238 |   }
1239 |   param {
1240 |     lr_mult: 0.0
1241 |     decay_mult: 0.0
1242 |   }
1243 |   param {
1244 |     lr_mult: 0.0
1245 |     decay_mult: 0.0
1246 |   }
1247 |   batch_norm_param {
1248 |     use_global_stats: true
1249 |   }
1250 | }
1251 | layer {
1252 |   name: "scale_5_1_dw"
1253 |   type: "Scale"
1254 |   bottom: "conv_5_1_dw"
1255 |   top: "conv_5_1_dw"
1256 |   scale_param {
1257 |     bias_term: true
1258 |   }
1259 | }
1260 | layer {
1261 |   name: "relu_5_1_dw"
1262 |   type: "ReLU"
1263 |   bottom: "conv_5_1_dw"
1264 |   top: "conv_5_1_dw"
1265 | }
1266 | layer {
1267 |   name: "conv_5_1_linear"
1268 |   type: "Convolution"
1269 |   bottom: "conv_5_1_dw"
1270 |   top: "conv_5_1_linear"
1271 |   param {
1272 |     lr_mult: 1.0
1273 |     decay_mult: 1.0
1274 |   }
1275 |   convolution_param {
1276 |     num_output: 64
1277 |     bias_term: false
1278 |     pad: 0
1279 |     kernel_size: 1
1280 |     stride: 1
1281 |     weight_filler {
1282 |       type: "msra"
1283 |     }
1284 |     bias_filler {
1285 |       type: "constant"
1286 |       value: 0.0
1287 |     }
1288 |   }
1289 | }
1290 | layer {
1291 |   name: "bn_5_1_linear"
1292 |   type: "BatchNorm"
1293 |   bottom: "conv_5_1_linear"
1294 |   top: "conv_5_1_linear"
1295 |   param {
1296 |     lr_mult: 0.0
1297 |     decay_mult: 0.0
1298 |   }
1299 |   param {
1300 |     lr_mult: 0.0
1301 |     decay_mult: 0.0
1302 |   }
1303 |   param {
1304 |     lr_mult: 0.0
1305 |     decay_mult: 0.0
1306 |   }
1307 |   batch_norm_param {
1308 |     use_global_stats: true
1309 |   }
1310 | }
1311 | layer {
1312 |   name: "scale_5_1_linear"
1313 |   type: "Scale"
1314 |   bottom: "conv_5_1_linear"
1315 |   top: "conv_5_1_linear"
1316 |   scale_param {
1317 |     bias_term: true
1318 |   }
1319 | }
1320 | layer {
1321 |   name: "conv_5_2_pw"
1322 |   type: "Convolution"
1323 |   bottom: "conv_5_1_linear"
1324 |   top: "conv_5_2_pw"
1325 |   param {
1326 |     lr_mult: 1.0
1327 |     decay_mult: 1.0
1328 |   }
1329 |   convolution_param {
1330 |     num_output: 384
1331 |     bias_term: false
1332 |     pad: 0
1333 |     kernel_size: 1
1334 |     stride: 1
1335 |     weight_filler {
1336 |       type: "msra"
1337 |     }
1338 |     bias_filler {
1339 |       type: "constant"
1340 |       value: 0.0
1341 |     }
1342 |   }
1343 | }
1344 | layer {
1345 |   name: "bn_5_2_pw"
1346 |   type: "BatchNorm"
1347 |   bottom: "conv_5_2_pw"
1348 |   top: "conv_5_2_pw"
1349 |   param {
1350 |     lr_mult: 0.0
1351 |     decay_mult: 0.0
1352 |   }
1353 |   param {
1354 |     lr_mult: 0.0
1355 |     decay_mult: 0.0
1356 |   }
1357 |   param {
1358 |     lr_mult: 0.0
1359 |     decay_mult: 0.0
1360 |   }
1361 |   batch_norm_param {
1362 |     use_global_stats: true
1363 |   }
1364 | }
1365 | layer {
1366 |   name: "scale_5_2_pw"
1367 |   type: "Scale"
1368 |   bottom: "conv_5_2_pw"
1369 |   top: "conv_5_2_pw"
1370 |   scale_param {
1371 |     bias_term: true
1372 |   }
1373 | }
1374 | layer {
1375 |   name: "relu_5_2_pw"
1376 |   type: "ReLU"
1377 |   bottom: "conv_5_2_pw"
1378 |   top: "conv_5_2_pw"
1379 | }
1380 | layer {
1381 |   name: "conv_5_2_dw"
1382 |   type: "Convolution"
1383 |   bottom: "conv_5_2_pw"
1384 |   top: "conv_5_2_dw"
1385 |   param {
1386 |     lr_mult: 1.0
1387 |     decay_mult: 1.0
1388 |   }
1389 |   convolution_param {
1390 |     num_output: 384
1391 |     bias_term: false
1392 |     pad: 1
1393 |     kernel_size: 3
1394 |     group: 384
1395 |     stride: 1
1396 |     weight_filler {
1397 |       type: "msra"
1398 |     }
1399 |     bias_filler {
1400 |       type: "constant"
1401 |       value: 0.0
1402 |     }
1403 |     engine: CAFFE
1404 |   }
1405 | }
1406 | layer {
1407 |   name: "bn_5_2_dw"
1408 |   type: "BatchNorm"
1409 |   bottom: "conv_5_2_dw"
1410 |   top: "conv_5_2_dw"
1411 |   param {
1412 |     lr_mult: 0.0
1413 |     decay_mult: 0.0
1414 |   }
1415 |   param {
1416 |     lr_mult: 0.0
1417 |     decay_mult: 0.0
1418 |   }
1419 |   param {
1420 |     lr_mult: 0.0
1421 |     decay_mult: 0.0
1422 |   }
1423 |   batch_norm_param {
1424 |     use_global_stats: true
1425 |   }
1426 | }
1427 | layer {
1428 |   name: "scale_5_2_dw"
1429 |   type: "Scale"
1430 |   bottom: "conv_5_2_dw"
1431 |   top: "conv_5_2_dw"
1432 |   scale_param {
1433 |     bias_term: true
1434 |   }
1435 | }
1436 | layer {
1437 |   name: "relu_5_2_dw"
1438 |   type: "ReLU"
1439 |   bottom: "conv_5_2_dw"
1440 |   top: "conv_5_2_dw"
1441 | }
1442 | layer {
1443 |   name: "conv_5_2_linear"
1444 |   type: "Convolution"
1445 |   bottom: "conv_5_2_dw"
1446 |   top: "conv_5_2_linear"
1447 |   param {
1448 |     lr_mult: 1.0
1449 |     decay_mult: 1.0
1450 |   }
1451 |   convolution_param {
1452 |     num_output: 64
1453 |     bias_term: false
1454 |     pad: 0
1455 |     kernel_size: 1
1456 |     stride: 1
1457 |     weight_filler {
1458 |       type: "msra"
1459 |     }
1460 |     bias_filler {
1461 |       type: "constant"
1462 |       value: 0.0
1463 |     }
1464 |   }
1465 | }
1466 | layer {
1467 |   name: "bn_5_2_linear"
1468 |   type: "BatchNorm"
1469 |   bottom: "conv_5_2_linear"
1470 |   top: "conv_5_2_linear"
1471 |   param {
1472 |     lr_mult: 0.0
1473 |     decay_mult: 0.0
1474 |   }
1475 |   param {
1476 |     lr_mult: 0.0
1477 |     decay_mult: 0.0
1478 |   }
1479 |   param {
1480 |     lr_mult: 0.0
1481 |     decay_mult: 0.0
1482 |   }
1483 |   batch_norm_param {
1484 |     use_global_stats: true
1485 |   }
1486 | }
1487 | layer {
1488 |   name: "scale_5_2_linear"
1489 |   type: "Scale"
1490 |   bottom: "conv_5_2_linear"
1491 |   top: "conv_5_2_linear"
1492 |   scale_param {
1493 |     bias_term: true
1494 |   }
1495 | }
1496 | layer {
1497 |   name: "add_5_2"
1498 |   type: "Eltwise"
1499 |   bottom: "conv_5_1_linear"
1500 |   bottom: "conv_5_2_linear"
1501 |   top: "add_5_2"
1502 | }
1503 | layer {
1504 |   name: "conv_5_3_pw"
1505 |   type: "Convolution"
1506 |   bottom: "add_5_2"
1507 |   top: "conv_5_3_pw"
1508 |   param {
1509 |     lr_mult: 1.0
1510 |     decay_mult: 1.0
1511 |   }
1512 |   convolution_param {
1513 |     num_output: 384
1514 |     bias_term: false
1515 |     pad: 0
1516 |     kernel_size: 1
1517 |     stride: 1
1518 |     weight_filler {
1519 |       type: "msra"
1520 |     }
1521 |     bias_filler {
1522 |       type: "constant"
1523 |       value: 0.0
1524 |     }
1525 |   }
1526 | }
1527 | layer {
1528 |   name: "bn_5_3_pw"
1529 |   type: "BatchNorm"
1530 |   bottom: "conv_5_3_pw"
1531 |   top: "conv_5_3_pw"
1532 |   param {
1533 |     lr_mult: 0.0
1534 |     decay_mult: 0.0
1535 |   }
1536 |   param {
1537 |     lr_mult: 0.0
1538 |     decay_mult: 0.0
1539 |   }
1540 |   param {
1541 |     lr_mult: 0.0
1542 |     decay_mult: 0.0
1543 |   }
1544 |   batch_norm_param {
1545 |     use_global_stats: true
1546 |   }
1547 | }
1548 | layer {
1549 |   name: "scale_5_3_pw"
1550 |   type: "Scale"
1551 |   bottom: "conv_5_3_pw"
1552 |   top: "conv_5_3_pw"
1553 |   scale_param {
1554 |     bias_term: true
1555 |   }
1556 | }
1557 | layer {
1558 |   name: "relu_5_3_pw"
1559 |   type: "ReLU"
1560 |   bottom: "conv_5_3_pw"
1561 |   top: "conv_5_3_pw"
1562 | }
1563 | layer {
1564 |   name: "conv_5_3_dw"
1565 |   type: "Convolution"
1566 |   bottom: "conv_5_3_pw"
1567 |   top: "conv_5_3_dw"
1568 |   param {
1569 |     lr_mult: 1.0
1570 |     decay_mult: 1.0
1571 |   }
1572 |   convolution_param {
1573 |     num_output: 384
1574 |     bias_term: false
1575 |     pad: 1
1576 |     kernel_size: 3
1577 |     group: 384
1578 |     stride: 1
1579 |     weight_filler {
1580 |       type: "msra"
1581 |     }
1582 |     bias_filler {
1583 |       type: "constant"
1584 |       value: 0.0
1585 |     }
1586 |     engine: CAFFE
1587 |   }
1588 | }
1589 | layer {
1590 |   name: "bn_5_3_dw"
1591 |   type: "BatchNorm"
1592 |   bottom: "conv_5_3_dw"
1593 |   top: "conv_5_3_dw"
1594 |   param {
1595 |     lr_mult: 0.0
1596 |     decay_mult: 0.0
1597 |   }
1598 |   param {
1599 |     lr_mult: 0.0
1600 |     decay_mult: 0.0
1601 |   }
1602 |   param {
1603 |     lr_mult: 0.0
1604 |     decay_mult: 0.0
1605 |   }
1606 |   batch_norm_param {
1607 |     use_global_stats: true
1608 |   }
1609 | }
1610 | layer {
1611 |   name: "scale_5_3_dw"
1612 |   type: "Scale"
1613 |   bottom: "conv_5_3_dw"
1614 |   top: "conv_5_3_dw"
1615 |   scale_param {
1616 |     bias_term: true
1617 |   }
1618 | }
1619 | layer {
1620 |   name: "relu_5_3_dw"
1621 |   type: "ReLU"
1622 |   bottom: "conv_5_3_dw"
1623 |   top: "conv_5_3_dw"
1624 | }
1625 | layer {
1626 |   name: "conv_5_3_linear"
1627 |   type: "Convolution"
1628 |   bottom: "conv_5_3_dw"
1629 |   top: "conv_5_3_linear"
1630 |   param {
1631 |     lr_mult: 1.0
1632 |     decay_mult: 1.0
1633 |   }
1634 |   convolution_param {
1635 |     num_output: 64
1636 |     bias_term: false
1637 |     pad: 0
1638 |     kernel_size: 1
1639 |     stride: 1
1640 |     weight_filler {
1641 |       type: "msra"
1642 |     }
1643 |     bias_filler {
1644 |       type: "constant"
1645 |       value: 0.0
1646 |     }
1647 |   }
1648 | }
1649 | layer {
1650 |   name: "bn_5_3_linear"
1651 |   type: "BatchNorm"
1652 |   bottom: "conv_5_3_linear"
1653 |   top: "conv_5_3_linear"
1654 |   param {
1655 |     lr_mult: 0.0
1656 |     decay_mult: 0.0
1657 |   }
1658 |   param {
1659 |     lr_mult: 0.0
1660 |     decay_mult: 0.0
1661 |   }
1662 |   param {
1663 |     lr_mult: 0.0
1664 |     decay_mult: 0.0
1665 |   }
1666 |   batch_norm_param {
1667 |     use_global_stats: true
1668 |   }
1669 | }
1670 | layer {
1671 |   name: "scale_5_3_linear"
1672 |   type: "Scale"
1673 |   bottom: "conv_5_3_linear"
1674 |   top: "conv_5_3_linear"
1675 |   scale_param {
1676 |     bias_term: true
1677 |   }
1678 | }
1679 | layer {
1680 |   name: "add_5_3"
1681 |   type: "Eltwise"
1682 |   bottom: "add_5_2"
1683 |   bottom: "conv_5_3_linear"
1684 |   top: "add_5_3"
1685 | }
1686 | layer {
1687 |   name: "conv_5_4_pw"
1688 |   type: "Convolution"
1689 |   bottom: "add_5_3"
1690 |   top: "conv_5_4_pw"
1691 |   param {
1692 |     lr_mult: 1.0
1693 |     decay_mult: 1.0
1694 |   }
1695 |   convolution_param {
1696 |     num_output: 384
1697 |     bias_term: false
1698 |     pad: 0
1699 |     kernel_size: 1
1700 |     stride: 1
1701 |     weight_filler {
1702 |       type: "msra"
1703 |     }
1704 |     bias_filler {
1705 |       type: "constant"
1706 |       value: 0.0
1707 |     }
1708 |   }
1709 | }
1710 | layer {
1711 |   name: "bn_5_4_pw"
1712 |   type: "BatchNorm"
1713 |   bottom: "conv_5_4_pw"
1714 |   top: "conv_5_4_pw"
1715 |   param {
1716 |     lr_mult: 0.0
1717 |     decay_mult: 0.0
1718 |   }
1719 |   param {
1720 |     lr_mult: 0.0
1721 |     decay_mult: 0.0
1722 |   }
1723 |   param {
1724 |     lr_mult: 0.0
1725 |     decay_mult: 0.0
1726 |   }
1727 |   batch_norm_param {
1728 |     use_global_stats: true
1729 |   }
1730 | }
1731 | layer {
1732 |   name: "scale_5_4_pw"
1733 |   type: "Scale"
1734 |   bottom: "conv_5_4_pw"
1735 |   top: "conv_5_4_pw"
1736 |   scale_param {
1737 |     bias_term: true
1738 |   }
1739 | }
1740 | layer {
1741 |   name: "relu_5_4_pw"
1742 |   type: "ReLU"
1743 |   bottom: "conv_5_4_pw"
1744 |   top: "conv_5_4_pw"
1745 | }
1746 | layer {
1747 |   name: "conv_5_4_dw"
1748 |   type: "Convolution"
1749 |   bottom: "conv_5_4_pw"
1750 |   top: "conv_5_4_dw"
1751 |   param {
1752 |     lr_mult: 1.0
1753 |     decay_mult: 1.0
1754 |   }
1755 |   convolution_param {
1756 |     num_output: 384
1757 |     bias_term: false
1758 |     pad: 1
1759 |     kernel_size: 3
1760 |     group: 384
1761 |     stride: 1
1762 |     weight_filler {
1763 |       type: "msra"
1764 |     }
1765 |     bias_filler {
1766 |       type: "constant"
1767 |       value: 0.0
1768 |     }
1769 |     engine: CAFFE
1770 |   }
1771 | }
1772 | layer {
1773 |   name: "bn_5_4_dw"
1774 |   type: "BatchNorm"
1775 |   bottom: "conv_5_4_dw"
1776 |   top: "conv_5_4_dw"
1777 |   param {
1778 |     lr_mult: 0.0
1779 |     decay_mult: 0.0
1780 |   }
1781 |   param {
1782 |     lr_mult: 0.0
1783 |     decay_mult: 0.0
1784 |   }
1785 |   param {
1786 |     lr_mult: 0.0
1787 |     decay_mult: 0.0
1788 |   }
1789 |   batch_norm_param {
1790 |     use_global_stats: true
1791 |   }
1792 | }
1793 | layer {
1794 |   name: "scale_5_4_dw"
1795 |   type: "Scale"
1796 |   bottom: "conv_5_4_dw"
1797 |   top: "conv_5_4_dw"
1798 |   scale_param {
1799 |     bias_term: true
1800 |   }
1801 | }
1802 | layer {
1803 |   name: "relu_5_4_dw"
1804 |   type: "ReLU"
1805 |   bottom: "conv_5_4_dw"
1806 |   top: "conv_5_4_dw"
1807 | }
1808 | layer {
1809 |   name: "conv_5_4_linear"
1810 |   type: "Convolution"
1811 |   bottom: "conv_5_4_dw"
1812 |   top: "conv_5_4_linear"
1813 |   param {
1814 |     lr_mult: 1.0
1815 |     decay_mult: 1.0
1816 |   }
1817 |   convolution_param {
1818 |     num_output: 64
1819 |     bias_term: false
1820 |     pad: 0
1821 |     kernel_size: 1
1822 |     stride: 1
1823 |     weight_filler {
1824 |       type: "msra"
1825 |     }
1826 |     bias_filler {
1827 |       type: "constant"
1828 |       value: 0.0
1829 |     }
1830 |   }
1831 | }
1832 | layer {
1833 |   name: "bn_5_4_linear"
1834 |   type: "BatchNorm"
1835 |   bottom: "conv_5_4_linear"
1836 |   top: "conv_5_4_linear"
1837 |   param {
1838 |     lr_mult: 0.0
1839 |     decay_mult: 0.0
1840 |   }
1841 |   param {
1842 |     lr_mult: 0.0
1843 |     decay_mult: 0.0
1844 |   }
1845 |   param {
1846 |     lr_mult: 0.0
1847 |     decay_mult: 0.0
1848 |   }
1849 |   batch_norm_param {
1850 |     use_global_stats: true
1851 |   }
1852 | }
1853 | layer {
1854 |   name: "scale_5_4_linear"
1855 |   type: "Scale"
1856 |   bottom: "conv_5_4_linear"
1857 |   top: "conv_5_4_linear"
1858 |   scale_param {
1859 |     bias_term: true
1860 |   }
1861 | }
1862 | layer {
1863 |   name: "add_5_4"
1864 |   type: "Eltwise"
1865 |   bottom: "add_5_3"
1866 |   bottom: "conv_5_4_linear"
1867 |   top: "add_5_4"
1868 | }
1869 | layer {
1870 |   name: "conv_6_1_pw"
1871 |   type: "Convolution"
1872 |   bottom: "add_5_4"
1873 |   top: "conv_6_1_pw"
1874 |   param {
1875 |     lr_mult: 1.0
1876 |     decay_mult: 1.0
1877 |   }
1878 |   convolution_param {
1879 |     num_output: 384
1880 |     bias_term: false
1881 |     pad: 0
1882 |     kernel_size: 1
1883 |     stride: 1
1884 |     weight_filler {
1885 |       type: "msra"
1886 |     }
1887 |     bias_filler {
1888 |       type: "constant"
1889 |       value: 0.0
1890 |     }
1891 |   }
1892 | }
1893 | layer {
1894 |   name: "bn_6_1_pw"
1895 |   type: "BatchNorm"
1896 |   bottom: "conv_6_1_pw"
1897 |   top: "conv_6_1_pw"
1898 |   param {
1899 |     lr_mult: 0.0
1900 |     decay_mult: 0.0
1901 |   }
1902 |   param {
1903 |     lr_mult: 0.0
1904 |     decay_mult: 0.0
1905 |   }
1906 |   param {
1907 |     lr_mult: 0.0
1908 |     decay_mult: 0.0
1909 |   }
1910 |   batch_norm_param {
1911 |     use_global_stats: true
1912 |   }
1913 | }
1914 | layer {
1915 |   name: "scale_6_1_pw"
1916 |   type: "Scale"
1917 |   bottom: "conv_6_1_pw"
1918 |   top: "conv_6_1_pw"
1919 |   scale_param {
1920 |     bias_term: true
1921 |   }
1922 | }
1923 | layer {
1924 |   name: "relu_6_1_pw"
1925 |   type: "ReLU"
1926 |   bottom: "conv_6_1_pw"
1927 |   top: "conv_6_1_pw"
1928 | }
1929 | layer {
1930 |   name: "conv_6_1_dw"
1931 |   type: "Convolution"
1932 |   bottom: "conv_6_1_pw"
1933 |   top: "conv_6_1_dw"
1934 |   param {
1935 |     lr_mult: 1.0
1936 |     decay_mult: 1.0
1937 |   }
1938 |   convolution_param {
1939 |     num_output: 384
1940 |     bias_term: false
1941 |     pad: 1
1942 |     kernel_size: 3
1943 |     group: 384
1944 |     stride: 1
1945 |     weight_filler {
1946 |       type: "msra"
1947 |     }
1948 |     bias_filler {
1949 |       type: "constant"
1950 |       value: 0.0
1951 |     }
1952 |     engine: CAFFE
1953 |   }
1954 | }
1955 | layer {
1956 |   name: "bn_6_1_dw"
1957 |   type: "BatchNorm"
1958 |   bottom: "conv_6_1_dw"
1959 |   top: "conv_6_1_dw"
1960 |   param {
1961 |     lr_mult: 0.0
1962 |     decay_mult: 0.0
1963 |   }
1964 |   param {
1965 |     lr_mult: 0.0
1966 |     decay_mult: 0.0
1967 |   }
1968 |   param {
1969 |     lr_mult: 0.0
1970 |     decay_mult: 0.0
1971 |   }
1972 |   batch_norm_param {
1973 |     use_global_stats: true
1974 |   }
1975 | }
1976 | layer {
1977 |   name: "scale_6_1_dw"
1978 |   type: "Scale"
1979 |   bottom: "conv_6_1_dw"
1980 |   top: "conv_6_1_dw"
1981 |   scale_param {
1982 |     bias_term: true
1983 |   }
1984 | }
1985 | layer {
1986 |   name: "relu_6_1_dw"
1987 |   type: "ReLU"
1988 |   bottom: "conv_6_1_dw"
1989 |   top: "conv_6_1_dw"
1990 | }
1991 | layer {
1992 |   name: "conv_6_1_linear"
1993 |   type: "Convolution"
1994 |   bottom: "conv_6_1_dw"
1995 |   top: "conv_6_1_linear"
1996 |   param {
1997 |     lr_mult: 1.0
1998 |     decay_mult: 1.0
1999 |   }
2000 |   convolution_param {
2001 |     num_output: 96
2002 |     bias_term: false
2003 |     pad: 0
2004 |     kernel_size: 1
2005 |     stride: 1
2006 |     weight_filler {
2007 |       type: "msra"
2008 |     }
2009 |     bias_filler {
2010 |       type: "constant"
2011 |       value: 0.0
2012 |     }
2013 |   }
2014 | }
2015 | layer {
2016 |   name: "bn_6_1_linear"
2017 |   type: "BatchNorm"
2018 |   bottom: "conv_6_1_linear"
2019 |   top: "conv_6_1_linear"
2020 |   param {
2021 |     lr_mult: 0.0
2022 |     decay_mult: 0.0
2023 |   }
2024 |   param {
2025 |     lr_mult: 0.0
2026 |     decay_mult: 0.0
2027 |   }
2028 |   param {
2029 |     lr_mult: 0.0
2030 |     decay_mult: 0.0
2031 |   }
2032 |   batch_norm_param {
2033 |     use_global_stats: true
2034 |   }
2035 | }
2036 | layer {
2037 |   name: "scale_6_1_linear"
2038 |   type: "Scale"
2039 |   bottom: "conv_6_1_linear"
2040 |   top: "conv_6_1_linear"
2041 |   scale_param {
2042 |     bias_term: true
2043 |   }
2044 | }
2045 | layer {
2046 |   name: "conv_6_2_pw"
2047 |   type: "Convolution"
2048 |   bottom: "conv_6_1_linear"
2049 |   top: "conv_6_2_pw"
2050 |   param {
2051 |     lr_mult: 1.0
2052 |     decay_mult: 1.0
2053 |   }
2054 |   convolution_param {
2055 |     num_output: 576
2056 |     bias_term: false
2057 |     pad: 0
2058 |     kernel_size: 1
2059 |     stride: 1
2060 |     weight_filler {
2061 |       type: "msra"
2062 |     }
2063 |     bias_filler {
2064 |       type: "constant"
2065 |       value: 0.0
2066 |     }
2067 |   }
2068 | }
2069 | layer {
2070 |   name: "bn_6_2_pw"
2071 |   type: "BatchNorm"
2072 |   bottom: "conv_6_2_pw"
2073 |   top: "conv_6_2_pw"
2074 |   param {
2075 |     lr_mult: 0.0
2076 |     decay_mult: 0.0
2077 |   }
2078 |   param {
2079 |     lr_mult: 0.0
2080 |     decay_mult: 0.0
2081 |   }
2082 |   param {
2083 |     lr_mult: 0.0
2084 |     decay_mult: 0.0
2085 |   }
2086 |   batch_norm_param {
2087 |     use_global_stats: true
2088 |   }
2089 | }
2090 | layer {
2091 |   name: "scale_6_2_pw"
2092 |   type: "Scale"
2093 |   bottom: "conv_6_2_pw"
2094 |   top: "conv_6_2_pw"
2095 |   scale_param {
2096 |     bias_term: true
2097 |   }
2098 | }
2099 | layer {
2100 |   name: "relu_6_2_pw"
2101 |   type: "ReLU"
2102 |   bottom: "conv_6_2_pw"
2103 |   top: "conv_6_2_pw"
2104 | }
2105 | layer {
2106 |   name: "conv_6_2_dw"
2107 |   type: "Convolution"
2108 |   bottom: "conv_6_2_pw"
2109 |   top: "conv_6_2_dw"
2110 |   param {
2111 |     lr_mult: 1.0
2112 |     decay_mult: 1.0
2113 |   }
2114 |   convolution_param {
2115 |     num_output: 576
2116 |     bias_term: false
2117 |     pad: 1
2118 |     kernel_size: 3
2119 |     group: 576
2120 |     stride: 1
2121 |     weight_filler {
2122 |       type: "msra"
2123 |     }
2124 |     bias_filler {
2125 |       type: "constant"
2126 |       value: 0.0
2127 |     }
2128 |     engine: CAFFE
2129 |   }
2130 | }
2131 | layer {
2132 |   name: "bn_6_2_dw"
2133 |   type: "BatchNorm"
2134 |   bottom: "conv_6_2_dw"
2135 |   top: "conv_6_2_dw"
2136 |   param {
2137 |     lr_mult: 0.0
2138 |     decay_mult: 0.0
2139 |   }
2140 |   param {
2141 |     lr_mult: 0.0
2142 |     decay_mult: 0.0
2143 |   }
2144 |   param {
2145 |     lr_mult: 0.0
2146 |     decay_mult: 0.0
2147 |   }
2148 |   batch_norm_param {
2149 |     use_global_stats: true
2150 |   }
2151 | }
2152 | layer {
2153 |   name: "scale_6_2_dw"
2154 |   type: "Scale"
2155 |   bottom: "conv_6_2_dw"
2156 |   top: "conv_6_2_dw"
2157 |   scale_param {
2158 |     bias_term: true
2159 |   }
2160 | }
2161 | layer {
2162 |   name: "relu_6_2_dw"
2163 |   type: "ReLU"
2164 |   bottom: "conv_6_2_dw"
2165 |   top: "conv_6_2_dw"
2166 | }
2167 | layer {
2168 |   name: "conv_6_2_linear"
2169 |   type: "Convolution"
2170 |   bottom: "conv_6_2_dw"
2171 |   top: "conv_6_2_linear"
2172 |   param {
2173 |     lr_mult: 1.0
2174 |     decay_mult: 1.0
2175 |   }
2176 |   convolution_param {
2177 |     num_output: 96
2178 |     bias_term: false
2179 |     pad: 0
2180 |     kernel_size: 1
2181 |     stride: 1
2182 |     weight_filler {
2183 |       type: "msra"
2184 |     }
2185 |     bias_filler {
2186 |       type: "constant"
2187 |       value: 0.0
2188 |     }
2189 |   }
2190 | }
2191 | layer {
2192 |   name: "bn_6_2_linear"
2193 |   type: "BatchNorm"
2194 |   bottom: "conv_6_2_linear"
2195 |   top: "conv_6_2_linear"
2196 |   param {
2197 |     lr_mult: 0.0
2198 |     decay_mult: 0.0
2199 |   }
2200 |   param {
2201 |     lr_mult: 0.0
2202 |     decay_mult: 0.0
2203 |   }
2204 |   param {
2205 |     lr_mult: 0.0
2206 |     decay_mult: 0.0
2207 |   }
2208 |   batch_norm_param {
2209 |     use_global_stats: true
2210 |   }
2211 | }
2212 | layer {
2213 |   name: "scale_6_2_linear"
2214 |   type: "Scale"
2215 |   bottom: "conv_6_2_linear"
2216 |   top: "conv_6_2_linear"
2217 |   scale_param {
2218 |     bias_term: true
2219 |   }
2220 | }
2221 | layer {
2222 |   name: "add_6_2"
2223 |   type: "Eltwise"
2224 |   bottom: "conv_6_1_linear"
2225 |   bottom: "conv_6_2_linear"
2226 |   top: "add_6_2"
2227 | }
2228 | layer {
2229 |   name: "conv_6_3_pw"
2230 |   type: "Convolution"
2231 |   bottom: "add_6_2"
2232 |   top: "conv_6_3_pw"
2233 |   param {
2234 |     lr_mult: 1.0
2235 |     decay_mult: 1.0
2236 |   }
2237 |   convolution_param {
2238 |     num_output: 576
2239 |     bias_term: false
2240 |     pad: 0
2241 |     kernel_size: 1
2242 |     stride: 1
2243 |     weight_filler {
2244 |       type: "msra"
2245 |     }
2246 |     bias_filler {
2247 |       type: "constant"
2248 |       value: 0.0
2249 |     }
2250 |   }
2251 | }
2252 | layer {
2253 |   name: "bn_6_3_pw"
2254 |   type: "BatchNorm"
2255 |   bottom: "conv_6_3_pw"
2256 |   top: "conv_6_3_pw"
2257 |   param {
2258 |     lr_mult: 0.0
2259 |     decay_mult: 0.0
2260 |   }
2261 |   param {
2262 |     lr_mult: 0.0
2263 |     decay_mult: 0.0
2264 |   }
2265 |   param {
2266 |     lr_mult: 0.0
2267 |     decay_mult: 0.0
2268 |   }
2269 |   batch_norm_param {
2270 |     use_global_stats: true
2271 |   }
2272 | }
2273 | layer {
2274 |   name: "scale_6_3_pw"
2275 |   type: "Scale"
2276 |   bottom: "conv_6_3_pw"
2277 |   top: "conv_6_3_pw"
2278 |   scale_param {
2279 |     bias_term: true
2280 |   }
2281 | }
2282 | layer {
2283 |   name: "relu_6_3_pw"
2284 |   type: "ReLU"
2285 |   bottom: "conv_6_3_pw"
2286 |   top: "conv_6_3_pw"
2287 | }
2288 | layer {
2289 |   name: "conv_6_3_dw"
2290 |   type: "Convolution"
2291 |   bottom: "conv_6_3_pw"
2292 |   top: "conv_6_3_dw"
2293 |   param {
2294 |     lr_mult: 1.0
2295 |     decay_mult: 1.0
2296 |   }
2297 |   convolution_param {
2298 |     num_output: 576
2299 |     bias_term: false
2300 |     pad: 1
2301 |     kernel_size: 3
2302 |     group: 576
2303 |     stride: 1
2304 |     weight_filler {
2305 |       type: "msra"
2306 |     }
2307 |     bias_filler {
2308 |       type: "constant"
2309 |       value: 0.0
2310 |     }
2311 |     engine: CAFFE
2312 |   }
2313 | }
2314 | layer {
2315 |   name: "bn_6_3_dw"
2316 |   type: "BatchNorm"
2317 |   bottom: "conv_6_3_dw"
2318 |   top: "conv_6_3_dw"
2319 |   param {
2320 |     lr_mult: 0.0
2321 |     decay_mult: 0.0
2322 |   }
2323 |   param {
2324 |     lr_mult: 0.0
2325 |     decay_mult: 0.0
2326 |   }
2327 |   param {
2328 |     lr_mult: 0.0
2329 |     decay_mult: 0.0
2330 |   }
2331 |   batch_norm_param {
2332 |     use_global_stats: true
2333 |   }
2334 | }
2335 | layer {
2336 |   name: "scale_6_3_dw"
2337 |   type: "Scale"
2338 |   bottom: "conv_6_3_dw"
2339 |   top: "conv_6_3_dw"
2340 |   scale_param {
2341 |     bias_term: true
2342 |   }
2343 | }
2344 | layer {
2345 |   name: "relu_6_3_dw"
2346 |   type: "ReLU"
2347 |   bottom: "conv_6_3_dw"
2348 |   top: "conv_6_3_dw"
2349 | }
2350 | layer {
2351 |   name: "conv_6_3_linear"
2352 |   type: "Convolution"
2353 |   bottom: "conv_6_3_dw"
2354 |   top: "conv_6_3_linear"
2355 |   param {
2356 |     lr_mult: 1.0
2357 |     decay_mult: 1.0
2358 |   }
2359 |   convolution_param {
2360 |     num_output: 96
2361 |     bias_term: false
2362 |     pad: 0
2363 |     kernel_size: 1
2364 |     stride: 1
2365 |     weight_filler {
2366 |       type: "msra"
2367 |     }
2368 |     bias_filler {
2369 |       type: "constant"
2370 |       value: 0.0
2371 |     }
2372 |   }
2373 | }
2374 | layer {
2375 |   name: "bn_6_3_linear"
2376 |   type: "BatchNorm"
2377 |   bottom: "conv_6_3_linear"
2378 |   top: "conv_6_3_linear"
2379 |   param {
2380 |     lr_mult: 0.0
2381 |     decay_mult: 0.0
2382 |   }
2383 |   param {
2384 |     lr_mult: 0.0
2385 |     decay_mult: 0.0
2386 |   }
2387 |   param {
2388 |     lr_mult: 0.0
2389 |     decay_mult: 0.0
2390 |   }
2391 |   batch_norm_param {
2392 |     use_global_stats: true
2393 |   }
2394 | }
2395 | layer {
2396 |   name: "scale_6_3_linear"
2397 |   type: "Scale"
2398 |   bottom: "conv_6_3_linear"
2399 |   top: "conv_6_3_linear"
2400 |   scale_param {
2401 |     bias_term: true
2402 |   }
2403 | }
2404 | layer {
2405 |   name: "add_6_3"
2406 |   type: "Eltwise"
2407 |   bottom: "add_6_2"
2408 |   bottom: "conv_6_3_linear"
2409 |   top: "add_6_3"
2410 | }
2411 | layer {
2412 |   name: "conv_7_1_pw"
2413 |   type: "Convolution"
2414 |   bottom: "add_6_3"
2415 |   top: "conv_7_1_pw"
2416 |   param {
2417 |     lr_mult: 1.0
2418 |     decay_mult: 1.0
2419 |   }
2420 |   convolution_param {
2421 |     num_output: 576
2422 |     bias_term: false
2423 |     pad: 0
2424 |     kernel_size: 1
2425 |     stride: 1
2426 |     weight_filler {
2427 |       type: "msra"
2428 |     }
2429 |     bias_filler {
2430 |       type: "constant"
2431 |       value: 0.0
2432 |     }
2433 |   }
2434 | }
2435 | layer {
2436 |   name: "bn_7_1_pw"
2437 |   type: "BatchNorm"
2438 |   bottom: "conv_7_1_pw"
2439 |   top: "conv_7_1_pw"
2440 |   param {
2441 |     lr_mult: 0.0
2442 |     decay_mult: 0.0
2443 |   }
2444 |   param {
2445 |     lr_mult: 0.0
2446 |     decay_mult: 0.0
2447 |   }
2448 |   param {
2449 |     lr_mult: 0.0
2450 |     decay_mult: 0.0
2451 |   }
2452 |   batch_norm_param {
2453 |     use_global_stats: true
2454 |   }
2455 | }
2456 | layer {
2457 |   name: "scale_7_1_pw"
2458 |   type: "Scale"
2459 |   bottom: "conv_7_1_pw"
2460 |   top: "conv_7_1_pw"
2461 |   scale_param {
2462 |     bias_term: true
2463 |   }
2464 | }
2465 | layer {
2466 |   name: "relu_7_1_pw"
2467 |   type: "ReLU"
2468 |   bottom: "conv_7_1_pw"
2469 |   top: "conv_7_1_pw"
2470 | }
2471 | layer {
2472 |   name: "conv_7_1_dw"
2473 |   type: "Convolution"
2474 |   bottom: "conv_7_1_pw"
2475 |   top: "conv_7_1_dw"
2476 |   param {
2477 |     lr_mult: 1.0
2478 |     decay_mult: 1.0
2479 |   }
2480 |   convolution_param {
2481 |     num_output: 576
2482 |     bias_term: false
2483 |     pad: 1
2484 |     kernel_size: 3
2485 |     group: 576
2486 |     stride: 2
2487 |     weight_filler {
2488 |       type: "msra"
2489 |     }
2490 |     bias_filler {
2491 |       type: "constant"
2492 |       value: 0.0
2493 |     }
2494 |     engine: CAFFE
2495 |   }
2496 | }
2497 | layer {
2498 |   name: "bn_7_1_dw"
2499 |   type: "BatchNorm"
2500 |   bottom: "conv_7_1_dw"
2501 |   top: "conv_7_1_dw"
2502 |   param {
2503 |     lr_mult: 0.0
2504 |     decay_mult: 0.0
2505 |   }
2506 |   param {
2507 |     lr_mult: 0.0
2508 |     decay_mult: 0.0
2509 |   }
2510 |   param {
2511 |     lr_mult: 0.0
2512 |     decay_mult: 0.0
2513 |   }
2514 |   batch_norm_param {
2515 |     use_global_stats: true
2516 |   }
2517 | }
2518 | layer {
2519 |   name: "scale_7_1_dw"
2520 |   type: "Scale"
2521 |   bottom: "conv_7_1_dw"
2522 |   top: "conv_7_1_dw"
2523 |   scale_param {
2524 |     bias_term: true
2525 |   }
2526 | }
2527 | layer {
2528 |   name: "relu_7_1_dw"
2529 |   type: "ReLU"
2530 |   bottom: "conv_7_1_dw"
2531 |   top: "conv_7_1_dw"
2532 | }
2533 | layer {
2534 |   name: "conv_7_1_linear"
2535 |   type: "Convolution"
2536 |   bottom: "conv_7_1_dw"
2537 |   top: "conv_7_1_linear"
2538 |   param {
2539 |     lr_mult: 1.0
2540 |     decay_mult: 1.0
2541 |   }
2542 |   convolution_param {
2543 |     num_output: 160
2544 |     bias_term: false
2545 |     pad: 0
2546 |     kernel_size: 1
2547 |     stride: 1
2548 |     weight_filler {
2549 |       type: "msra"
2550 |     }
2551 |     bias_filler {
2552 |       type: "constant"
2553 |       value: 0.0
2554 |     }
2555 |   }
2556 | }
2557 | layer {
2558 |   name: "bn_7_1_linear"
2559 |   type: "BatchNorm"
2560 |   bottom: "conv_7_1_linear"
2561 |   top: "conv_7_1_linear"
2562 |   param {
2563 |     lr_mult: 0.0
2564 |     decay_mult: 0.0
2565 |   }
2566 |   param {
2567 |     lr_mult: 0.0
2568 |     decay_mult: 0.0
2569 |   }
2570 |   param {
2571 |     lr_mult: 0.0
2572 |     decay_mult: 0.0
2573 |   }
2574 |   batch_norm_param {
2575 |     use_global_stats: true
2576 |   }
2577 | }
2578 | layer {
2579 |   name: "scale_7_1_linear"
2580 |   type: "Scale"
2581 |   bottom: "conv_7_1_linear"
2582 |   top: "conv_7_1_linear"
2583 |   scale_param {
2584 |     bias_term: true
2585 |   }
2586 | }
2587 | layer {
2588 |   name: "conv_7_2_pw"
2589 |   type: "Convolution"
2590 |   bottom: "conv_7_1_linear"
2591 |   top: "conv_7_2_pw"
2592 |   param {
2593 |     lr_mult: 1.0
2594 |     decay_mult: 1.0
2595 |   }
2596 |   convolution_param {
2597 |     num_output: 960
2598 |     bias_term: false
2599 |     pad: 0
2600 |     kernel_size: 1
2601 |     stride: 1
2602 |     weight_filler {
2603 |       type: "msra"
2604 |     }
2605 |     bias_filler {
2606 |       type: "constant"
2607 |       value: 0.0
2608 |     }
2609 |   }
2610 | }
2611 | layer {
2612 |   name: "bn_7_2_pw"
2613 |   type: "BatchNorm"
2614 |   bottom: "conv_7_2_pw"
2615 |   top: "conv_7_2_pw"
2616 |   param {
2617 |     lr_mult: 0.0
2618 |     decay_mult: 0.0
2619 |   }
2620 |   param {
2621 |     lr_mult: 0.0
2622 |     decay_mult: 0.0
2623 |   }
2624 |   param {
2625 |     lr_mult: 0.0
2626 |     decay_mult: 0.0
2627 |   }
2628 |   batch_norm_param {
2629 |     use_global_stats: true
2630 |   }
2631 | }
2632 | layer {
2633 |   name: "scale_7_2_pw"
2634 |   type: "Scale"
2635 |   bottom: "conv_7_2_pw"
2636 |   top: "conv_7_2_pw"
2637 |   scale_param {
2638 |     bias_term: true
2639 |   }
2640 | }
2641 | layer {
2642 |   name: "relu_7_2_pw"
2643 |   type: "ReLU"
2644 |   bottom: "conv_7_2_pw"
2645 |   top: "conv_7_2_pw"
2646 | }
2647 | layer {
2648 |   name: "conv_7_2_dw"
2649 |   type: "Convolution"
2650 |   bottom: "conv_7_2_pw"
2651 |   top: "conv_7_2_dw"
2652 |   param {
2653 |     lr_mult: 1.0
2654 |     decay_mult: 1.0
2655 |   }
2656 |   convolution_param {
2657 |     num_output: 960
2658 |     bias_term: false
2659 |     pad: 1
2660 |     kernel_size: 3
2661 |     group: 960
2662 |     stride: 1
2663 |     weight_filler {
2664 |       type: "msra"
2665 |     }
2666 |     bias_filler {
2667 |       type: "constant"
2668 |       value: 0.0
2669 |     }
2670 |     engine: CAFFE
2671 |   }
2672 | }
2673 | layer {
2674 |   name: "bn_7_2_dw"
2675 |   type: "BatchNorm"
2676 |   bottom: "conv_7_2_dw"
2677 |   top: "conv_7_2_dw"
2678 |   param {
2679 |     lr_mult: 0.0
2680 |     decay_mult: 0.0
2681 |   }
2682 |   param {
2683 |     lr_mult: 0.0
2684 |     decay_mult: 0.0
2685 |   }
2686 |   param {
2687 |     lr_mult: 0.0
2688 |     decay_mult: 0.0
2689 |   }
2690 |   batch_norm_param {
2691 |     use_global_stats: true
2692 |   }
2693 | }
2694 | layer {
2695 |   name: "scale_7_2_dw"
2696 |   type: "Scale"
2697 |   bottom: "conv_7_2_dw"
2698 |   top: "conv_7_2_dw"
2699 |   scale_param {
2700 |     bias_term: true
2701 |   }
2702 | }
2703 | layer {
2704 |   name: "relu_7_2_dw"
2705 |   type: "ReLU"
2706 |   bottom: "conv_7_2_dw"
2707 |   top: "conv_7_2_dw"
2708 | }
2709 | layer {
2710 |   name: "conv_7_2_linear"
2711 |   type: "Convolution"
2712 |   bottom: "conv_7_2_dw"
2713 |   top: "conv_7_2_linear"
2714 |   param {
2715 |     lr_mult: 1.0
2716 |     decay_mult: 1.0
2717 |   }
2718 |   convolution_param {
2719 |     num_output: 160
2720 |     bias_term: false
2721 |     pad: 0
2722 |     kernel_size: 1
2723 |     stride: 1
2724 |     weight_filler {
2725 |       type: "msra"
2726 |     }
2727 |     bias_filler {
2728 |       type: "constant"
2729 |       value: 0.0
2730 |     }
2731 |   }
2732 | }
2733 | layer {
2734 |   name: "bn_7_2_linear"
2735 |   type: "BatchNorm"
2736 |   bottom: "conv_7_2_linear"
2737 |   top: "conv_7_2_linear"
2738 |   param {
2739 |     lr_mult: 0.0
2740 |     decay_mult: 0.0
2741 |   }
2742 |   param {
2743 |     lr_mult: 0.0
2744 |     decay_mult: 0.0
2745 |   }
2746 |   param {
2747 |     lr_mult: 0.0
2748 |     decay_mult: 0.0
2749 |   }
2750 |   batch_norm_param {
2751 |     use_global_stats: true
2752 |   }
2753 | }
2754 | layer {
2755 |   name: "scale_7_2_linear"
2756 |   type: "Scale"
2757 |   bottom: "conv_7_2_linear"
2758 |   top: "conv_7_2_linear"
2759 |   scale_param {
2760 |     bias_term: true
2761 |   }
2762 | }
2763 | layer {
2764 |   name: "add_7_2"
2765 |   type: "Eltwise"
2766 |   bottom: "conv_7_1_linear"
2767 |   bottom: "conv_7_2_linear"
2768 |   top: "add_7_2"
2769 | }
2770 | layer {
2771 |   name: "conv_7_3_pw"
2772 |   type: "Convolution"
2773 |   bottom: "add_7_2"
2774 |   top: "conv_7_3_pw"
2775 |   param {
2776 |     lr_mult: 1.0
2777 |     decay_mult: 1.0
2778 |   }
2779 |   convolution_param {
2780 |     num_output: 960
2781 |     bias_term: false
2782 |     pad: 0
2783 |     kernel_size: 1
2784 |     stride: 1
2785 |     weight_filler {
2786 |       type: "msra"
2787 |     }
2788 |     bias_filler {
2789 |       type: "constant"
2790 |       value: 0.0
2791 |     }
2792 |   }
2793 | }
2794 | layer {
2795 |   name: "bn_7_3_pw"
2796 |   type: "BatchNorm"
2797 |   bottom: "conv_7_3_pw"
2798 |   top: "conv_7_3_pw"
2799 |   param {
2800 |     lr_mult: 0.0
2801 |     decay_mult: 0.0
2802 |   }
2803 |   param {
2804 |     lr_mult: 0.0
2805 |     decay_mult: 0.0
2806 |   }
2807 |   param {
2808 |     lr_mult: 0.0
2809 |     decay_mult: 0.0
2810 |   }
2811 |   batch_norm_param {
2812 |     use_global_stats: true
2813 |   }
2814 | }
2815 | layer {
2816 |   name: "scale_7_3_pw"
2817 |   type: "Scale"
2818 |   bottom: "conv_7_3_pw"
2819 |   top: "conv_7_3_pw"
2820 |   scale_param {
2821 |     bias_term: true
2822 |   }
2823 | }
2824 | layer {
2825 |   name: "relu_7_3_pw"
2826 |   type: "ReLU"
2827 |   bottom: "conv_7_3_pw"
2828 |   top: "conv_7_3_pw"
2829 | }
2830 | layer {
2831 |   name: "conv_7_3_dw"
2832 |   type: "Convolution"
2833 |   bottom: "conv_7_3_pw"
2834 |   top: "conv_7_3_dw"
2835 |   param {
2836 |     lr_mult: 1.0
2837 |     decay_mult: 1.0
2838 |   }
2839 |   convolution_param {
2840 |     num_output: 960
2841 |     bias_term: false
2842 |     pad: 1
2843 |     kernel_size: 3
2844 |     group: 960
2845 |     stride: 1
2846 |     weight_filler {
2847 |       type: "msra"
2848 |     }
2849 |     bias_filler {
2850 |       type: "constant"
2851 |       value: 0.0
2852 |     }
2853 |     engine: CAFFE
2854 |   }
2855 | }
2856 | layer {
2857 |   name: "bn_7_3_dw"
2858 |   type: "BatchNorm"
2859 |   bottom: "conv_7_3_dw"
2860 |   top: "conv_7_3_dw"
2861 |   param {
2862 |     lr_mult: 0.0
2863 |     decay_mult: 0.0
2864 |   }
2865 |   param {
2866 |     lr_mult: 0.0
2867 |     decay_mult: 0.0
2868 |   }
2869 |   param {
2870 |     lr_mult: 0.0
2871 |     decay_mult: 0.0
2872 |   }
2873 |   batch_norm_param {
2874 |     use_global_stats: true
2875 |   }
2876 | }
2877 | layer {
2878 |   name: "scale_7_3_dw"
2879 |   type: "Scale"
2880 |   bottom: "conv_7_3_dw"
2881 |   top: "conv_7_3_dw"
2882 |   scale_param {
2883 |     bias_term: true
2884 |   }
2885 | }
2886 | layer {
2887 |   name: "relu_7_3_dw"
2888 |   type: "ReLU"
2889 |   bottom: "conv_7_3_dw"
2890 |   top: "conv_7_3_dw"
2891 | }
2892 | layer {
2893 |   name: "conv_7_3_linear"
2894 |   type: "Convolution"
2895 |   bottom: "conv_7_3_dw"
2896 |   top: "conv_7_3_linear"
2897 |   param {
2898 |     lr_mult: 1.0
2899 |     decay_mult: 1.0
2900 |   }
2901 |   convolution_param {
2902 |     num_output: 160
2903 |     bias_term: false
2904 |     pad: 0
2905 |     kernel_size: 1
2906 |     stride: 1
2907 |     weight_filler {
2908 |       type: "msra"
2909 |     }
2910 |     bias_filler {
2911 |       type: "constant"
2912 |       value: 0.0
2913 |     }
2914 |   }
2915 | }
2916 | layer {
2917 |   name: "bn_7_3_linear"
2918 |   type: "BatchNorm"
2919 |   bottom: "conv_7_3_linear"
2920 |   top: "conv_7_3_linear"
2921 |   param {
2922 |     lr_mult: 0.0
2923 |     decay_mult: 0.0
2924 |   }
2925 |   param {
2926 |     lr_mult: 0.0
2927 |     decay_mult: 0.0
2928 |   }
2929 |   param {
2930 |     lr_mult: 0.0
2931 |     decay_mult: 0.0
2932 |   }
2933 |   batch_norm_param {
2934 |     use_global_stats: true
2935 |   }
2936 | }
2937 | layer {
2938 |   name: "scale_7_3_linear"
2939 |   type: "Scale"
2940 |   bottom: "conv_7_3_linear"
2941 |   top: "conv_7_3_linear"
2942 |   scale_param {
2943 |     bias_term: true
2944 |   }
2945 | }
2946 | layer {
2947 |   name: "add_7_3"
2948 |   type: "Eltwise"
2949 |   bottom: "add_7_2"
2950 |   bottom: "conv_7_3_linear"
2951 |   top: "add_7_3"
2952 | }
2953 | layer {
2954 |   name: "conv_8_1_pw"
2955 |   type: "Convolution"
2956 |   bottom: "add_7_3"
2957 |   top: "conv_8_1_pw"
2958 |   param {
2959 |     lr_mult: 1.0
2960 |     decay_mult: 1.0
2961 |   }
2962 |   convolution_param {
2963 |     num_output: 960
2964 |     bias_term: false
2965 |     pad: 0
2966 |     kernel_size: 1
2967 |     stride: 1
2968 |     weight_filler {
2969 |       type: "msra"
2970 |     }
2971 |     bias_filler {
2972 |       type: "constant"
2973 |       value: 0.0
2974 |     }
2975 |   }
2976 | }
2977 | layer {
2978 |   name: "bn_8_1_pw"
2979 |   type: "BatchNorm"
2980 |   bottom: "conv_8_1_pw"
2981 |   top: "conv_8_1_pw"
2982 |   param {
2983 |     lr_mult: 0.0
2984 |     decay_mult: 0.0
2985 |   }
2986 |   param {
2987 |     lr_mult: 0.0
2988 |     decay_mult: 0.0
2989 |   }
2990 |   param {
2991 |     lr_mult: 0.0
2992 |     decay_mult: 0.0
2993 |   }
2994 |   batch_norm_param {
2995 |     use_global_stats: true
2996 |   }
2997 | }
2998 | layer {
2999 |   name: "scale_8_1_pw"
3000 |   type: "Scale"
3001 |   bottom: "conv_8_1_pw"
3002 |   top: "conv_8_1_pw"
3003 |   scale_param {
3004 |     bias_term: true
3005 |   }
3006 | }
3007 | layer {
3008 |   name: "relu_8_1_pw"
3009 |   type: "ReLU"
3010 |   bottom: "conv_8_1_pw"
3011 |   top: "conv_8_1_pw"
3012 | }
3013 | layer {
3014 |   name: "conv_8_1_dw"
3015 |   type: "Convolution"
3016 |   bottom: "conv_8_1_pw"
3017 |   top: "conv_8_1_dw"
3018 |   param {
3019 |     lr_mult: 1.0
3020 |     decay_mult: 1.0
3021 |   }
3022 |   convolution_param {
3023 |     num_output: 960
3024 |     bias_term: false
3025 |     pad: 1
3026 |     kernel_size: 3
3027 |     group: 960
3028 |     stride: 1
3029 |     weight_filler {
3030 |       type: "msra"
3031 |     }
3032 |     bias_filler {
3033 |       type: "constant"
3034 |       value: 0.0
3035 |     }
3036 |     engine: CAFFE
3037 |   }
3038 | }
3039 | layer {
3040 |   name: "bn_8_1_dw"
3041 |   type: "BatchNorm"
3042 |   bottom: "conv_8_1_dw"
3043 |   top: "conv_8_1_dw"
3044 |   param {
3045 |     lr_mult: 0.0
3046 |     decay_mult: 0.0
3047 |   }
3048 |   param {
3049 |     lr_mult: 0.0
3050 |     decay_mult: 0.0
3051 |   }
3052 |   param {
3053 |     lr_mult: 0.0
3054 |     decay_mult: 0.0
3055 |   }
3056 |   batch_norm_param {
3057 |     use_global_stats: true
3058 |   }
3059 | }
3060 | layer {
3061 |   name: "scale_8_1_dw"
3062 |   type: "Scale"
3063 |   bottom: "conv_8_1_dw"
3064 |   top: "conv_8_1_dw"
3065 |   scale_param {
3066 |     bias_term: true
3067 |   }
3068 | }
3069 | layer {
3070 |   name: "relu_8_1_dw"
3071 |   type: "ReLU"
3072 |   bottom: "conv_8_1_dw"
3073 |   top: "conv_8_1_dw"
3074 | }
3075 | layer {
3076 |   name: "conv_8_1_linear"
3077 |   type: "Convolution"
3078 |   bottom: "conv_8_1_dw"
3079 |   top: "conv_8_1_linear"
3080 |   param {
3081 |     lr_mult: 1.0
3082 |     decay_mult: 1.0
3083 |   }
3084 |   convolution_param {
3085 |     num_output: 320
3086 |     bias_term: false
3087 |     pad: 0
3088 |     kernel_size: 1
3089 |     stride: 1
3090 |     weight_filler {
3091 |       type: "msra"
3092 |     }
3093 |     bias_filler {
3094 |       type: "constant"
3095 |       value: 0.0
3096 |     }
3097 |   }
3098 | }
3099 | layer {
3100 |   name: "bn_8_1_linear"
3101 |   type: "BatchNorm"
3102 |   bottom: "conv_8_1_linear"
3103 |   top: "conv_8_1_linear"
3104 |   param {
3105 |     lr_mult: 0.0
3106 |     decay_mult: 0.0
3107 |   }
3108 |   param {
3109 |     lr_mult: 0.0
3110 |     decay_mult: 0.0
3111 |   }
3112 |   param {
3113 |     lr_mult: 0.0
3114 |     decay_mult: 0.0
3115 |   }
3116 |   batch_norm_param {
3117 |     use_global_stats: true
3118 |   }
3119 | }
3120 | layer {
3121 |   name: "scale_8_1_linear"
3122 |   type: "Scale"
3123 |   bottom: "conv_8_1_linear"
3124 |   top: "conv_8_1_linear"
3125 |   scale_param {
3126 |     bias_term: true
3127 |   }
3128 | }
3129 | layer {
3130 |   name: "conv9"
3131 |   type: "Convolution"
3132 |   bottom: "conv_8_1_linear"
3133 |   top: "conv9"
3134 |   param {
3135 |     lr_mult: 1.0
3136 |     decay_mult: 1.0
3137 |   }
3138 |   convolution_param {
3139 |     num_output: 1280
3140 |     pad: 0
3141 |     kernel_size: 1
3142 |     stride: 1
3143 |     weight_filler {
3144 |       type: "msra"
3145 |     }
3146 |     bias_filler {
3147 |       type: "constant"
3148 |       value: 0.0
3149 |     }
3150 |   }
3151 | }
3152 | layer {
3153 |   name: "pool9"
3154 |   type: "Pooling"
3155 |   bottom: "conv9"
3156 |   top: "pool9"
3157 |   pooling_param {
3158 |     pool: AVE
3159 |     stride: 1
3160 |     global_pooling: true
3161 |   }
3162 | }
3163 | layer {
3164 |   name: "dropout"
3165 |   type: "Dropout"
3166 |   bottom: "pool9"
3167 |   top: "pool9"
3168 |   dropout_param {
3169 |     dropout_ratio : 0.2
3170 |   }
3171 | }
3172 | layer {
3173 |   name: "conv10"
3174 |   type: "Convolution"
3175 |   bottom: "pool9"
3176 |   top: "conv10"
3177 |   param {
3178 |     lr_mult: 1.0
3179 |     decay_mult: 1.0
3180 |   }
3181 |   convolution_param {
3182 |     num_output: 1000
3183 |     pad: 0
3184 |     kernel_size: 1
3185 |     stride: 1
3186 |     weight_filler {
3187 |       type: "msra"
3188 |     }
3189 |     bias_filler {
3190 |       type: "constant"
3191 |       value: 0.0
3192 |     }
3193 |   }
3194 | }
3195 | layer {
3196 |   name: "loss"
3197 |   type: "Softmax"
3198 |   bottom: "conv10"
3199 |   top: "loss"
3200 | }
3201 | 
3202 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # MobileNet-v2-caffe
 2 | MobileNet-v2 experimental network description for caffe.
 3 | 
 4 | ### Update 2018-08-18
 5 | 1. Add other Mobilenet-v2 variants
 6 | 2. Suggestion: cudnn v7 has supported depthwise 3x3 when `group == input_channel`, you may speed up your training process by using the latest cudnn v7. 
 7 | 
 8 | ### Update
 9 | 1. Google has released a series of mobilenet-v2 models. So reference pretrained model from tensorflow/model repository.
10 | 2. MobileNet-V2 has accepted by CVPR 2018. The latest ilsvrc12 top1 accuracy is **72.0%**. 
11 | 3. According to google official model, mobilenet-v2 downsampled feature map early.
12 | 4. shortcuts are placed except the first inverted residual bottleneck sequence.
13 | 
14 | ### Note
15 | There are some unclear details about the network architechture.
16 | 1. ~~bottleneck sequence 5's input size doesn't match its prior sequence's stride.~~
17 | 2. ~~how to deal with shortcuts or residual when the input channel and ouptut chanel are not the same. (Currently, we add shortcuts for all bottlenecks in the bottleneck sequence except the first one.)~~
18 | 3. The paper says there are 19 bottlenecks, while there only 17 bottlene in Table 2. 
19 | 
20 | ### Suggestion
21 | 1. Strongly recommend that reimplement the paper use mxnet, pytorch, tensorflow, other than caffe, since there are optimized depthwise conv layer. 
22 | 2. Don't forget set the weight decay 4e-5. 
23 | 3. inception data augmentation helps.
24 | 


--------------------------------------------------------------------------------
/mobilenetv2-0.35/test.prototxt:
--------------------------------------------------------------------------------
   1 | layer {
   2 |   name: "data"
   3 |   type: "MemoryData"
   4 |   top: "data"
   5 |   top: "label"
   6 |   transform_param {
   7 |     crop_size: 224
   8 |   }
   9 |   memory_data_param {
  10 |     batch_size: 1
  11 |     channels: 3
  12 |     height: 224
  13 |     width: 224
  14 |   }
  15 | }
  16 | layer {
  17 |   name: "conv_1_1"
  18 |   type: "Convolution"
  19 |   bottom: "data"
  20 |   top: "conv_1_1"
  21 |   param {
  22 |     lr_mult: 1.0
  23 |     decay_mult: 1.0
  24 |   }
  25 |   convolution_param {
  26 |     num_output: 16
  27 |     bias_term: true
  28 |     pad: 1
  29 |     kernel_size: 3
  30 |     stride: 2
  31 |     weight_filler {
  32 |       type: "msra"
  33 |     }
  34 |     bias_filler {
  35 |       type: "constant"
  36 |       value: 0.0
  37 |     }
  38 |   }
  39 | }
  40 | layer {
  41 |   name: "bn_1_1"
  42 |   type: "BatchNorm"
  43 |   bottom: "conv_1_1"
  44 |   top: "conv_1_1"
  45 |   param {
  46 |     lr_mult: 0.0
  47 |     decay_mult: 0.0
  48 |   }
  49 |   param {
  50 |     lr_mult: 0.0
  51 |     decay_mult: 0.0
  52 |   }
  53 |   param {
  54 |     lr_mult: 0.0
  55 |     decay_mult: 0.0
  56 |   }
  57 |   batch_norm_param {
  58 |     use_global_stats: true
  59 |   }
  60 | }
  61 | layer {
  62 |   name: "scale_1_1"
  63 |   type: "Scale"
  64 |   bottom: "conv_1_1"
  65 |   top: "conv_1_1"
  66 |   scale_param {
  67 |     bias_term: true
  68 |   }
  69 | }
  70 | layer {
  71 |   name: "relu_1_1"
  72 |   type: "ReLU"
  73 |   bottom: "conv_1_1"
  74 |   top: "conv_1_1"
  75 | }
  76 | layer {
  77 |   name: "conv_2_1_pw"
  78 |   type: "Convolution"
  79 |   bottom: "conv_1_1"
  80 |   top: "conv_2_1_pw"
  81 |   param {
  82 |     lr_mult: 1.0
  83 |     decay_mult: 1.0
  84 |   }
  85 |   convolution_param {
  86 |     num_output: 16
  87 |     bias_term: false
  88 |     pad: 0
  89 |     kernel_size: 1
  90 |     stride: 1
  91 |     weight_filler {
  92 |       type: "msra"
  93 |     }
  94 |     bias_filler {
  95 |       type: "constant"
  96 |       value: 0.0
  97 |     }
  98 |   }
  99 | }
 100 | layer {
 101 |   name: "bn_2_1_pw"
 102 |   type: "BatchNorm"
 103 |   bottom: "conv_2_1_pw"
 104 |   top: "conv_2_1_pw"
 105 |   param {
 106 |     lr_mult: 0.0
 107 |     decay_mult: 0.0
 108 |   }
 109 |   param {
 110 |     lr_mult: 0.0
 111 |     decay_mult: 0.0
 112 |   }
 113 |   param {
 114 |     lr_mult: 0.0
 115 |     decay_mult: 0.0
 116 |   }
 117 |   batch_norm_param {
 118 |     use_global_stats: true
 119 |   }
 120 | }
 121 | layer {
 122 |   name: "scale_2_1_pw"
 123 |   type: "Scale"
 124 |   bottom: "conv_2_1_pw"
 125 |   top: "conv_2_1_pw"
 126 |   scale_param {
 127 |     bias_term: true
 128 |   }
 129 | }
 130 | layer {
 131 |   name: "relu_2_1_pw"
 132 |   type: "ReLU"
 133 |   bottom: "conv_2_1_pw"
 134 |   top: "conv_2_1_pw"
 135 | }
 136 | layer {
 137 |   name: "conv_2_1_dw"
 138 |   type: "Convolution"
 139 |   bottom: "conv_2_1_pw"
 140 |   top: "conv_2_1_dw"
 141 |   param {
 142 |     lr_mult: 1.0
 143 |     decay_mult: 1.0
 144 |   }
 145 |   convolution_param {
 146 |     num_output: 16
 147 |     bias_term: false
 148 |     pad: 1
 149 |     kernel_size: 3
 150 |     group: 16
 151 |     stride: 1
 152 |     weight_filler {
 153 |       type: "msra"
 154 |     }
 155 |     bias_filler {
 156 |       type: "constant"
 157 |       value: 0.0
 158 |     }
 159 |     engine: CAFFE
 160 |   }
 161 | }
 162 | layer {
 163 |   name: "bn_2_1_dw"
 164 |   type: "BatchNorm"
 165 |   bottom: "conv_2_1_dw"
 166 |   top: "conv_2_1_dw"
 167 |   param {
 168 |     lr_mult: 0.0
 169 |     decay_mult: 0.0
 170 |   }
 171 |   param {
 172 |     lr_mult: 0.0
 173 |     decay_mult: 0.0
 174 |   }
 175 |   param {
 176 |     lr_mult: 0.0
 177 |     decay_mult: 0.0
 178 |   }
 179 |   batch_norm_param {
 180 |     use_global_stats: true
 181 |   }
 182 | }
 183 | layer {
 184 |   name: "scale_2_1_dw"
 185 |   type: "Scale"
 186 |   bottom: "conv_2_1_dw"
 187 |   top: "conv_2_1_dw"
 188 |   scale_param {
 189 |     bias_term: true
 190 |   }
 191 | }
 192 | layer {
 193 |   name: "relu_2_1_dw"
 194 |   type: "ReLU"
 195 |   bottom: "conv_2_1_dw"
 196 |   top: "conv_2_1_dw"
 197 | }
 198 | layer {
 199 |   name: "conv_2_1_linear"
 200 |   type: "Convolution"
 201 |   bottom: "conv_2_1_dw"
 202 |   top: "conv_2_1_linear"
 203 |   param {
 204 |     lr_mult: 1.0
 205 |     decay_mult: 1.0
 206 |   }
 207 |   convolution_param {
 208 |     num_output: 8
 209 |     bias_term: false
 210 |     pad: 0
 211 |     kernel_size: 1
 212 |     stride: 1
 213 |     weight_filler {
 214 |       type: "msra"
 215 |     }
 216 |     bias_filler {
 217 |       type: "constant"
 218 |       value: 0.0
 219 |     }
 220 |   }
 221 | }
 222 | layer {
 223 |   name: "bn_2_1_linear"
 224 |   type: "BatchNorm"
 225 |   bottom: "conv_2_1_linear"
 226 |   top: "conv_2_1_linear"
 227 |   param {
 228 |     lr_mult: 0.0
 229 |     decay_mult: 0.0
 230 |   }
 231 |   param {
 232 |     lr_mult: 0.0
 233 |     decay_mult: 0.0
 234 |   }
 235 |   param {
 236 |     lr_mult: 0.0
 237 |     decay_mult: 0.0
 238 |   }
 239 |   batch_norm_param {
 240 |     use_global_stats: true
 241 |   }
 242 | }
 243 | layer {
 244 |   name: "scale_2_1_linear"
 245 |   type: "Scale"
 246 |   bottom: "conv_2_1_linear"
 247 |   top: "conv_2_1_linear"
 248 |   scale_param {
 249 |     bias_term: true
 250 |   }
 251 | }
 252 | layer {
 253 |   name: "conv_3_1_pw"
 254 |   type: "Convolution"
 255 |   bottom: "conv_2_1_linear"
 256 |   top: "conv_3_1_pw"
 257 |   param {
 258 |     lr_mult: 1.0
 259 |     decay_mult: 1.0
 260 |   }
 261 |   convolution_param {
 262 |     num_output: 48
 263 |     bias_term: false
 264 |     pad: 0
 265 |     kernel_size: 1
 266 |     stride: 1
 267 |     weight_filler {
 268 |       type: "msra"
 269 |     }
 270 |     bias_filler {
 271 |       type: "constant"
 272 |       value: 0.0
 273 |     }
 274 |   }
 275 | }
 276 | layer {
 277 |   name: "bn_3_1_pw"
 278 |   type: "BatchNorm"
 279 |   bottom: "conv_3_1_pw"
 280 |   top: "conv_3_1_pw"
 281 |   param {
 282 |     lr_mult: 0.0
 283 |     decay_mult: 0.0
 284 |   }
 285 |   param {
 286 |     lr_mult: 0.0
 287 |     decay_mult: 0.0
 288 |   }
 289 |   param {
 290 |     lr_mult: 0.0
 291 |     decay_mult: 0.0
 292 |   }
 293 |   batch_norm_param {
 294 |     use_global_stats: true
 295 |   }
 296 | }
 297 | layer {
 298 |   name: "scale_3_1_pw"
 299 |   type: "Scale"
 300 |   bottom: "conv_3_1_pw"
 301 |   top: "conv_3_1_pw"
 302 |   scale_param {
 303 |     bias_term: true
 304 |   }
 305 | }
 306 | layer {
 307 |   name: "relu_3_1_pw"
 308 |   type: "ReLU"
 309 |   bottom: "conv_3_1_pw"
 310 |   top: "conv_3_1_pw"
 311 | }
 312 | layer {
 313 |   name: "conv_3_1_dw"
 314 |   type: "Convolution"
 315 |   bottom: "conv_3_1_pw"
 316 |   top: "conv_3_1_dw"
 317 |   param {
 318 |     lr_mult: 1.0
 319 |     decay_mult: 1.0
 320 |   }
 321 |   convolution_param {
 322 |     num_output: 48
 323 |     bias_term: false
 324 |     pad: 1
 325 |     kernel_size: 3
 326 |     group: 48
 327 |     stride: 2
 328 |     weight_filler {
 329 |       type: "msra"
 330 |     }
 331 |     bias_filler {
 332 |       type: "constant"
 333 |       value: 0.0
 334 |     }
 335 |     engine: CAFFE
 336 |   }
 337 | }
 338 | layer {
 339 |   name: "bn_3_1_dw"
 340 |   type: "BatchNorm"
 341 |   bottom: "conv_3_1_dw"
 342 |   top: "conv_3_1_dw"
 343 |   param {
 344 |     lr_mult: 0.0
 345 |     decay_mult: 0.0
 346 |   }
 347 |   param {
 348 |     lr_mult: 0.0
 349 |     decay_mult: 0.0
 350 |   }
 351 |   param {
 352 |     lr_mult: 0.0
 353 |     decay_mult: 0.0
 354 |   }
 355 |   batch_norm_param {
 356 |     use_global_stats: true
 357 |   }
 358 | }
 359 | layer {
 360 |   name: "scale_3_1_dw"
 361 |   type: "Scale"
 362 |   bottom: "conv_3_1_dw"
 363 |   top: "conv_3_1_dw"
 364 |   scale_param {
 365 |     bias_term: true
 366 |   }
 367 | }
 368 | layer {
 369 |   name: "relu_3_1_dw"
 370 |   type: "ReLU"
 371 |   bottom: "conv_3_1_dw"
 372 |   top: "conv_3_1_dw"
 373 | }
 374 | layer {
 375 |   name: "conv_3_1_linear"
 376 |   type: "Convolution"
 377 |   bottom: "conv_3_1_dw"
 378 |   top: "conv_3_1_linear"
 379 |   param {
 380 |     lr_mult: 1.0
 381 |     decay_mult: 1.0
 382 |   }
 383 |   convolution_param {
 384 |     num_output: 8
 385 |     bias_term: false
 386 |     pad: 0
 387 |     kernel_size: 1
 388 |     stride: 1
 389 |     weight_filler {
 390 |       type: "msra"
 391 |     }
 392 |     bias_filler {
 393 |       type: "constant"
 394 |       value: 0.0
 395 |     }
 396 |   }
 397 | }
 398 | layer {
 399 |   name: "bn_3_1_linear"
 400 |   type: "BatchNorm"
 401 |   bottom: "conv_3_1_linear"
 402 |   top: "conv_3_1_linear"
 403 |   param {
 404 |     lr_mult: 0.0
 405 |     decay_mult: 0.0
 406 |   }
 407 |   param {
 408 |     lr_mult: 0.0
 409 |     decay_mult: 0.0
 410 |   }
 411 |   param {
 412 |     lr_mult: 0.0
 413 |     decay_mult: 0.0
 414 |   }
 415 |   batch_norm_param {
 416 |     use_global_stats: true
 417 |   }
 418 | }
 419 | layer {
 420 |   name: "scale_3_1_linear"
 421 |   type: "Scale"
 422 |   bottom: "conv_3_1_linear"
 423 |   top: "conv_3_1_linear"
 424 |   scale_param {
 425 |     bias_term: true
 426 |   }
 427 | }
 428 | layer {
 429 |   name: "conv_3_2_pw"
 430 |   type: "Convolution"
 431 |   bottom: "conv_3_1_linear"
 432 |   top: "conv_3_2_pw"
 433 |   param {
 434 |     lr_mult: 1.0
 435 |     decay_mult: 1.0
 436 |   }
 437 |   convolution_param {
 438 |     num_output: 48
 439 |     bias_term: false
 440 |     pad: 0
 441 |     kernel_size: 1
 442 |     stride: 1
 443 |     weight_filler {
 444 |       type: "msra"
 445 |     }
 446 |     bias_filler {
 447 |       type: "constant"
 448 |       value: 0.0
 449 |     }
 450 |   }
 451 | }
 452 | layer {
 453 |   name: "bn_3_2_pw"
 454 |   type: "BatchNorm"
 455 |   bottom: "conv_3_2_pw"
 456 |   top: "conv_3_2_pw"
 457 |   param {
 458 |     lr_mult: 0.0
 459 |     decay_mult: 0.0
 460 |   }
 461 |   param {
 462 |     lr_mult: 0.0
 463 |     decay_mult: 0.0
 464 |   }
 465 |   param {
 466 |     lr_mult: 0.0
 467 |     decay_mult: 0.0
 468 |   }
 469 |   batch_norm_param {
 470 |     use_global_stats: true
 471 |   }
 472 | }
 473 | layer {
 474 |   name: "scale_3_2_pw"
 475 |   type: "Scale"
 476 |   bottom: "conv_3_2_pw"
 477 |   top: "conv_3_2_pw"
 478 |   scale_param {
 479 |     bias_term: true
 480 |   }
 481 | }
 482 | layer {
 483 |   name: "relu_3_2_pw"
 484 |   type: "ReLU"
 485 |   bottom: "conv_3_2_pw"
 486 |   top: "conv_3_2_pw"
 487 | }
 488 | layer {
 489 |   name: "conv_3_2_dw"
 490 |   type: "Convolution"
 491 |   bottom: "conv_3_2_pw"
 492 |   top: "conv_3_2_dw"
 493 |   param {
 494 |     lr_mult: 1.0
 495 |     decay_mult: 1.0
 496 |   }
 497 |   convolution_param {
 498 |     num_output: 48
 499 |     bias_term: false
 500 |     pad: 1
 501 |     kernel_size: 3
 502 |     group: 48
 503 |     stride: 1
 504 |     weight_filler {
 505 |       type: "msra"
 506 |     }
 507 |     bias_filler {
 508 |       type: "constant"
 509 |       value: 0.0
 510 |     }
 511 |     engine: CAFFE
 512 |   }
 513 | }
 514 | layer {
 515 |   name: "bn_3_2_dw"
 516 |   type: "BatchNorm"
 517 |   bottom: "conv_3_2_dw"
 518 |   top: "conv_3_2_dw"
 519 |   param {
 520 |     lr_mult: 0.0
 521 |     decay_mult: 0.0
 522 |   }
 523 |   param {
 524 |     lr_mult: 0.0
 525 |     decay_mult: 0.0
 526 |   }
 527 |   param {
 528 |     lr_mult: 0.0
 529 |     decay_mult: 0.0
 530 |   }
 531 |   batch_norm_param {
 532 |     use_global_stats: true
 533 |   }
 534 | }
 535 | layer {
 536 |   name: "scale_3_2_dw"
 537 |   type: "Scale"
 538 |   bottom: "conv_3_2_dw"
 539 |   top: "conv_3_2_dw"
 540 |   scale_param {
 541 |     bias_term: true
 542 |   }
 543 | }
 544 | layer {
 545 |   name: "relu_3_2_dw"
 546 |   type: "ReLU"
 547 |   bottom: "conv_3_2_dw"
 548 |   top: "conv_3_2_dw"
 549 | }
 550 | layer {
 551 |   name: "conv_3_2_linear"
 552 |   type: "Convolution"
 553 |   bottom: "conv_3_2_dw"
 554 |   top: "conv_3_2_linear"
 555 |   param {
 556 |     lr_mult: 1.0
 557 |     decay_mult: 1.0
 558 |   }
 559 |   convolution_param {
 560 |     num_output: 8
 561 |     bias_term: false
 562 |     pad: 0
 563 |     kernel_size: 1
 564 |     stride: 1
 565 |     weight_filler {
 566 |       type: "msra"
 567 |     }
 568 |     bias_filler {
 569 |       type: "constant"
 570 |       value: 0.0
 571 |     }
 572 |   }
 573 | }
 574 | layer {
 575 |   name: "bn_3_2_linear"
 576 |   type: "BatchNorm"
 577 |   bottom: "conv_3_2_linear"
 578 |   top: "conv_3_2_linear"
 579 |   param {
 580 |     lr_mult: 0.0
 581 |     decay_mult: 0.0
 582 |   }
 583 |   param {
 584 |     lr_mult: 0.0
 585 |     decay_mult: 0.0
 586 |   }
 587 |   param {
 588 |     lr_mult: 0.0
 589 |     decay_mult: 0.0
 590 |   }
 591 |   batch_norm_param {
 592 |     use_global_stats: true
 593 |   }
 594 | }
 595 | layer {
 596 |   name: "scale_3_2_linear"
 597 |   type: "Scale"
 598 |   bottom: "conv_3_2_linear"
 599 |   top: "conv_3_2_linear"
 600 |   scale_param {
 601 |     bias_term: true
 602 |   }
 603 | }
 604 | layer {
 605 |   name: "add_3_2"
 606 |   type: "Eltwise"
 607 |   bottom: "conv_3_1_linear"
 608 |   bottom: "conv_3_2_linear"
 609 |   top: "add_3_2"
 610 | }
 611 | layer {
 612 |   name: "conv_4_1_pw"
 613 |   type: "Convolution"
 614 |   bottom: "add_3_2"
 615 |   top: "conv_4_1_pw"
 616 |   param {
 617 |     lr_mult: 1.0
 618 |     decay_mult: 1.0
 619 |   }
 620 |   convolution_param {
 621 |     num_output: 48
 622 |     bias_term: false
 623 |     pad: 0
 624 |     kernel_size: 1
 625 |     stride: 1
 626 |     weight_filler {
 627 |       type: "msra"
 628 |     }
 629 |     bias_filler {
 630 |       type: "constant"
 631 |       value: 0.0
 632 |     }
 633 |   }
 634 | }
 635 | layer {
 636 |   name: "bn_4_1_pw"
 637 |   type: "BatchNorm"
 638 |   bottom: "conv_4_1_pw"
 639 |   top: "conv_4_1_pw"
 640 |   param {
 641 |     lr_mult: 0.0
 642 |     decay_mult: 0.0
 643 |   }
 644 |   param {
 645 |     lr_mult: 0.0
 646 |     decay_mult: 0.0
 647 |   }
 648 |   param {
 649 |     lr_mult: 0.0
 650 |     decay_mult: 0.0
 651 |   }
 652 |   batch_norm_param {
 653 |     use_global_stats: true
 654 |   }
 655 | }
 656 | layer {
 657 |   name: "scale_4_1_pw"
 658 |   type: "Scale"
 659 |   bottom: "conv_4_1_pw"
 660 |   top: "conv_4_1_pw"
 661 |   scale_param {
 662 |     bias_term: true
 663 |   }
 664 | }
 665 | layer {
 666 |   name: "relu_4_1_pw"
 667 |   type: "ReLU"
 668 |   bottom: "conv_4_1_pw"
 669 |   top: "conv_4_1_pw"
 670 | }
 671 | layer {
 672 |   name: "conv_4_1_dw"
 673 |   type: "Convolution"
 674 |   bottom: "conv_4_1_pw"
 675 |   top: "conv_4_1_dw"
 676 |   param {
 677 |     lr_mult: 1.0
 678 |     decay_mult: 1.0
 679 |   }
 680 |   convolution_param {
 681 |     num_output: 48
 682 |     bias_term: false
 683 |     pad: 1
 684 |     kernel_size: 3
 685 |     group: 48
 686 |     stride: 2
 687 |     weight_filler {
 688 |       type: "msra"
 689 |     }
 690 |     bias_filler {
 691 |       type: "constant"
 692 |       value: 0.0
 693 |     }
 694 |     engine: CAFFE
 695 |   }
 696 | }
 697 | layer {
 698 |   name: "bn_4_1_dw"
 699 |   type: "BatchNorm"
 700 |   bottom: "conv_4_1_dw"
 701 |   top: "conv_4_1_dw"
 702 |   param {
 703 |     lr_mult: 0.0
 704 |     decay_mult: 0.0
 705 |   }
 706 |   param {
 707 |     lr_mult: 0.0
 708 |     decay_mult: 0.0
 709 |   }
 710 |   param {
 711 |     lr_mult: 0.0
 712 |     decay_mult: 0.0
 713 |   }
 714 |   batch_norm_param {
 715 |     use_global_stats: true
 716 |   }
 717 | }
 718 | layer {
 719 |   name: "scale_4_1_dw"
 720 |   type: "Scale"
 721 |   bottom: "conv_4_1_dw"
 722 |   top: "conv_4_1_dw"
 723 |   scale_param {
 724 |     bias_term: true
 725 |   }
 726 | }
 727 | layer {
 728 |   name: "relu_4_1_dw"
 729 |   type: "ReLU"
 730 |   bottom: "conv_4_1_dw"
 731 |   top: "conv_4_1_dw"
 732 | }
 733 | layer {
 734 |   name: "conv_4_1_linear"
 735 |   type: "Convolution"
 736 |   bottom: "conv_4_1_dw"
 737 |   top: "conv_4_1_linear"
 738 |   param {
 739 |     lr_mult: 1.0
 740 |     decay_mult: 1.0
 741 |   }
 742 |   convolution_param {
 743 |     num_output: 16
 744 |     bias_term: false
 745 |     pad: 0
 746 |     kernel_size: 1
 747 |     stride: 1
 748 |     weight_filler {
 749 |       type: "msra"
 750 |     }
 751 |     bias_filler {
 752 |       type: "constant"
 753 |       value: 0.0
 754 |     }
 755 |   }
 756 | }
 757 | layer {
 758 |   name: "bn_4_1_linear"
 759 |   type: "BatchNorm"
 760 |   bottom: "conv_4_1_linear"
 761 |   top: "conv_4_1_linear"
 762 |   param {
 763 |     lr_mult: 0.0
 764 |     decay_mult: 0.0
 765 |   }
 766 |   param {
 767 |     lr_mult: 0.0
 768 |     decay_mult: 0.0
 769 |   }
 770 |   param {
 771 |     lr_mult: 0.0
 772 |     decay_mult: 0.0
 773 |   }
 774 |   batch_norm_param {
 775 |     use_global_stats: true
 776 |   }
 777 | }
 778 | layer {
 779 |   name: "scale_4_1_linear"
 780 |   type: "Scale"
 781 |   bottom: "conv_4_1_linear"
 782 |   top: "conv_4_1_linear"
 783 |   scale_param {
 784 |     bias_term: true
 785 |   }
 786 | }
 787 | layer {
 788 |   name: "conv_4_2_pw"
 789 |   type: "Convolution"
 790 |   bottom: "conv_4_1_linear"
 791 |   top: "conv_4_2_pw"
 792 |   param {
 793 |     lr_mult: 1.0
 794 |     decay_mult: 1.0
 795 |   }
 796 |   convolution_param {
 797 |     num_output: 96
 798 |     bias_term: false
 799 |     pad: 0
 800 |     kernel_size: 1
 801 |     stride: 1
 802 |     weight_filler {
 803 |       type: "msra"
 804 |     }
 805 |     bias_filler {
 806 |       type: "constant"
 807 |       value: 0.0
 808 |     }
 809 |   }
 810 | }
 811 | layer {
 812 |   name: "bn_4_2_pw"
 813 |   type: "BatchNorm"
 814 |   bottom: "conv_4_2_pw"
 815 |   top: "conv_4_2_pw"
 816 |   param {
 817 |     lr_mult: 0.0
 818 |     decay_mult: 0.0
 819 |   }
 820 |   param {
 821 |     lr_mult: 0.0
 822 |     decay_mult: 0.0
 823 |   }
 824 |   param {
 825 |     lr_mult: 0.0
 826 |     decay_mult: 0.0
 827 |   }
 828 |   batch_norm_param {
 829 |     use_global_stats: true
 830 |   }
 831 | }
 832 | layer {
 833 |   name: "scale_4_2_pw"
 834 |   type: "Scale"
 835 |   bottom: "conv_4_2_pw"
 836 |   top: "conv_4_2_pw"
 837 |   scale_param {
 838 |     bias_term: true
 839 |   }
 840 | }
 841 | layer {
 842 |   name: "relu_4_2_pw"
 843 |   type: "ReLU"
 844 |   bottom: "conv_4_2_pw"
 845 |   top: "conv_4_2_pw"
 846 | }
 847 | layer {
 848 |   name: "conv_4_2_dw"
 849 |   type: "Convolution"
 850 |   bottom: "conv_4_2_pw"
 851 |   top: "conv_4_2_dw"
 852 |   param {
 853 |     lr_mult: 1.0
 854 |     decay_mult: 1.0
 855 |   }
 856 |   convolution_param {
 857 |     num_output: 96
 858 |     bias_term: false
 859 |     pad: 1
 860 |     kernel_size: 3
 861 |     group: 96
 862 |     stride: 1
 863 |     weight_filler {
 864 |       type: "msra"
 865 |     }
 866 |     bias_filler {
 867 |       type: "constant"
 868 |       value: 0.0
 869 |     }
 870 |     engine: CAFFE
 871 |   }
 872 | }
 873 | layer {
 874 |   name: "bn_4_2_dw"
 875 |   type: "BatchNorm"
 876 |   bottom: "conv_4_2_dw"
 877 |   top: "conv_4_2_dw"
 878 |   param {
 879 |     lr_mult: 0.0
 880 |     decay_mult: 0.0
 881 |   }
 882 |   param {
 883 |     lr_mult: 0.0
 884 |     decay_mult: 0.0
 885 |   }
 886 |   param {
 887 |     lr_mult: 0.0
 888 |     decay_mult: 0.0
 889 |   }
 890 |   batch_norm_param {
 891 |     use_global_stats: true
 892 |   }
 893 | }
 894 | layer {
 895 |   name: "scale_4_2_dw"
 896 |   type: "Scale"
 897 |   bottom: "conv_4_2_dw"
 898 |   top: "conv_4_2_dw"
 899 |   scale_param {
 900 |     bias_term: true
 901 |   }
 902 | }
 903 | layer {
 904 |   name: "relu_4_2_dw"
 905 |   type: "ReLU"
 906 |   bottom: "conv_4_2_dw"
 907 |   top: "conv_4_2_dw"
 908 | }
 909 | layer {
 910 |   name: "conv_4_2_linear"
 911 |   type: "Convolution"
 912 |   bottom: "conv_4_2_dw"
 913 |   top: "conv_4_2_linear"
 914 |   param {
 915 |     lr_mult: 1.0
 916 |     decay_mult: 1.0
 917 |   }
 918 |   convolution_param {
 919 |     num_output: 16
 920 |     bias_term: false
 921 |     pad: 0
 922 |     kernel_size: 1
 923 |     stride: 1
 924 |     weight_filler {
 925 |       type: "msra"
 926 |     }
 927 |     bias_filler {
 928 |       type: "constant"
 929 |       value: 0.0
 930 |     }
 931 |   }
 932 | }
 933 | layer {
 934 |   name: "bn_4_2_linear"
 935 |   type: "BatchNorm"
 936 |   bottom: "conv_4_2_linear"
 937 |   top: "conv_4_2_linear"
 938 |   param {
 939 |     lr_mult: 0.0
 940 |     decay_mult: 0.0
 941 |   }
 942 |   param {
 943 |     lr_mult: 0.0
 944 |     decay_mult: 0.0
 945 |   }
 946 |   param {
 947 |     lr_mult: 0.0
 948 |     decay_mult: 0.0
 949 |   }
 950 |   batch_norm_param {
 951 |     use_global_stats: true
 952 |   }
 953 | }
 954 | layer {
 955 |   name: "scale_4_2_linear"
 956 |   type: "Scale"
 957 |   bottom: "conv_4_2_linear"
 958 |   top: "conv_4_2_linear"
 959 |   scale_param {
 960 |     bias_term: true
 961 |   }
 962 | }
 963 | layer {
 964 |   name: "add_4_2"
 965 |   type: "Eltwise"
 966 |   bottom: "conv_4_1_linear"
 967 |   bottom: "conv_4_2_linear"
 968 |   top: "add_4_2"
 969 | }
 970 | layer {
 971 |   name: "conv_4_3_pw"
 972 |   type: "Convolution"
 973 |   bottom: "add_4_2"
 974 |   top: "conv_4_3_pw"
 975 |   param {
 976 |     lr_mult: 1.0
 977 |     decay_mult: 1.0
 978 |   }
 979 |   convolution_param {
 980 |     num_output: 96
 981 |     bias_term: false
 982 |     pad: 0
 983 |     kernel_size: 1
 984 |     stride: 1
 985 |     weight_filler {
 986 |       type: "msra"
 987 |     }
 988 |     bias_filler {
 989 |       type: "constant"
 990 |       value: 0.0
 991 |     }
 992 |   }
 993 | }
 994 | layer {
 995 |   name: "bn_4_3_pw"
 996 |   type: "BatchNorm"
 997 |   bottom: "conv_4_3_pw"
 998 |   top: "conv_4_3_pw"
 999 |   param {
1000 |     lr_mult: 0.0
1001 |     decay_mult: 0.0
1002 |   }
1003 |   param {
1004 |     lr_mult: 0.0
1005 |     decay_mult: 0.0
1006 |   }
1007 |   param {
1008 |     lr_mult: 0.0
1009 |     decay_mult: 0.0
1010 |   }
1011 |   batch_norm_param {
1012 |     use_global_stats: true
1013 |   }
1014 | }
1015 | layer {
1016 |   name: "scale_4_3_pw"
1017 |   type: "Scale"
1018 |   bottom: "conv_4_3_pw"
1019 |   top: "conv_4_3_pw"
1020 |   scale_param {
1021 |     bias_term: true
1022 |   }
1023 | }
1024 | layer {
1025 |   name: "relu_4_3_pw"
1026 |   type: "ReLU"
1027 |   bottom: "conv_4_3_pw"
1028 |   top: "conv_4_3_pw"
1029 | }
1030 | layer {
1031 |   name: "conv_4_3_dw"
1032 |   type: "Convolution"
1033 |   bottom: "conv_4_3_pw"
1034 |   top: "conv_4_3_dw"
1035 |   param {
1036 |     lr_mult: 1.0
1037 |     decay_mult: 1.0
1038 |   }
1039 |   convolution_param {
1040 |     num_output: 96
1041 |     bias_term: false
1042 |     pad: 1
1043 |     kernel_size: 3
1044 |     group: 96
1045 |     stride: 1
1046 |     weight_filler {
1047 |       type: "msra"
1048 |     }
1049 |     bias_filler {
1050 |       type: "constant"
1051 |       value: 0.0
1052 |     }
1053 |     engine: CAFFE
1054 |   }
1055 | }
1056 | layer {
1057 |   name: "bn_4_3_dw"
1058 |   type: "BatchNorm"
1059 |   bottom: "conv_4_3_dw"
1060 |   top: "conv_4_3_dw"
1061 |   param {
1062 |     lr_mult: 0.0
1063 |     decay_mult: 0.0
1064 |   }
1065 |   param {
1066 |     lr_mult: 0.0
1067 |     decay_mult: 0.0
1068 |   }
1069 |   param {
1070 |     lr_mult: 0.0
1071 |     decay_mult: 0.0
1072 |   }
1073 |   batch_norm_param {
1074 |     use_global_stats: true
1075 |   }
1076 | }
1077 | layer {
1078 |   name: "scale_4_3_dw"
1079 |   type: "Scale"
1080 |   bottom: "conv_4_3_dw"
1081 |   top: "conv_4_3_dw"
1082 |   scale_param {
1083 |     bias_term: true
1084 |   }
1085 | }
1086 | layer {
1087 |   name: "relu_4_3_dw"
1088 |   type: "ReLU"
1089 |   bottom: "conv_4_3_dw"
1090 |   top: "conv_4_3_dw"
1091 | }
1092 | layer {
1093 |   name: "conv_4_3_linear"
1094 |   type: "Convolution"
1095 |   bottom: "conv_4_3_dw"
1096 |   top: "conv_4_3_linear"
1097 |   param {
1098 |     lr_mult: 1.0
1099 |     decay_mult: 1.0
1100 |   }
1101 |   convolution_param {
1102 |     num_output: 16
1103 |     bias_term: false
1104 |     pad: 0
1105 |     kernel_size: 1
1106 |     stride: 1
1107 |     weight_filler {
1108 |       type: "msra"
1109 |     }
1110 |     bias_filler {
1111 |       type: "constant"
1112 |       value: 0.0
1113 |     }
1114 |   }
1115 | }
1116 | layer {
1117 |   name: "bn_4_3_linear"
1118 |   type: "BatchNorm"
1119 |   bottom: "conv_4_3_linear"
1120 |   top: "conv_4_3_linear"
1121 |   param {
1122 |     lr_mult: 0.0
1123 |     decay_mult: 0.0
1124 |   }
1125 |   param {
1126 |     lr_mult: 0.0
1127 |     decay_mult: 0.0
1128 |   }
1129 |   param {
1130 |     lr_mult: 0.0
1131 |     decay_mult: 0.0
1132 |   }
1133 |   batch_norm_param {
1134 |     use_global_stats: true
1135 |   }
1136 | }
1137 | layer {
1138 |   name: "scale_4_3_linear"
1139 |   type: "Scale"
1140 |   bottom: "conv_4_3_linear"
1141 |   top: "conv_4_3_linear"
1142 |   scale_param {
1143 |     bias_term: true
1144 |   }
1145 | }
1146 | layer {
1147 |   name: "add_4_3"
1148 |   type: "Eltwise"
1149 |   bottom: "add_4_2"
1150 |   bottom: "conv_4_3_linear"
1151 |   top: "add_4_3"
1152 | }
1153 | layer {
1154 |   name: "conv_5_1_pw"
1155 |   type: "Convolution"
1156 |   bottom: "add_4_3"
1157 |   top: "conv_5_1_pw"
1158 |   param {
1159 |     lr_mult: 1.0
1160 |     decay_mult: 1.0
1161 |   }
1162 |   convolution_param {
1163 |     num_output: 96
1164 |     bias_term: false
1165 |     pad: 0
1166 |     kernel_size: 1
1167 |     stride: 1
1168 |     weight_filler {
1169 |       type: "msra"
1170 |     }
1171 |     bias_filler {
1172 |       type: "constant"
1173 |       value: 0.0
1174 |     }
1175 |   }
1176 | }
1177 | layer {
1178 |   name: "bn_5_1_pw"
1179 |   type: "BatchNorm"
1180 |   bottom: "conv_5_1_pw"
1181 |   top: "conv_5_1_pw"
1182 |   param {
1183 |     lr_mult: 0.0
1184 |     decay_mult: 0.0
1185 |   }
1186 |   param {
1187 |     lr_mult: 0.0
1188 |     decay_mult: 0.0
1189 |   }
1190 |   param {
1191 |     lr_mult: 0.0
1192 |     decay_mult: 0.0
1193 |   }
1194 |   batch_norm_param {
1195 |     use_global_stats: true
1196 |   }
1197 | }
1198 | layer {
1199 |   name: "scale_5_1_pw"
1200 |   type: "Scale"
1201 |   bottom: "conv_5_1_pw"
1202 |   top: "conv_5_1_pw"
1203 |   scale_param {
1204 |     bias_term: true
1205 |   }
1206 | }
1207 | layer {
1208 |   name: "relu_5_1_pw"
1209 |   type: "ReLU"
1210 |   bottom: "conv_5_1_pw"
1211 |   top: "conv_5_1_pw"
1212 | }
1213 | layer {
1214 |   name: "conv_5_1_dw"
1215 |   type: "Convolution"
1216 |   bottom: "conv_5_1_pw"
1217 |   top: "conv_5_1_dw"
1218 |   param {
1219 |     lr_mult: 1.0
1220 |     decay_mult: 1.0
1221 |   }
1222 |   convolution_param {
1223 |     num_output: 96
1224 |     bias_term: false
1225 |     pad: 1
1226 |     kernel_size: 3
1227 |     group: 96
1228 |     stride: 2
1229 |     weight_filler {
1230 |       type: "msra"
1231 |     }
1232 |     bias_filler {
1233 |       type: "constant"
1234 |       value: 0.0
1235 |     }
1236 |     engine: CAFFE
1237 |   }
1238 | }
1239 | layer {
1240 |   name: "bn_5_1_dw"
1241 |   type: "BatchNorm"
1242 |   bottom: "conv_5_1_dw"
1243 |   top: "conv_5_1_dw"
1244 |   param {
1245 |     lr_mult: 0.0
1246 |     decay_mult: 0.0
1247 |   }
1248 |   param {
1249 |     lr_mult: 0.0
1250 |     decay_mult: 0.0
1251 |   }
1252 |   param {
1253 |     lr_mult: 0.0
1254 |     decay_mult: 0.0
1255 |   }
1256 |   batch_norm_param {
1257 |     use_global_stats: true
1258 |   }
1259 | }
1260 | layer {
1261 |   name: "scale_5_1_dw"
1262 |   type: "Scale"
1263 |   bottom: "conv_5_1_dw"
1264 |   top: "conv_5_1_dw"
1265 |   scale_param {
1266 |     bias_term: true
1267 |   }
1268 | }
1269 | layer {
1270 |   name: "relu_5_1_dw"
1271 |   type: "ReLU"
1272 |   bottom: "conv_5_1_dw"
1273 |   top: "conv_5_1_dw"
1274 | }
1275 | layer {
1276 |   name: "conv_5_1_linear"
1277 |   type: "Convolution"
1278 |   bottom: "conv_5_1_dw"
1279 |   top: "conv_5_1_linear"
1280 |   param {
1281 |     lr_mult: 1.0
1282 |     decay_mult: 1.0
1283 |   }
1284 |   convolution_param {
1285 |     num_output: 24
1286 |     bias_term: false
1287 |     pad: 0
1288 |     kernel_size: 1
1289 |     stride: 1
1290 |     weight_filler {
1291 |       type: "msra"
1292 |     }
1293 |     bias_filler {
1294 |       type: "constant"
1295 |       value: 0.0
1296 |     }
1297 |   }
1298 | }
1299 | layer {
1300 |   name: "bn_5_1_linear"
1301 |   type: "BatchNorm"
1302 |   bottom: "conv_5_1_linear"
1303 |   top: "conv_5_1_linear"
1304 |   param {
1305 |     lr_mult: 0.0
1306 |     decay_mult: 0.0
1307 |   }
1308 |   param {
1309 |     lr_mult: 0.0
1310 |     decay_mult: 0.0
1311 |   }
1312 |   param {
1313 |     lr_mult: 0.0
1314 |     decay_mult: 0.0
1315 |   }
1316 |   batch_norm_param {
1317 |     use_global_stats: true
1318 |   }
1319 | }
1320 | layer {
1321 |   name: "scale_5_1_linear"
1322 |   type: "Scale"
1323 |   bottom: "conv_5_1_linear"
1324 |   top: "conv_5_1_linear"
1325 |   scale_param {
1326 |     bias_term: true
1327 |   }
1328 | }
1329 | layer {
1330 |   name: "conv_5_2_pw"
1331 |   type: "Convolution"
1332 |   bottom: "conv_5_1_linear"
1333 |   top: "conv_5_2_pw"
1334 |   param {
1335 |     lr_mult: 1.0
1336 |     decay_mult: 1.0
1337 |   }
1338 |   convolution_param {
1339 |     num_output: 144
1340 |     bias_term: false
1341 |     pad: 0
1342 |     kernel_size: 1
1343 |     stride: 1
1344 |     weight_filler {
1345 |       type: "msra"
1346 |     }
1347 |     bias_filler {
1348 |       type: "constant"
1349 |       value: 0.0
1350 |     }
1351 |   }
1352 | }
1353 | layer {
1354 |   name: "bn_5_2_pw"
1355 |   type: "BatchNorm"
1356 |   bottom: "conv_5_2_pw"
1357 |   top: "conv_5_2_pw"
1358 |   param {
1359 |     lr_mult: 0.0
1360 |     decay_mult: 0.0
1361 |   }
1362 |   param {
1363 |     lr_mult: 0.0
1364 |     decay_mult: 0.0
1365 |   }
1366 |   param {
1367 |     lr_mult: 0.0
1368 |     decay_mult: 0.0
1369 |   }
1370 |   batch_norm_param {
1371 |     use_global_stats: true
1372 |   }
1373 | }
1374 | layer {
1375 |   name: "scale_5_2_pw"
1376 |   type: "Scale"
1377 |   bottom: "conv_5_2_pw"
1378 |   top: "conv_5_2_pw"
1379 |   scale_param {
1380 |     bias_term: true
1381 |   }
1382 | }
1383 | layer {
1384 |   name: "relu_5_2_pw"
1385 |   type: "ReLU"
1386 |   bottom: "conv_5_2_pw"
1387 |   top: "conv_5_2_pw"
1388 | }
1389 | layer {
1390 |   name: "conv_5_2_dw"
1391 |   type: "Convolution"
1392 |   bottom: "conv_5_2_pw"
1393 |   top: "conv_5_2_dw"
1394 |   param {
1395 |     lr_mult: 1.0
1396 |     decay_mult: 1.0
1397 |   }
1398 |   convolution_param {
1399 |     num_output: 144
1400 |     bias_term: false
1401 |     pad: 1
1402 |     kernel_size: 3
1403 |     group: 144
1404 |     stride: 1
1405 |     weight_filler {
1406 |       type: "msra"
1407 |     }
1408 |     bias_filler {
1409 |       type: "constant"
1410 |       value: 0.0
1411 |     }
1412 |     engine: CAFFE
1413 |   }
1414 | }
1415 | layer {
1416 |   name: "bn_5_2_dw"
1417 |   type: "BatchNorm"
1418 |   bottom: "conv_5_2_dw"
1419 |   top: "conv_5_2_dw"
1420 |   param {
1421 |     lr_mult: 0.0
1422 |     decay_mult: 0.0
1423 |   }
1424 |   param {
1425 |     lr_mult: 0.0
1426 |     decay_mult: 0.0
1427 |   }
1428 |   param {
1429 |     lr_mult: 0.0
1430 |     decay_mult: 0.0
1431 |   }
1432 |   batch_norm_param {
1433 |     use_global_stats: true
1434 |   }
1435 | }
1436 | layer {
1437 |   name: "scale_5_2_dw"
1438 |   type: "Scale"
1439 |   bottom: "conv_5_2_dw"
1440 |   top: "conv_5_2_dw"
1441 |   scale_param {
1442 |     bias_term: true
1443 |   }
1444 | }
1445 | layer {
1446 |   name: "relu_5_2_dw"
1447 |   type: "ReLU"
1448 |   bottom: "conv_5_2_dw"
1449 |   top: "conv_5_2_dw"
1450 | }
1451 | layer {
1452 |   name: "conv_5_2_linear"
1453 |   type: "Convolution"
1454 |   bottom: "conv_5_2_dw"
1455 |   top: "conv_5_2_linear"
1456 |   param {
1457 |     lr_mult: 1.0
1458 |     decay_mult: 1.0
1459 |   }
1460 |   convolution_param {
1461 |     num_output: 24
1462 |     bias_term: false
1463 |     pad: 0
1464 |     kernel_size: 1
1465 |     stride: 1
1466 |     weight_filler {
1467 |       type: "msra"
1468 |     }
1469 |     bias_filler {
1470 |       type: "constant"
1471 |       value: 0.0
1472 |     }
1473 |   }
1474 | }
1475 | layer {
1476 |   name: "bn_5_2_linear"
1477 |   type: "BatchNorm"
1478 |   bottom: "conv_5_2_linear"
1479 |   top: "conv_5_2_linear"
1480 |   param {
1481 |     lr_mult: 0.0
1482 |     decay_mult: 0.0
1483 |   }
1484 |   param {
1485 |     lr_mult: 0.0
1486 |     decay_mult: 0.0
1487 |   }
1488 |   param {
1489 |     lr_mult: 0.0
1490 |     decay_mult: 0.0
1491 |   }
1492 |   batch_norm_param {
1493 |     use_global_stats: true
1494 |   }
1495 | }
1496 | layer {
1497 |   name: "scale_5_2_linear"
1498 |   type: "Scale"
1499 |   bottom: "conv_5_2_linear"
1500 |   top: "conv_5_2_linear"
1501 |   scale_param {
1502 |     bias_term: true
1503 |   }
1504 | }
1505 | layer {
1506 |   name: "add_5_2"
1507 |   type: "Eltwise"
1508 |   bottom: "conv_5_1_linear"
1509 |   bottom: "conv_5_2_linear"
1510 |   top: "add_5_2"
1511 | }
1512 | layer {
1513 |   name: "conv_5_3_pw"
1514 |   type: "Convolution"
1515 |   bottom: "add_5_2"
1516 |   top: "conv_5_3_pw"
1517 |   param {
1518 |     lr_mult: 1.0
1519 |     decay_mult: 1.0
1520 |   }
1521 |   convolution_param {
1522 |     num_output: 144
1523 |     bias_term: false
1524 |     pad: 0
1525 |     kernel_size: 1
1526 |     stride: 1
1527 |     weight_filler {
1528 |       type: "msra"
1529 |     }
1530 |     bias_filler {
1531 |       type: "constant"
1532 |       value: 0.0
1533 |     }
1534 |   }
1535 | }
1536 | layer {
1537 |   name: "bn_5_3_pw"
1538 |   type: "BatchNorm"
1539 |   bottom: "conv_5_3_pw"
1540 |   top: "conv_5_3_pw"
1541 |   param {
1542 |     lr_mult: 0.0
1543 |     decay_mult: 0.0
1544 |   }
1545 |   param {
1546 |     lr_mult: 0.0
1547 |     decay_mult: 0.0
1548 |   }
1549 |   param {
1550 |     lr_mult: 0.0
1551 |     decay_mult: 0.0
1552 |   }
1553 |   batch_norm_param {
1554 |     use_global_stats: true
1555 |   }
1556 | }
1557 | layer {
1558 |   name: "scale_5_3_pw"
1559 |   type: "Scale"
1560 |   bottom: "conv_5_3_pw"
1561 |   top: "conv_5_3_pw"
1562 |   scale_param {
1563 |     bias_term: true
1564 |   }
1565 | }
1566 | layer {
1567 |   name: "relu_5_3_pw"
1568 |   type: "ReLU"
1569 |   bottom: "conv_5_3_pw"
1570 |   top: "conv_5_3_pw"
1571 | }
1572 | layer {
1573 |   name: "conv_5_3_dw"
1574 |   type: "Convolution"
1575 |   bottom: "conv_5_3_pw"
1576 |   top: "conv_5_3_dw"
1577 |   param {
1578 |     lr_mult: 1.0
1579 |     decay_mult: 1.0
1580 |   }
1581 |   convolution_param {
1582 |     num_output: 144
1583 |     bias_term: false
1584 |     pad: 1
1585 |     kernel_size: 3
1586 |     group: 144
1587 |     stride: 1
1588 |     weight_filler {
1589 |       type: "msra"
1590 |     }
1591 |     bias_filler {
1592 |       type: "constant"
1593 |       value: 0.0
1594 |     }
1595 |     engine: CAFFE
1596 |   }
1597 | }
1598 | layer {
1599 |   name: "bn_5_3_dw"
1600 |   type: "BatchNorm"
1601 |   bottom: "conv_5_3_dw"
1602 |   top: "conv_5_3_dw"
1603 |   param {
1604 |     lr_mult: 0.0
1605 |     decay_mult: 0.0
1606 |   }
1607 |   param {
1608 |     lr_mult: 0.0
1609 |     decay_mult: 0.0
1610 |   }
1611 |   param {
1612 |     lr_mult: 0.0
1613 |     decay_mult: 0.0
1614 |   }
1615 |   batch_norm_param {
1616 |     use_global_stats: true
1617 |   }
1618 | }
1619 | layer {
1620 |   name: "scale_5_3_dw"
1621 |   type: "Scale"
1622 |   bottom: "conv_5_3_dw"
1623 |   top: "conv_5_3_dw"
1624 |   scale_param {
1625 |     bias_term: true
1626 |   }
1627 | }
1628 | layer {
1629 |   name: "relu_5_3_dw"
1630 |   type: "ReLU"
1631 |   bottom: "conv_5_3_dw"
1632 |   top: "conv_5_3_dw"
1633 | }
1634 | layer {
1635 |   name: "conv_5_3_linear"
1636 |   type: "Convolution"
1637 |   bottom: "conv_5_3_dw"
1638 |   top: "conv_5_3_linear"
1639 |   param {
1640 |     lr_mult: 1.0
1641 |     decay_mult: 1.0
1642 |   }
1643 |   convolution_param {
1644 |     num_output: 24
1645 |     bias_term: false
1646 |     pad: 0
1647 |     kernel_size: 1
1648 |     stride: 1
1649 |     weight_filler {
1650 |       type: "msra"
1651 |     }
1652 |     bias_filler {
1653 |       type: "constant"
1654 |       value: 0.0
1655 |     }
1656 |   }
1657 | }
1658 | layer {
1659 |   name: "bn_5_3_linear"
1660 |   type: "BatchNorm"
1661 |   bottom: "conv_5_3_linear"
1662 |   top: "conv_5_3_linear"
1663 |   param {
1664 |     lr_mult: 0.0
1665 |     decay_mult: 0.0
1666 |   }
1667 |   param {
1668 |     lr_mult: 0.0
1669 |     decay_mult: 0.0
1670 |   }
1671 |   param {
1672 |     lr_mult: 0.0
1673 |     decay_mult: 0.0
1674 |   }
1675 |   batch_norm_param {
1676 |     use_global_stats: true
1677 |   }
1678 | }
1679 | layer {
1680 |   name: "scale_5_3_linear"
1681 |   type: "Scale"
1682 |   bottom: "conv_5_3_linear"
1683 |   top: "conv_5_3_linear"
1684 |   scale_param {
1685 |     bias_term: true
1686 |   }
1687 | }
1688 | layer {
1689 |   name: "add_5_3"
1690 |   type: "Eltwise"
1691 |   bottom: "add_5_2"
1692 |   bottom: "conv_5_3_linear"
1693 |   top: "add_5_3"
1694 | }
1695 | layer {
1696 |   name: "conv_5_4_pw"
1697 |   type: "Convolution"
1698 |   bottom: "add_5_3"
1699 |   top: "conv_5_4_pw"
1700 |   param {
1701 |     lr_mult: 1.0
1702 |     decay_mult: 1.0
1703 |   }
1704 |   convolution_param {
1705 |     num_output: 144
1706 |     bias_term: false
1707 |     pad: 0
1708 |     kernel_size: 1
1709 |     stride: 1
1710 |     weight_filler {
1711 |       type: "msra"
1712 |     }
1713 |     bias_filler {
1714 |       type: "constant"
1715 |       value: 0.0
1716 |     }
1717 |   }
1718 | }
1719 | layer {
1720 |   name: "bn_5_4_pw"
1721 |   type: "BatchNorm"
1722 |   bottom: "conv_5_4_pw"
1723 |   top: "conv_5_4_pw"
1724 |   param {
1725 |     lr_mult: 0.0
1726 |     decay_mult: 0.0
1727 |   }
1728 |   param {
1729 |     lr_mult: 0.0
1730 |     decay_mult: 0.0
1731 |   }
1732 |   param {
1733 |     lr_mult: 0.0
1734 |     decay_mult: 0.0
1735 |   }
1736 |   batch_norm_param {
1737 |     use_global_stats: true
1738 |   }
1739 | }
1740 | layer {
1741 |   name: "scale_5_4_pw"
1742 |   type: "Scale"
1743 |   bottom: "conv_5_4_pw"
1744 |   top: "conv_5_4_pw"
1745 |   scale_param {
1746 |     bias_term: true
1747 |   }
1748 | }
1749 | layer {
1750 |   name: "relu_5_4_pw"
1751 |   type: "ReLU"
1752 |   bottom: "conv_5_4_pw"
1753 |   top: "conv_5_4_pw"
1754 | }
1755 | layer {
1756 |   name: "conv_5_4_dw"
1757 |   type: "Convolution"
1758 |   bottom: "conv_5_4_pw"
1759 |   top: "conv_5_4_dw"
1760 |   param {
1761 |     lr_mult: 1.0
1762 |     decay_mult: 1.0
1763 |   }
1764 |   convolution_param {
1765 |     num_output: 144
1766 |     bias_term: false
1767 |     pad: 1
1768 |     kernel_size: 3
1769 |     group: 144
1770 |     stride: 1
1771 |     weight_filler {
1772 |       type: "msra"
1773 |     }
1774 |     bias_filler {
1775 |       type: "constant"
1776 |       value: 0.0
1777 |     }
1778 |     engine: CAFFE
1779 |   }
1780 | }
1781 | layer {
1782 |   name: "bn_5_4_dw"
1783 |   type: "BatchNorm"
1784 |   bottom: "conv_5_4_dw"
1785 |   top: "conv_5_4_dw"
1786 |   param {
1787 |     lr_mult: 0.0
1788 |     decay_mult: 0.0
1789 |   }
1790 |   param {
1791 |     lr_mult: 0.0
1792 |     decay_mult: 0.0
1793 |   }
1794 |   param {
1795 |     lr_mult: 0.0
1796 |     decay_mult: 0.0
1797 |   }
1798 |   batch_norm_param {
1799 |     use_global_stats: true
1800 |   }
1801 | }
1802 | layer {
1803 |   name: "scale_5_4_dw"
1804 |   type: "Scale"
1805 |   bottom: "conv_5_4_dw"
1806 |   top: "conv_5_4_dw"
1807 |   scale_param {
1808 |     bias_term: true
1809 |   }
1810 | }
1811 | layer {
1812 |   name: "relu_5_4_dw"
1813 |   type: "ReLU"
1814 |   bottom: "conv_5_4_dw"
1815 |   top: "conv_5_4_dw"
1816 | }
1817 | layer {
1818 |   name: "conv_5_4_linear"
1819 |   type: "Convolution"
1820 |   bottom: "conv_5_4_dw"
1821 |   top: "conv_5_4_linear"
1822 |   param {
1823 |     lr_mult: 1.0
1824 |     decay_mult: 1.0
1825 |   }
1826 |   convolution_param {
1827 |     num_output: 24
1828 |     bias_term: false
1829 |     pad: 0
1830 |     kernel_size: 1
1831 |     stride: 1
1832 |     weight_filler {
1833 |       type: "msra"
1834 |     }
1835 |     bias_filler {
1836 |       type: "constant"
1837 |       value: 0.0
1838 |     }
1839 |   }
1840 | }
1841 | layer {
1842 |   name: "bn_5_4_linear"
1843 |   type: "BatchNorm"
1844 |   bottom: "conv_5_4_linear"
1845 |   top: "conv_5_4_linear"
1846 |   param {
1847 |     lr_mult: 0.0
1848 |     decay_mult: 0.0
1849 |   }
1850 |   param {
1851 |     lr_mult: 0.0
1852 |     decay_mult: 0.0
1853 |   }
1854 |   param {
1855 |     lr_mult: 0.0
1856 |     decay_mult: 0.0
1857 |   }
1858 |   batch_norm_param {
1859 |     use_global_stats: true
1860 |   }
1861 | }
1862 | layer {
1863 |   name: "scale_5_4_linear"
1864 |   type: "Scale"
1865 |   bottom: "conv_5_4_linear"
1866 |   top: "conv_5_4_linear"
1867 |   scale_param {
1868 |     bias_term: true
1869 |   }
1870 | }
1871 | layer {
1872 |   name: "add_5_4"
1873 |   type: "Eltwise"
1874 |   bottom: "add_5_3"
1875 |   bottom: "conv_5_4_linear"
1876 |   top: "add_5_4"
1877 | }
1878 | layer {
1879 |   name: "conv_6_1_pw"
1880 |   type: "Convolution"
1881 |   bottom: "add_5_4"
1882 |   top: "conv_6_1_pw"
1883 |   param {
1884 |     lr_mult: 1.0
1885 |     decay_mult: 1.0
1886 |   }
1887 |   convolution_param {
1888 |     num_output: 144
1889 |     bias_term: false
1890 |     pad: 0
1891 |     kernel_size: 1
1892 |     stride: 1
1893 |     weight_filler {
1894 |       type: "msra"
1895 |     }
1896 |     bias_filler {
1897 |       type: "constant"
1898 |       value: 0.0
1899 |     }
1900 |   }
1901 | }
1902 | layer {
1903 |   name: "bn_6_1_pw"
1904 |   type: "BatchNorm"
1905 |   bottom: "conv_6_1_pw"
1906 |   top: "conv_6_1_pw"
1907 |   param {
1908 |     lr_mult: 0.0
1909 |     decay_mult: 0.0
1910 |   }
1911 |   param {
1912 |     lr_mult: 0.0
1913 |     decay_mult: 0.0
1914 |   }
1915 |   param {
1916 |     lr_mult: 0.0
1917 |     decay_mult: 0.0
1918 |   }
1919 |   batch_norm_param {
1920 |     use_global_stats: true
1921 |   }
1922 | }
1923 | layer {
1924 |   name: "scale_6_1_pw"
1925 |   type: "Scale"
1926 |   bottom: "conv_6_1_pw"
1927 |   top: "conv_6_1_pw"
1928 |   scale_param {
1929 |     bias_term: true
1930 |   }
1931 | }
1932 | layer {
1933 |   name: "relu_6_1_pw"
1934 |   type: "ReLU"
1935 |   bottom: "conv_6_1_pw"
1936 |   top: "conv_6_1_pw"
1937 | }
1938 | layer {
1939 |   name: "conv_6_1_dw"
1940 |   type: "Convolution"
1941 |   bottom: "conv_6_1_pw"
1942 |   top: "conv_6_1_dw"
1943 |   param {
1944 |     lr_mult: 1.0
1945 |     decay_mult: 1.0
1946 |   }
1947 |   convolution_param {
1948 |     num_output: 144
1949 |     bias_term: false
1950 |     pad: 1
1951 |     kernel_size: 3
1952 |     group: 144
1953 |     stride: 1
1954 |     weight_filler {
1955 |       type: "msra"
1956 |     }
1957 |     bias_filler {
1958 |       type: "constant"
1959 |       value: 0.0
1960 |     }
1961 |     engine: CAFFE
1962 |   }
1963 | }
1964 | layer {
1965 |   name: "bn_6_1_dw"
1966 |   type: "BatchNorm"
1967 |   bottom: "conv_6_1_dw"
1968 |   top: "conv_6_1_dw"
1969 |   param {
1970 |     lr_mult: 0.0
1971 |     decay_mult: 0.0
1972 |   }
1973 |   param {
1974 |     lr_mult: 0.0
1975 |     decay_mult: 0.0
1976 |   }
1977 |   param {
1978 |     lr_mult: 0.0
1979 |     decay_mult: 0.0
1980 |   }
1981 |   batch_norm_param {
1982 |     use_global_stats: true
1983 |   }
1984 | }
1985 | layer {
1986 |   name: "scale_6_1_dw"
1987 |   type: "Scale"
1988 |   bottom: "conv_6_1_dw"
1989 |   top: "conv_6_1_dw"
1990 |   scale_param {
1991 |     bias_term: true
1992 |   }
1993 | }
1994 | layer {
1995 |   name: "relu_6_1_dw"
1996 |   type: "ReLU"
1997 |   bottom: "conv_6_1_dw"
1998 |   top: "conv_6_1_dw"
1999 | }
2000 | layer {
2001 |   name: "conv_6_1_linear"
2002 |   type: "Convolution"
2003 |   bottom: "conv_6_1_dw"
2004 |   top: "conv_6_1_linear"
2005 |   param {
2006 |     lr_mult: 1.0
2007 |     decay_mult: 1.0
2008 |   }
2009 |   convolution_param {
2010 |     num_output: 32
2011 |     bias_term: false
2012 |     pad: 0
2013 |     kernel_size: 1
2014 |     stride: 1
2015 |     weight_filler {
2016 |       type: "msra"
2017 |     }
2018 |     bias_filler {
2019 |       type: "constant"
2020 |       value: 0.0
2021 |     }
2022 |   }
2023 | }
2024 | layer {
2025 |   name: "bn_6_1_linear"
2026 |   type: "BatchNorm"
2027 |   bottom: "conv_6_1_linear"
2028 |   top: "conv_6_1_linear"
2029 |   param {
2030 |     lr_mult: 0.0
2031 |     decay_mult: 0.0
2032 |   }
2033 |   param {
2034 |     lr_mult: 0.0
2035 |     decay_mult: 0.0
2036 |   }
2037 |   param {
2038 |     lr_mult: 0.0
2039 |     decay_mult: 0.0
2040 |   }
2041 |   batch_norm_param {
2042 |     use_global_stats: true
2043 |   }
2044 | }
2045 | layer {
2046 |   name: "scale_6_1_linear"
2047 |   type: "Scale"
2048 |   bottom: "conv_6_1_linear"
2049 |   top: "conv_6_1_linear"
2050 |   scale_param {
2051 |     bias_term: true
2052 |   }
2053 | }
2054 | layer {
2055 |   name: "conv_6_2_pw"
2056 |   type: "Convolution"
2057 |   bottom: "conv_6_1_linear"
2058 |   top: "conv_6_2_pw"
2059 |   param {
2060 |     lr_mult: 1.0
2061 |     decay_mult: 1.0
2062 |   }
2063 |   convolution_param {
2064 |     num_output: 192
2065 |     bias_term: false
2066 |     pad: 0
2067 |     kernel_size: 1
2068 |     stride: 1
2069 |     weight_filler {
2070 |       type: "msra"
2071 |     }
2072 |     bias_filler {
2073 |       type: "constant"
2074 |       value: 0.0
2075 |     }
2076 |   }
2077 | }
2078 | layer {
2079 |   name: "bn_6_2_pw"
2080 |   type: "BatchNorm"
2081 |   bottom: "conv_6_2_pw"
2082 |   top: "conv_6_2_pw"
2083 |   param {
2084 |     lr_mult: 0.0
2085 |     decay_mult: 0.0
2086 |   }
2087 |   param {
2088 |     lr_mult: 0.0
2089 |     decay_mult: 0.0
2090 |   }
2091 |   param {
2092 |     lr_mult: 0.0
2093 |     decay_mult: 0.0
2094 |   }
2095 |   batch_norm_param {
2096 |     use_global_stats: true
2097 |   }
2098 | }
2099 | layer {
2100 |   name: "scale_6_2_pw"
2101 |   type: "Scale"
2102 |   bottom: "conv_6_2_pw"
2103 |   top: "conv_6_2_pw"
2104 |   scale_param {
2105 |     bias_term: true
2106 |   }
2107 | }
2108 | layer {
2109 |   name: "relu_6_2_pw"
2110 |   type: "ReLU"
2111 |   bottom: "conv_6_2_pw"
2112 |   top: "conv_6_2_pw"
2113 | }
2114 | layer {
2115 |   name: "conv_6_2_dw"
2116 |   type: "Convolution"
2117 |   bottom: "conv_6_2_pw"
2118 |   top: "conv_6_2_dw"
2119 |   param {
2120 |     lr_mult: 1.0
2121 |     decay_mult: 1.0
2122 |   }
2123 |   convolution_param {
2124 |     num_output: 192
2125 |     bias_term: false
2126 |     pad: 1
2127 |     kernel_size: 3
2128 |     group: 192
2129 |     stride: 1
2130 |     weight_filler {
2131 |       type: "msra"
2132 |     }
2133 |     bias_filler {
2134 |       type: "constant"
2135 |       value: 0.0
2136 |     }
2137 |     engine: CAFFE
2138 |   }
2139 | }
2140 | layer {
2141 |   name: "bn_6_2_dw"
2142 |   type: "BatchNorm"
2143 |   bottom: "conv_6_2_dw"
2144 |   top: "conv_6_2_dw"
2145 |   param {
2146 |     lr_mult: 0.0
2147 |     decay_mult: 0.0
2148 |   }
2149 |   param {
2150 |     lr_mult: 0.0
2151 |     decay_mult: 0.0
2152 |   }
2153 |   param {
2154 |     lr_mult: 0.0
2155 |     decay_mult: 0.0
2156 |   }
2157 |   batch_norm_param {
2158 |     use_global_stats: true
2159 |   }
2160 | }
2161 | layer {
2162 |   name: "scale_6_2_dw"
2163 |   type: "Scale"
2164 |   bottom: "conv_6_2_dw"
2165 |   top: "conv_6_2_dw"
2166 |   scale_param {
2167 |     bias_term: true
2168 |   }
2169 | }
2170 | layer {
2171 |   name: "relu_6_2_dw"
2172 |   type: "ReLU"
2173 |   bottom: "conv_6_2_dw"
2174 |   top: "conv_6_2_dw"
2175 | }
2176 | layer {
2177 |   name: "conv_6_2_linear"
2178 |   type: "Convolution"
2179 |   bottom: "conv_6_2_dw"
2180 |   top: "conv_6_2_linear"
2181 |   param {
2182 |     lr_mult: 1.0
2183 |     decay_mult: 1.0
2184 |   }
2185 |   convolution_param {
2186 |     num_output: 32
2187 |     bias_term: false
2188 |     pad: 0
2189 |     kernel_size: 1
2190 |     stride: 1
2191 |     weight_filler {
2192 |       type: "msra"
2193 |     }
2194 |     bias_filler {
2195 |       type: "constant"
2196 |       value: 0.0
2197 |     }
2198 |   }
2199 | }
2200 | layer {
2201 |   name: "bn_6_2_linear"
2202 |   type: "BatchNorm"
2203 |   bottom: "conv_6_2_linear"
2204 |   top: "conv_6_2_linear"
2205 |   param {
2206 |     lr_mult: 0.0
2207 |     decay_mult: 0.0
2208 |   }
2209 |   param {
2210 |     lr_mult: 0.0
2211 |     decay_mult: 0.0
2212 |   }
2213 |   param {
2214 |     lr_mult: 0.0
2215 |     decay_mult: 0.0
2216 |   }
2217 |   batch_norm_param {
2218 |     use_global_stats: true
2219 |   }
2220 | }
2221 | layer {
2222 |   name: "scale_6_2_linear"
2223 |   type: "Scale"
2224 |   bottom: "conv_6_2_linear"
2225 |   top: "conv_6_2_linear"
2226 |   scale_param {
2227 |     bias_term: true
2228 |   }
2229 | }
2230 | layer {
2231 |   name: "add_6_2"
2232 |   type: "Eltwise"
2233 |   bottom: "conv_6_1_linear"
2234 |   bottom: "conv_6_2_linear"
2235 |   top: "add_6_2"
2236 | }
2237 | layer {
2238 |   name: "conv_6_3_pw"
2239 |   type: "Convolution"
2240 |   bottom: "add_6_2"
2241 |   top: "conv_6_3_pw"
2242 |   param {
2243 |     lr_mult: 1.0
2244 |     decay_mult: 1.0
2245 |   }
2246 |   convolution_param {
2247 |     num_output: 192
2248 |     bias_term: false
2249 |     pad: 0
2250 |     kernel_size: 1
2251 |     stride: 1
2252 |     weight_filler {
2253 |       type: "msra"
2254 |     }
2255 |     bias_filler {
2256 |       type: "constant"
2257 |       value: 0.0
2258 |     }
2259 |   }
2260 | }
2261 | layer {
2262 |   name: "bn_6_3_pw"
2263 |   type: "BatchNorm"
2264 |   bottom: "conv_6_3_pw"
2265 |   top: "conv_6_3_pw"
2266 |   param {
2267 |     lr_mult: 0.0
2268 |     decay_mult: 0.0
2269 |   }
2270 |   param {
2271 |     lr_mult: 0.0
2272 |     decay_mult: 0.0
2273 |   }
2274 |   param {
2275 |     lr_mult: 0.0
2276 |     decay_mult: 0.0
2277 |   }
2278 |   batch_norm_param {
2279 |     use_global_stats: true
2280 |   }
2281 | }
2282 | layer {
2283 |   name: "scale_6_3_pw"
2284 |   type: "Scale"
2285 |   bottom: "conv_6_3_pw"
2286 |   top: "conv_6_3_pw"
2287 |   scale_param {
2288 |     bias_term: true
2289 |   }
2290 | }
2291 | layer {
2292 |   name: "relu_6_3_pw"
2293 |   type: "ReLU"
2294 |   bottom: "conv_6_3_pw"
2295 |   top: "conv_6_3_pw"
2296 | }
2297 | layer {
2298 |   name: "conv_6_3_dw"
2299 |   type: "Convolution"
2300 |   bottom: "conv_6_3_pw"
2301 |   top: "conv_6_3_dw"
2302 |   param {
2303 |     lr_mult: 1.0
2304 |     decay_mult: 1.0
2305 |   }
2306 |   convolution_param {
2307 |     num_output: 192
2308 |     bias_term: false
2309 |     pad: 1
2310 |     kernel_size: 3
2311 |     group: 192
2312 |     stride: 1
2313 |     weight_filler {
2314 |       type: "msra"
2315 |     }
2316 |     bias_filler {
2317 |       type: "constant"
2318 |       value: 0.0
2319 |     }
2320 |     engine: CAFFE
2321 |   }
2322 | }
2323 | layer {
2324 |   name: "bn_6_3_dw"
2325 |   type: "BatchNorm"
2326 |   bottom: "conv_6_3_dw"
2327 |   top: "conv_6_3_dw"
2328 |   param {
2329 |     lr_mult: 0.0
2330 |     decay_mult: 0.0
2331 |   }
2332 |   param {
2333 |     lr_mult: 0.0
2334 |     decay_mult: 0.0
2335 |   }
2336 |   param {
2337 |     lr_mult: 0.0
2338 |     decay_mult: 0.0
2339 |   }
2340 |   batch_norm_param {
2341 |     use_global_stats: true
2342 |   }
2343 | }
2344 | layer {
2345 |   name: "scale_6_3_dw"
2346 |   type: "Scale"
2347 |   bottom: "conv_6_3_dw"
2348 |   top: "conv_6_3_dw"
2349 |   scale_param {
2350 |     bias_term: true
2351 |   }
2352 | }
2353 | layer {
2354 |   name: "relu_6_3_dw"
2355 |   type: "ReLU"
2356 |   bottom: "conv_6_3_dw"
2357 |   top: "conv_6_3_dw"
2358 | }
2359 | layer {
2360 |   name: "conv_6_3_linear"
2361 |   type: "Convolution"
2362 |   bottom: "conv_6_3_dw"
2363 |   top: "conv_6_3_linear"
2364 |   param {
2365 |     lr_mult: 1.0
2366 |     decay_mult: 1.0
2367 |   }
2368 |   convolution_param {
2369 |     num_output: 32
2370 |     bias_term: false
2371 |     pad: 0
2372 |     kernel_size: 1
2373 |     stride: 1
2374 |     weight_filler {
2375 |       type: "msra"
2376 |     }
2377 |     bias_filler {
2378 |       type: "constant"
2379 |       value: 0.0
2380 |     }
2381 |   }
2382 | }
2383 | layer {
2384 |   name: "bn_6_3_linear"
2385 |   type: "BatchNorm"
2386 |   bottom: "conv_6_3_linear"
2387 |   top: "conv_6_3_linear"
2388 |   param {
2389 |     lr_mult: 0.0
2390 |     decay_mult: 0.0
2391 |   }
2392 |   param {
2393 |     lr_mult: 0.0
2394 |     decay_mult: 0.0
2395 |   }
2396 |   param {
2397 |     lr_mult: 0.0
2398 |     decay_mult: 0.0
2399 |   }
2400 |   batch_norm_param {
2401 |     use_global_stats: true
2402 |   }
2403 | }
2404 | layer {
2405 |   name: "scale_6_3_linear"
2406 |   type: "Scale"
2407 |   bottom: "conv_6_3_linear"
2408 |   top: "conv_6_3_linear"
2409 |   scale_param {
2410 |     bias_term: true
2411 |   }
2412 | }
2413 | layer {
2414 |   name: "add_6_3"
2415 |   type: "Eltwise"
2416 |   bottom: "add_6_2"
2417 |   bottom: "conv_6_3_linear"
2418 |   top: "add_6_3"
2419 | }
2420 | layer {
2421 |   name: "conv_7_1_pw"
2422 |   type: "Convolution"
2423 |   bottom: "add_6_3"
2424 |   top: "conv_7_1_pw"
2425 |   param {
2426 |     lr_mult: 1.0
2427 |     decay_mult: 1.0
2428 |   }
2429 |   convolution_param {
2430 |     num_output: 192
2431 |     bias_term: false
2432 |     pad: 0
2433 |     kernel_size: 1
2434 |     stride: 1
2435 |     weight_filler {
2436 |       type: "msra"
2437 |     }
2438 |     bias_filler {
2439 |       type: "constant"
2440 |       value: 0.0
2441 |     }
2442 |   }
2443 | }
2444 | layer {
2445 |   name: "bn_7_1_pw"
2446 |   type: "BatchNorm"
2447 |   bottom: "conv_7_1_pw"
2448 |   top: "conv_7_1_pw"
2449 |   param {
2450 |     lr_mult: 0.0
2451 |     decay_mult: 0.0
2452 |   }
2453 |   param {
2454 |     lr_mult: 0.0
2455 |     decay_mult: 0.0
2456 |   }
2457 |   param {
2458 |     lr_mult: 0.0
2459 |     decay_mult: 0.0
2460 |   }
2461 |   batch_norm_param {
2462 |     use_global_stats: true
2463 |   }
2464 | }
2465 | layer {
2466 |   name: "scale_7_1_pw"
2467 |   type: "Scale"
2468 |   bottom: "conv_7_1_pw"
2469 |   top: "conv_7_1_pw"
2470 |   scale_param {
2471 |     bias_term: true
2472 |   }
2473 | }
2474 | layer {
2475 |   name: "relu_7_1_pw"
2476 |   type: "ReLU"
2477 |   bottom: "conv_7_1_pw"
2478 |   top: "conv_7_1_pw"
2479 | }
2480 | layer {
2481 |   name: "conv_7_1_dw"
2482 |   type: "Convolution"
2483 |   bottom: "conv_7_1_pw"
2484 |   top: "conv_7_1_dw"
2485 |   param {
2486 |     lr_mult: 1.0
2487 |     decay_mult: 1.0
2488 |   }
2489 |   convolution_param {
2490 |     num_output: 192
2491 |     bias_term: false
2492 |     pad: 1
2493 |     kernel_size: 3
2494 |     group: 192
2495 |     stride: 2
2496 |     weight_filler {
2497 |       type: "msra"
2498 |     }
2499 |     bias_filler {
2500 |       type: "constant"
2501 |       value: 0.0
2502 |     }
2503 |     engine: CAFFE
2504 |   }
2505 | }
2506 | layer {
2507 |   name: "bn_7_1_dw"
2508 |   type: "BatchNorm"
2509 |   bottom: "conv_7_1_dw"
2510 |   top: "conv_7_1_dw"
2511 |   param {
2512 |     lr_mult: 0.0
2513 |     decay_mult: 0.0
2514 |   }
2515 |   param {
2516 |     lr_mult: 0.0
2517 |     decay_mult: 0.0
2518 |   }
2519 |   param {
2520 |     lr_mult: 0.0
2521 |     decay_mult: 0.0
2522 |   }
2523 |   batch_norm_param {
2524 |     use_global_stats: true
2525 |   }
2526 | }
2527 | layer {
2528 |   name: "scale_7_1_dw"
2529 |   type: "Scale"
2530 |   bottom: "conv_7_1_dw"
2531 |   top: "conv_7_1_dw"
2532 |   scale_param {
2533 |     bias_term: true
2534 |   }
2535 | }
2536 | layer {
2537 |   name: "relu_7_1_dw"
2538 |   type: "ReLU"
2539 |   bottom: "conv_7_1_dw"
2540 |   top: "conv_7_1_dw"
2541 | }
2542 | layer {
2543 |   name: "conv_7_1_linear"
2544 |   type: "Convolution"
2545 |   bottom: "conv_7_1_dw"
2546 |   top: "conv_7_1_linear"
2547 |   param {
2548 |     lr_mult: 1.0
2549 |     decay_mult: 1.0
2550 |   }
2551 |   convolution_param {
2552 |     num_output: 56
2553 |     bias_term: false
2554 |     pad: 0
2555 |     kernel_size: 1
2556 |     stride: 1
2557 |     weight_filler {
2558 |       type: "msra"
2559 |     }
2560 |     bias_filler {
2561 |       type: "constant"
2562 |       value: 0.0
2563 |     }
2564 |   }
2565 | }
2566 | layer {
2567 |   name: "bn_7_1_linear"
2568 |   type: "BatchNorm"
2569 |   bottom: "conv_7_1_linear"
2570 |   top: "conv_7_1_linear"
2571 |   param {
2572 |     lr_mult: 0.0
2573 |     decay_mult: 0.0
2574 |   }
2575 |   param {
2576 |     lr_mult: 0.0
2577 |     decay_mult: 0.0
2578 |   }
2579 |   param {
2580 |     lr_mult: 0.0
2581 |     decay_mult: 0.0
2582 |   }
2583 |   batch_norm_param {
2584 |     use_global_stats: true
2585 |   }
2586 | }
2587 | layer {
2588 |   name: "scale_7_1_linear"
2589 |   type: "Scale"
2590 |   bottom: "conv_7_1_linear"
2591 |   top: "conv_7_1_linear"
2592 |   scale_param {
2593 |     bias_term: true
2594 |   }
2595 | }
2596 | layer {
2597 |   name: "conv_7_2_pw"
2598 |   type: "Convolution"
2599 |   bottom: "conv_7_1_linear"
2600 |   top: "conv_7_2_pw"
2601 |   param {
2602 |     lr_mult: 1.0
2603 |     decay_mult: 1.0
2604 |   }
2605 |   convolution_param {
2606 |     num_output: 336
2607 |     bias_term: false
2608 |     pad: 0
2609 |     kernel_size: 1
2610 |     stride: 1
2611 |     weight_filler {
2612 |       type: "msra"
2613 |     }
2614 |     bias_filler {
2615 |       type: "constant"
2616 |       value: 0.0
2617 |     }
2618 |   }
2619 | }
2620 | layer {
2621 |   name: "bn_7_2_pw"
2622 |   type: "BatchNorm"
2623 |   bottom: "conv_7_2_pw"
2624 |   top: "conv_7_2_pw"
2625 |   param {
2626 |     lr_mult: 0.0
2627 |     decay_mult: 0.0
2628 |   }
2629 |   param {
2630 |     lr_mult: 0.0
2631 |     decay_mult: 0.0
2632 |   }
2633 |   param {
2634 |     lr_mult: 0.0
2635 |     decay_mult: 0.0
2636 |   }
2637 |   batch_norm_param {
2638 |     use_global_stats: true
2639 |   }
2640 | }
2641 | layer {
2642 |   name: "scale_7_2_pw"
2643 |   type: "Scale"
2644 |   bottom: "conv_7_2_pw"
2645 |   top: "conv_7_2_pw"
2646 |   scale_param {
2647 |     bias_term: true
2648 |   }
2649 | }
2650 | layer {
2651 |   name: "relu_7_2_pw"
2652 |   type: "ReLU"
2653 |   bottom: "conv_7_2_pw"
2654 |   top: "conv_7_2_pw"
2655 | }
2656 | layer {
2657 |   name: "conv_7_2_dw"
2658 |   type: "Convolution"
2659 |   bottom: "conv_7_2_pw"
2660 |   top: "conv_7_2_dw"
2661 |   param {
2662 |     lr_mult: 1.0
2663 |     decay_mult: 1.0
2664 |   }
2665 |   convolution_param {
2666 |     num_output: 336
2667 |     bias_term: false
2668 |     pad: 1
2669 |     kernel_size: 3
2670 |     group: 336
2671 |     stride: 1
2672 |     weight_filler {
2673 |       type: "msra"
2674 |     }
2675 |     bias_filler {
2676 |       type: "constant"
2677 |       value: 0.0
2678 |     }
2679 |     engine: CAFFE
2680 |   }
2681 | }
2682 | layer {
2683 |   name: "bn_7_2_dw"
2684 |   type: "BatchNorm"
2685 |   bottom: "conv_7_2_dw"
2686 |   top: "conv_7_2_dw"
2687 |   param {
2688 |     lr_mult: 0.0
2689 |     decay_mult: 0.0
2690 |   }
2691 |   param {
2692 |     lr_mult: 0.0
2693 |     decay_mult: 0.0
2694 |   }
2695 |   param {
2696 |     lr_mult: 0.0
2697 |     decay_mult: 0.0
2698 |   }
2699 |   batch_norm_param {
2700 |     use_global_stats: true
2701 |   }
2702 | }
2703 | layer {
2704 |   name: "scale_7_2_dw"
2705 |   type: "Scale"
2706 |   bottom: "conv_7_2_dw"
2707 |   top: "conv_7_2_dw"
2708 |   scale_param {
2709 |     bias_term: true
2710 |   }
2711 | }
2712 | layer {
2713 |   name: "relu_7_2_dw"
2714 |   type: "ReLU"
2715 |   bottom: "conv_7_2_dw"
2716 |   top: "conv_7_2_dw"
2717 | }
2718 | layer {
2719 |   name: "conv_7_2_linear"
2720 |   type: "Convolution"
2721 |   bottom: "conv_7_2_dw"
2722 |   top: "conv_7_2_linear"
2723 |   param {
2724 |     lr_mult: 1.0
2725 |     decay_mult: 1.0
2726 |   }
2727 |   convolution_param {
2728 |     num_output: 56
2729 |     bias_term: false
2730 |     pad: 0
2731 |     kernel_size: 1
2732 |     stride: 1
2733 |     weight_filler {
2734 |       type: "msra"
2735 |     }
2736 |     bias_filler {
2737 |       type: "constant"
2738 |       value: 0.0
2739 |     }
2740 |   }
2741 | }
2742 | layer {
2743 |   name: "bn_7_2_linear"
2744 |   type: "BatchNorm"
2745 |   bottom: "conv_7_2_linear"
2746 |   top: "conv_7_2_linear"
2747 |   param {
2748 |     lr_mult: 0.0
2749 |     decay_mult: 0.0
2750 |   }
2751 |   param {
2752 |     lr_mult: 0.0
2753 |     decay_mult: 0.0
2754 |   }
2755 |   param {
2756 |     lr_mult: 0.0
2757 |     decay_mult: 0.0
2758 |   }
2759 |   batch_norm_param {
2760 |     use_global_stats: true
2761 |   }
2762 | }
2763 | layer {
2764 |   name: "scale_7_2_linear"
2765 |   type: "Scale"
2766 |   bottom: "conv_7_2_linear"
2767 |   top: "conv_7_2_linear"
2768 |   scale_param {
2769 |     bias_term: true
2770 |   }
2771 | }
2772 | layer {
2773 |   name: "add_7_2"
2774 |   type: "Eltwise"
2775 |   bottom: "conv_7_1_linear"
2776 |   bottom: "conv_7_2_linear"
2777 |   top: "add_7_2"
2778 | }
2779 | layer {
2780 |   name: "conv_7_3_pw"
2781 |   type: "Convolution"
2782 |   bottom: "add_7_2"
2783 |   top: "conv_7_3_pw"
2784 |   param {
2785 |     lr_mult: 1.0
2786 |     decay_mult: 1.0
2787 |   }
2788 |   convolution_param {
2789 |     num_output: 336
2790 |     bias_term: false
2791 |     pad: 0
2792 |     kernel_size: 1
2793 |     stride: 1
2794 |     weight_filler {
2795 |       type: "msra"
2796 |     }
2797 |     bias_filler {
2798 |       type: "constant"
2799 |       value: 0.0
2800 |     }
2801 |   }
2802 | }
2803 | layer {
2804 |   name: "bn_7_3_pw"
2805 |   type: "BatchNorm"
2806 |   bottom: "conv_7_3_pw"
2807 |   top: "conv_7_3_pw"
2808 |   param {
2809 |     lr_mult: 0.0
2810 |     decay_mult: 0.0
2811 |   }
2812 |   param {
2813 |     lr_mult: 0.0
2814 |     decay_mult: 0.0
2815 |   }
2816 |   param {
2817 |     lr_mult: 0.0
2818 |     decay_mult: 0.0
2819 |   }
2820 |   batch_norm_param {
2821 |     use_global_stats: true
2822 |   }
2823 | }
2824 | layer {
2825 |   name: "scale_7_3_pw"
2826 |   type: "Scale"
2827 |   bottom: "conv_7_3_pw"
2828 |   top: "conv_7_3_pw"
2829 |   scale_param {
2830 |     bias_term: true
2831 |   }
2832 | }
2833 | layer {
2834 |   name: "relu_7_3_pw"
2835 |   type: "ReLU"
2836 |   bottom: "conv_7_3_pw"
2837 |   top: "conv_7_3_pw"
2838 | }
2839 | layer {
2840 |   name: "conv_7_3_dw"
2841 |   type: "Convolution"
2842 |   bottom: "conv_7_3_pw"
2843 |   top: "conv_7_3_dw"
2844 |   param {
2845 |     lr_mult: 1.0
2846 |     decay_mult: 1.0
2847 |   }
2848 |   convolution_param {
2849 |     num_output: 336
2850 |     bias_term: false
2851 |     pad: 1
2852 |     kernel_size: 3
2853 |     group: 336
2854 |     stride: 1
2855 |     weight_filler {
2856 |       type: "msra"
2857 |     }
2858 |     bias_filler {
2859 |       type: "constant"
2860 |       value: 0.0
2861 |     }
2862 |     engine: CAFFE
2863 |   }
2864 | }
2865 | layer {
2866 |   name: "bn_7_3_dw"
2867 |   type: "BatchNorm"
2868 |   bottom: "conv_7_3_dw"
2869 |   top: "conv_7_3_dw"
2870 |   param {
2871 |     lr_mult: 0.0
2872 |     decay_mult: 0.0
2873 |   }
2874 |   param {
2875 |     lr_mult: 0.0
2876 |     decay_mult: 0.0
2877 |   }
2878 |   param {
2879 |     lr_mult: 0.0
2880 |     decay_mult: 0.0
2881 |   }
2882 |   batch_norm_param {
2883 |     use_global_stats: true
2884 |   }
2885 | }
2886 | layer {
2887 |   name: "scale_7_3_dw"
2888 |   type: "Scale"
2889 |   bottom: "conv_7_3_dw"
2890 |   top: "conv_7_3_dw"
2891 |   scale_param {
2892 |     bias_term: true
2893 |   }
2894 | }
2895 | layer {
2896 |   name: "relu_7_3_dw"
2897 |   type: "ReLU"
2898 |   bottom: "conv_7_3_dw"
2899 |   top: "conv_7_3_dw"
2900 | }
2901 | layer {
2902 |   name: "conv_7_3_linear"
2903 |   type: "Convolution"
2904 |   bottom: "conv_7_3_dw"
2905 |   top: "conv_7_3_linear"
2906 |   param {
2907 |     lr_mult: 1.0
2908 |     decay_mult: 1.0
2909 |   }
2910 |   convolution_param {
2911 |     num_output: 56
2912 |     bias_term: false
2913 |     pad: 0
2914 |     kernel_size: 1
2915 |     stride: 1
2916 |     weight_filler {
2917 |       type: "msra"
2918 |     }
2919 |     bias_filler {
2920 |       type: "constant"
2921 |       value: 0.0
2922 |     }
2923 |   }
2924 | }
2925 | layer {
2926 |   name: "bn_7_3_linear"
2927 |   type: "BatchNorm"
2928 |   bottom: "conv_7_3_linear"
2929 |   top: "conv_7_3_linear"
2930 |   param {
2931 |     lr_mult: 0.0
2932 |     decay_mult: 0.0
2933 |   }
2934 |   param {
2935 |     lr_mult: 0.0
2936 |     decay_mult: 0.0
2937 |   }
2938 |   param {
2939 |     lr_mult: 0.0
2940 |     decay_mult: 0.0
2941 |   }
2942 |   batch_norm_param {
2943 |     use_global_stats: true
2944 |   }
2945 | }
2946 | layer {
2947 |   name: "scale_7_3_linear"
2948 |   type: "Scale"
2949 |   bottom: "conv_7_3_linear"
2950 |   top: "conv_7_3_linear"
2951 |   scale_param {
2952 |     bias_term: true
2953 |   }
2954 | }
2955 | layer {
2956 |   name: "add_7_3"
2957 |   type: "Eltwise"
2958 |   bottom: "add_7_2"
2959 |   bottom: "conv_7_3_linear"
2960 |   top: "add_7_3"
2961 | }
2962 | layer {
2963 |   name: "conv_8_1_pw"
2964 |   type: "Convolution"
2965 |   bottom: "add_7_3"
2966 |   top: "conv_8_1_pw"
2967 |   param {
2968 |     lr_mult: 1.0
2969 |     decay_mult: 1.0
2970 |   }
2971 |   convolution_param {
2972 |     num_output: 336
2973 |     bias_term: false
2974 |     pad: 0
2975 |     kernel_size: 1
2976 |     stride: 1
2977 |     weight_filler {
2978 |       type: "msra"
2979 |     }
2980 |     bias_filler {
2981 |       type: "constant"
2982 |       value: 0.0
2983 |     }
2984 |   }
2985 | }
2986 | layer {
2987 |   name: "bn_8_1_pw"
2988 |   type: "BatchNorm"
2989 |   bottom: "conv_8_1_pw"
2990 |   top: "conv_8_1_pw"
2991 |   param {
2992 |     lr_mult: 0.0
2993 |     decay_mult: 0.0
2994 |   }
2995 |   param {
2996 |     lr_mult: 0.0
2997 |     decay_mult: 0.0
2998 |   }
2999 |   param {
3000 |     lr_mult: 0.0
3001 |     decay_mult: 0.0
3002 |   }
3003 |   batch_norm_param {
3004 |     use_global_stats: true
3005 |   }
3006 | }
3007 | layer {
3008 |   name: "scale_8_1_pw"
3009 |   type: "Scale"
3010 |   bottom: "conv_8_1_pw"
3011 |   top: "conv_8_1_pw"
3012 |   scale_param {
3013 |     bias_term: true
3014 |   }
3015 | }
3016 | layer {
3017 |   name: "relu_8_1_pw"
3018 |   type: "ReLU"
3019 |   bottom: "conv_8_1_pw"
3020 |   top: "conv_8_1_pw"
3021 | }
3022 | layer {
3023 |   name: "conv_8_1_dw"
3024 |   type: "Convolution"
3025 |   bottom: "conv_8_1_pw"
3026 |   top: "conv_8_1_dw"
3027 |   param {
3028 |     lr_mult: 1.0
3029 |     decay_mult: 1.0
3030 |   }
3031 |   convolution_param {
3032 |     num_output: 336
3033 |     bias_term: false
3034 |     pad: 1
3035 |     kernel_size: 3
3036 |     group: 336
3037 |     stride: 1
3038 |     weight_filler {
3039 |       type: "msra"
3040 |     }
3041 |     bias_filler {
3042 |       type: "constant"
3043 |       value: 0.0
3044 |     }
3045 |     engine: CAFFE
3046 |   }
3047 | }
3048 | layer {
3049 |   name: "bn_8_1_dw"
3050 |   type: "BatchNorm"
3051 |   bottom: "conv_8_1_dw"
3052 |   top: "conv_8_1_dw"
3053 |   param {
3054 |     lr_mult: 0.0
3055 |     decay_mult: 0.0
3056 |   }
3057 |   param {
3058 |     lr_mult: 0.0
3059 |     decay_mult: 0.0
3060 |   }
3061 |   param {
3062 |     lr_mult: 0.0
3063 |     decay_mult: 0.0
3064 |   }
3065 |   batch_norm_param {
3066 |     use_global_stats: true
3067 |   }
3068 | }
3069 | layer {
3070 |   name: "scale_8_1_dw"
3071 |   type: "Scale"
3072 |   bottom: "conv_8_1_dw"
3073 |   top: "conv_8_1_dw"
3074 |   scale_param {
3075 |     bias_term: true
3076 |   }
3077 | }
3078 | layer {
3079 |   name: "relu_8_1_dw"
3080 |   type: "ReLU"
3081 |   bottom: "conv_8_1_dw"
3082 |   top: "conv_8_1_dw"
3083 | }
3084 | layer {
3085 |   name: "conv_8_1_linear"
3086 |   type: "Convolution"
3087 |   bottom: "conv_8_1_dw"
3088 |   top: "conv_8_1_linear"
3089 |   param {
3090 |     lr_mult: 1.0
3091 |     decay_mult: 1.0
3092 |   }
3093 |   convolution_param {
3094 |     num_output: 112
3095 |     bias_term: false
3096 |     pad: 0
3097 |     kernel_size: 1
3098 |     stride: 1
3099 |     weight_filler {
3100 |       type: "msra"
3101 |     }
3102 |     bias_filler {
3103 |       type: "constant"
3104 |       value: 0.0
3105 |     }
3106 |   }
3107 | }
3108 | layer {
3109 |   name: "bn_8_1_linear"
3110 |   type: "BatchNorm"
3111 |   bottom: "conv_8_1_linear"
3112 |   top: "conv_8_1_linear"
3113 |   param {
3114 |     lr_mult: 0.0
3115 |     decay_mult: 0.0
3116 |   }
3117 |   param {
3118 |     lr_mult: 0.0
3119 |     decay_mult: 0.0
3120 |   }
3121 |   param {
3122 |     lr_mult: 0.0
3123 |     decay_mult: 0.0
3124 |   }
3125 |   batch_norm_param {
3126 |     use_global_stats: true
3127 |   }
3128 | }
3129 | layer {
3130 |   name: "scale_8_1_linear"
3131 |   type: "Scale"
3132 |   bottom: "conv_8_1_linear"
3133 |   top: "conv_8_1_linear"
3134 |   scale_param {
3135 |     bias_term: true
3136 |   }
3137 | }
3138 | layer {
3139 |   name: "conv_conv9"
3140 |   type: "Convolution"
3141 |   bottom: "conv_8_1_linear"
3142 |   top: "conv_conv9"
3143 |   param {
3144 |     lr_mult: 1.0
3145 |     decay_mult: 1.0
3146 |   }
3147 |   convolution_param {
3148 |     num_output: 1280
3149 |     bias_term: true
3150 |     pad: 0
3151 |     kernel_size: 1
3152 |     stride: 1
3153 |     weight_filler {
3154 |       type: "msra"
3155 |     }
3156 |     bias_filler {
3157 |       type: "constant"
3158 |       value: 0.0
3159 |     }
3160 |   }
3161 | }
3162 | layer {
3163 |   name: "bn_conv9"
3164 |   type: "BatchNorm"
3165 |   bottom: "conv_conv9"
3166 |   top: "conv_conv9"
3167 |   param {
3168 |     lr_mult: 0.0
3169 |     decay_mult: 0.0
3170 |   }
3171 |   param {
3172 |     lr_mult: 0.0
3173 |     decay_mult: 0.0
3174 |   }
3175 |   param {
3176 |     lr_mult: 0.0
3177 |     decay_mult: 0.0
3178 |   }
3179 |   batch_norm_param {
3180 |     use_global_stats: true
3181 |   }
3182 | }
3183 | layer {
3184 |   name: "scale_conv9"
3185 |   type: "Scale"
3186 |   bottom: "conv_conv9"
3187 |   top: "conv_conv9"
3188 |   scale_param {
3189 |     bias_term: true
3190 |   }
3191 | }
3192 | layer {
3193 |   name: "relu_conv9"
3194 |   type: "ReLU"
3195 |   bottom: "conv_conv9"
3196 |   top: "conv_conv9"
3197 | }
3198 | layer {
3199 |   name: "pool9"
3200 |   type: "Pooling"
3201 |   bottom: "conv_conv9"
3202 |   top: "pool9"
3203 |   pooling_param {
3204 |     pool: AVE
3205 |     stride: 1
3206 |     global_pooling: true
3207 |   }
3208 | }
3209 | layer {
3210 |   name: "conv10"
3211 |   type: "Convolution"
3212 |   bottom: "pool9"
3213 |   top: "conv10"
3214 |   param {
3215 |     lr_mult: 1.0
3216 |     decay_mult: 1.0
3217 |   }
3218 |   convolution_param {
3219 |     num_output: 1000
3220 |     pad: 0
3221 |     kernel_size: 1
3222 |     stride: 1
3223 |     weight_filler {
3224 |       type: "msra"
3225 |     }
3226 |     bias_filler {
3227 |       type: "constant"
3228 |       value: 0.0
3229 |     }
3230 |   }
3231 | }
3232 | layer {
3233 |   name: "loss"
3234 |   type: "SoftmaxWithLoss"
3235 |   bottom: "conv10"
3236 |   bottom: "label"
3237 |   top: "loss"
3238 | }
3239 | 
3240 | 


--------------------------------------------------------------------------------
/mobilenetv2-0.5/test.prototxt:
--------------------------------------------------------------------------------
   1 | layer {
   2 |   name: "data"
   3 |   type: "MemoryData"
   4 |   top: "data"
   5 |   top: "label"
   6 |   transform_param {
   7 |     crop_size: 224
   8 |   }
   9 |   memory_data_param {
  10 |     batch_size: 1
  11 |     channels: 3
  12 |     height: 224
  13 |     width: 224
  14 |   }
  15 | }
  16 | layer {
  17 |   name: "conv_1_1"
  18 |   type: "Convolution"
  19 |   bottom: "data"
  20 |   top: "conv_1_1"
  21 |   param {
  22 |     lr_mult: 1.0
  23 |     decay_mult: 1.0
  24 |   }
  25 |   convolution_param {
  26 |     num_output: 16
  27 |     bias_term: true
  28 |     pad: 1
  29 |     kernel_size: 3
  30 |     stride: 2
  31 |     weight_filler {
  32 |       type: "msra"
  33 |     }
  34 |     bias_filler {
  35 |       type: "constant"
  36 |       value: 0.0
  37 |     }
  38 |   }
  39 | }
  40 | layer {
  41 |   name: "bn_1_1"
  42 |   type: "BatchNorm"
  43 |   bottom: "conv_1_1"
  44 |   top: "conv_1_1"
  45 |   param {
  46 |     lr_mult: 0.0
  47 |     decay_mult: 0.0
  48 |   }
  49 |   param {
  50 |     lr_mult: 0.0
  51 |     decay_mult: 0.0
  52 |   }
  53 |   param {
  54 |     lr_mult: 0.0
  55 |     decay_mult: 0.0
  56 |   }
  57 |   batch_norm_param {
  58 |     use_global_stats: true
  59 |   }
  60 | }
  61 | layer {
  62 |   name: "scale_1_1"
  63 |   type: "Scale"
  64 |   bottom: "conv_1_1"
  65 |   top: "conv_1_1"
  66 |   scale_param {
  67 |     bias_term: true
  68 |   }
  69 | }
  70 | layer {
  71 |   name: "relu_1_1"
  72 |   type: "ReLU"
  73 |   bottom: "conv_1_1"
  74 |   top: "conv_1_1"
  75 | }
  76 | layer {
  77 |   name: "conv_2_1_pw"
  78 |   type: "Convolution"
  79 |   bottom: "conv_1_1"
  80 |   top: "conv_2_1_pw"
  81 |   param {
  82 |     lr_mult: 1.0
  83 |     decay_mult: 1.0
  84 |   }
  85 |   convolution_param {
  86 |     num_output: 16
  87 |     bias_term: false
  88 |     pad: 0
  89 |     kernel_size: 1
  90 |     stride: 1
  91 |     weight_filler {
  92 |       type: "msra"
  93 |     }
  94 |     bias_filler {
  95 |       type: "constant"
  96 |       value: 0.0
  97 |     }
  98 |   }
  99 | }
 100 | layer {
 101 |   name: "bn_2_1_pw"
 102 |   type: "BatchNorm"
 103 |   bottom: "conv_2_1_pw"
 104 |   top: "conv_2_1_pw"
 105 |   param {
 106 |     lr_mult: 0.0
 107 |     decay_mult: 0.0
 108 |   }
 109 |   param {
 110 |     lr_mult: 0.0
 111 |     decay_mult: 0.0
 112 |   }
 113 |   param {
 114 |     lr_mult: 0.0
 115 |     decay_mult: 0.0
 116 |   }
 117 |   batch_norm_param {
 118 |     use_global_stats: true
 119 |   }
 120 | }
 121 | layer {
 122 |   name: "scale_2_1_pw"
 123 |   type: "Scale"
 124 |   bottom: "conv_2_1_pw"
 125 |   top: "conv_2_1_pw"
 126 |   scale_param {
 127 |     bias_term: true
 128 |   }
 129 | }
 130 | layer {
 131 |   name: "relu_2_1_pw"
 132 |   type: "ReLU"
 133 |   bottom: "conv_2_1_pw"
 134 |   top: "conv_2_1_pw"
 135 | }
 136 | layer {
 137 |   name: "conv_2_1_dw"
 138 |   type: "Convolution"
 139 |   bottom: "conv_2_1_pw"
 140 |   top: "conv_2_1_dw"
 141 |   param {
 142 |     lr_mult: 1.0
 143 |     decay_mult: 1.0
 144 |   }
 145 |   convolution_param {
 146 |     num_output: 16
 147 |     bias_term: false
 148 |     pad: 1
 149 |     kernel_size: 3
 150 |     group: 16
 151 |     stride: 1
 152 |     weight_filler {
 153 |       type: "msra"
 154 |     }
 155 |     bias_filler {
 156 |       type: "constant"
 157 |       value: 0.0
 158 |     }
 159 |     engine: CAFFE
 160 |   }
 161 | }
 162 | layer {
 163 |   name: "bn_2_1_dw"
 164 |   type: "BatchNorm"
 165 |   bottom: "conv_2_1_dw"
 166 |   top: "conv_2_1_dw"
 167 |   param {
 168 |     lr_mult: 0.0
 169 |     decay_mult: 0.0
 170 |   }
 171 |   param {
 172 |     lr_mult: 0.0
 173 |     decay_mult: 0.0
 174 |   }
 175 |   param {
 176 |     lr_mult: 0.0
 177 |     decay_mult: 0.0
 178 |   }
 179 |   batch_norm_param {
 180 |     use_global_stats: true
 181 |   }
 182 | }
 183 | layer {
 184 |   name: "scale_2_1_dw"
 185 |   type: "Scale"
 186 |   bottom: "conv_2_1_dw"
 187 |   top: "conv_2_1_dw"
 188 |   scale_param {
 189 |     bias_term: true
 190 |   }
 191 | }
 192 | layer {
 193 |   name: "relu_2_1_dw"
 194 |   type: "ReLU"
 195 |   bottom: "conv_2_1_dw"
 196 |   top: "conv_2_1_dw"
 197 | }
 198 | layer {
 199 |   name: "conv_2_1_linear"
 200 |   type: "Convolution"
 201 |   bottom: "conv_2_1_dw"
 202 |   top: "conv_2_1_linear"
 203 |   param {
 204 |     lr_mult: 1.0
 205 |     decay_mult: 1.0
 206 |   }
 207 |   convolution_param {
 208 |     num_output: 8
 209 |     bias_term: false
 210 |     pad: 0
 211 |     kernel_size: 1
 212 |     stride: 1
 213 |     weight_filler {
 214 |       type: "msra"
 215 |     }
 216 |     bias_filler {
 217 |       type: "constant"
 218 |       value: 0.0
 219 |     }
 220 |   }
 221 | }
 222 | layer {
 223 |   name: "bn_2_1_linear"
 224 |   type: "BatchNorm"
 225 |   bottom: "conv_2_1_linear"
 226 |   top: "conv_2_1_linear"
 227 |   param {
 228 |     lr_mult: 0.0
 229 |     decay_mult: 0.0
 230 |   }
 231 |   param {
 232 |     lr_mult: 0.0
 233 |     decay_mult: 0.0
 234 |   }
 235 |   param {
 236 |     lr_mult: 0.0
 237 |     decay_mult: 0.0
 238 |   }
 239 |   batch_norm_param {
 240 |     use_global_stats: true
 241 |   }
 242 | }
 243 | layer {
 244 |   name: "scale_2_1_linear"
 245 |   type: "Scale"
 246 |   bottom: "conv_2_1_linear"
 247 |   top: "conv_2_1_linear"
 248 |   scale_param {
 249 |     bias_term: true
 250 |   }
 251 | }
 252 | layer {
 253 |   name: "conv_3_1_pw"
 254 |   type: "Convolution"
 255 |   bottom: "conv_2_1_linear"
 256 |   top: "conv_3_1_pw"
 257 |   param {
 258 |     lr_mult: 1.0
 259 |     decay_mult: 1.0
 260 |   }
 261 |   convolution_param {
 262 |     num_output: 48
 263 |     bias_term: false
 264 |     pad: 0
 265 |     kernel_size: 1
 266 |     stride: 1
 267 |     weight_filler {
 268 |       type: "msra"
 269 |     }
 270 |     bias_filler {
 271 |       type: "constant"
 272 |       value: 0.0
 273 |     }
 274 |   }
 275 | }
 276 | layer {
 277 |   name: "bn_3_1_pw"
 278 |   type: "BatchNorm"
 279 |   bottom: "conv_3_1_pw"
 280 |   top: "conv_3_1_pw"
 281 |   param {
 282 |     lr_mult: 0.0
 283 |     decay_mult: 0.0
 284 |   }
 285 |   param {
 286 |     lr_mult: 0.0
 287 |     decay_mult: 0.0
 288 |   }
 289 |   param {
 290 |     lr_mult: 0.0
 291 |     decay_mult: 0.0
 292 |   }
 293 |   batch_norm_param {
 294 |     use_global_stats: true
 295 |   }
 296 | }
 297 | layer {
 298 |   name: "scale_3_1_pw"
 299 |   type: "Scale"
 300 |   bottom: "conv_3_1_pw"
 301 |   top: "conv_3_1_pw"
 302 |   scale_param {
 303 |     bias_term: true
 304 |   }
 305 | }
 306 | layer {
 307 |   name: "relu_3_1_pw"
 308 |   type: "ReLU"
 309 |   bottom: "conv_3_1_pw"
 310 |   top: "conv_3_1_pw"
 311 | }
 312 | layer {
 313 |   name: "conv_3_1_dw"
 314 |   type: "Convolution"
 315 |   bottom: "conv_3_1_pw"
 316 |   top: "conv_3_1_dw"
 317 |   param {
 318 |     lr_mult: 1.0
 319 |     decay_mult: 1.0
 320 |   }
 321 |   convolution_param {
 322 |     num_output: 48
 323 |     bias_term: false
 324 |     pad: 1
 325 |     kernel_size: 3
 326 |     group: 48
 327 |     stride: 2
 328 |     weight_filler {
 329 |       type: "msra"
 330 |     }
 331 |     bias_filler {
 332 |       type: "constant"
 333 |       value: 0.0
 334 |     }
 335 |     engine: CAFFE
 336 |   }
 337 | }
 338 | layer {
 339 |   name: "bn_3_1_dw"
 340 |   type: "BatchNorm"
 341 |   bottom: "conv_3_1_dw"
 342 |   top: "conv_3_1_dw"
 343 |   param {
 344 |     lr_mult: 0.0
 345 |     decay_mult: 0.0
 346 |   }
 347 |   param {
 348 |     lr_mult: 0.0
 349 |     decay_mult: 0.0
 350 |   }
 351 |   param {
 352 |     lr_mult: 0.0
 353 |     decay_mult: 0.0
 354 |   }
 355 |   batch_norm_param {
 356 |     use_global_stats: true
 357 |   }
 358 | }
 359 | layer {
 360 |   name: "scale_3_1_dw"
 361 |   type: "Scale"
 362 |   bottom: "conv_3_1_dw"
 363 |   top: "conv_3_1_dw"
 364 |   scale_param {
 365 |     bias_term: true
 366 |   }
 367 | }
 368 | layer {
 369 |   name: "relu_3_1_dw"
 370 |   type: "ReLU"
 371 |   bottom: "conv_3_1_dw"
 372 |   top: "conv_3_1_dw"
 373 | }
 374 | layer {
 375 |   name: "conv_3_1_linear"
 376 |   type: "Convolution"
 377 |   bottom: "conv_3_1_dw"
 378 |   top: "conv_3_1_linear"
 379 |   param {
 380 |     lr_mult: 1.0
 381 |     decay_mult: 1.0
 382 |   }
 383 |   convolution_param {
 384 |     num_output: 16
 385 |     bias_term: false
 386 |     pad: 0
 387 |     kernel_size: 1
 388 |     stride: 1
 389 |     weight_filler {
 390 |       type: "msra"
 391 |     }
 392 |     bias_filler {
 393 |       type: "constant"
 394 |       value: 0.0
 395 |     }
 396 |   }
 397 | }
 398 | layer {
 399 |   name: "bn_3_1_linear"
 400 |   type: "BatchNorm"
 401 |   bottom: "conv_3_1_linear"
 402 |   top: "conv_3_1_linear"
 403 |   param {
 404 |     lr_mult: 0.0
 405 |     decay_mult: 0.0
 406 |   }
 407 |   param {
 408 |     lr_mult: 0.0
 409 |     decay_mult: 0.0
 410 |   }
 411 |   param {
 412 |     lr_mult: 0.0
 413 |     decay_mult: 0.0
 414 |   }
 415 |   batch_norm_param {
 416 |     use_global_stats: true
 417 |   }
 418 | }
 419 | layer {
 420 |   name: "scale_3_1_linear"
 421 |   type: "Scale"
 422 |   bottom: "conv_3_1_linear"
 423 |   top: "conv_3_1_linear"
 424 |   scale_param {
 425 |     bias_term: true
 426 |   }
 427 | }
 428 | layer {
 429 |   name: "conv_3_2_pw"
 430 |   type: "Convolution"
 431 |   bottom: "conv_3_1_linear"
 432 |   top: "conv_3_2_pw"
 433 |   param {
 434 |     lr_mult: 1.0
 435 |     decay_mult: 1.0
 436 |   }
 437 |   convolution_param {
 438 |     num_output: 96
 439 |     bias_term: false
 440 |     pad: 0
 441 |     kernel_size: 1
 442 |     stride: 1
 443 |     weight_filler {
 444 |       type: "msra"
 445 |     }
 446 |     bias_filler {
 447 |       type: "constant"
 448 |       value: 0.0
 449 |     }
 450 |   }
 451 | }
 452 | layer {
 453 |   name: "bn_3_2_pw"
 454 |   type: "BatchNorm"
 455 |   bottom: "conv_3_2_pw"
 456 |   top: "conv_3_2_pw"
 457 |   param {
 458 |     lr_mult: 0.0
 459 |     decay_mult: 0.0
 460 |   }
 461 |   param {
 462 |     lr_mult: 0.0
 463 |     decay_mult: 0.0
 464 |   }
 465 |   param {
 466 |     lr_mult: 0.0
 467 |     decay_mult: 0.0
 468 |   }
 469 |   batch_norm_param {
 470 |     use_global_stats: true
 471 |   }
 472 | }
 473 | layer {
 474 |   name: "scale_3_2_pw"
 475 |   type: "Scale"
 476 |   bottom: "conv_3_2_pw"
 477 |   top: "conv_3_2_pw"
 478 |   scale_param {
 479 |     bias_term: true
 480 |   }
 481 | }
 482 | layer {
 483 |   name: "relu_3_2_pw"
 484 |   type: "ReLU"
 485 |   bottom: "conv_3_2_pw"
 486 |   top: "conv_3_2_pw"
 487 | }
 488 | layer {
 489 |   name: "conv_3_2_dw"
 490 |   type: "Convolution"
 491 |   bottom: "conv_3_2_pw"
 492 |   top: "conv_3_2_dw"
 493 |   param {
 494 |     lr_mult: 1.0
 495 |     decay_mult: 1.0
 496 |   }
 497 |   convolution_param {
 498 |     num_output: 96
 499 |     bias_term: false
 500 |     pad: 1
 501 |     kernel_size: 3
 502 |     group: 96
 503 |     stride: 1
 504 |     weight_filler {
 505 |       type: "msra"
 506 |     }
 507 |     bias_filler {
 508 |       type: "constant"
 509 |       value: 0.0
 510 |     }
 511 |     engine: CAFFE
 512 |   }
 513 | }
 514 | layer {
 515 |   name: "bn_3_2_dw"
 516 |   type: "BatchNorm"
 517 |   bottom: "conv_3_2_dw"
 518 |   top: "conv_3_2_dw"
 519 |   param {
 520 |     lr_mult: 0.0
 521 |     decay_mult: 0.0
 522 |   }
 523 |   param {
 524 |     lr_mult: 0.0
 525 |     decay_mult: 0.0
 526 |   }
 527 |   param {
 528 |     lr_mult: 0.0
 529 |     decay_mult: 0.0
 530 |   }
 531 |   batch_norm_param {
 532 |     use_global_stats: true
 533 |   }
 534 | }
 535 | layer {
 536 |   name: "scale_3_2_dw"
 537 |   type: "Scale"
 538 |   bottom: "conv_3_2_dw"
 539 |   top: "conv_3_2_dw"
 540 |   scale_param {
 541 |     bias_term: true
 542 |   }
 543 | }
 544 | layer {
 545 |   name: "relu_3_2_dw"
 546 |   type: "ReLU"
 547 |   bottom: "conv_3_2_dw"
 548 |   top: "conv_3_2_dw"
 549 | }
 550 | layer {
 551 |   name: "conv_3_2_linear"
 552 |   type: "Convolution"
 553 |   bottom: "conv_3_2_dw"
 554 |   top: "conv_3_2_linear"
 555 |   param {
 556 |     lr_mult: 1.0
 557 |     decay_mult: 1.0
 558 |   }
 559 |   convolution_param {
 560 |     num_output: 16
 561 |     bias_term: false
 562 |     pad: 0
 563 |     kernel_size: 1
 564 |     stride: 1
 565 |     weight_filler {
 566 |       type: "msra"
 567 |     }
 568 |     bias_filler {
 569 |       type: "constant"
 570 |       value: 0.0
 571 |     }
 572 |   }
 573 | }
 574 | layer {
 575 |   name: "bn_3_2_linear"
 576 |   type: "BatchNorm"
 577 |   bottom: "conv_3_2_linear"
 578 |   top: "conv_3_2_linear"
 579 |   param {
 580 |     lr_mult: 0.0
 581 |     decay_mult: 0.0
 582 |   }
 583 |   param {
 584 |     lr_mult: 0.0
 585 |     decay_mult: 0.0
 586 |   }
 587 |   param {
 588 |     lr_mult: 0.0
 589 |     decay_mult: 0.0
 590 |   }
 591 |   batch_norm_param {
 592 |     use_global_stats: true
 593 |   }
 594 | }
 595 | layer {
 596 |   name: "scale_3_2_linear"
 597 |   type: "Scale"
 598 |   bottom: "conv_3_2_linear"
 599 |   top: "conv_3_2_linear"
 600 |   scale_param {
 601 |     bias_term: true
 602 |   }
 603 | }
 604 | layer {
 605 |   name: "add_3_2"
 606 |   type: "Eltwise"
 607 |   bottom: "conv_3_1_linear"
 608 |   bottom: "conv_3_2_linear"
 609 |   top: "add_3_2"
 610 | }
 611 | layer {
 612 |   name: "conv_4_1_pw"
 613 |   type: "Convolution"
 614 |   bottom: "add_3_2"
 615 |   top: "conv_4_1_pw"
 616 |   param {
 617 |     lr_mult: 1.0
 618 |     decay_mult: 1.0
 619 |   }
 620 |   convolution_param {
 621 |     num_output: 96
 622 |     bias_term: false
 623 |     pad: 0
 624 |     kernel_size: 1
 625 |     stride: 1
 626 |     weight_filler {
 627 |       type: "msra"
 628 |     }
 629 |     bias_filler {
 630 |       type: "constant"
 631 |       value: 0.0
 632 |     }
 633 |   }
 634 | }
 635 | layer {
 636 |   name: "bn_4_1_pw"
 637 |   type: "BatchNorm"
 638 |   bottom: "conv_4_1_pw"
 639 |   top: "conv_4_1_pw"
 640 |   param {
 641 |     lr_mult: 0.0
 642 |     decay_mult: 0.0
 643 |   }
 644 |   param {
 645 |     lr_mult: 0.0
 646 |     decay_mult: 0.0
 647 |   }
 648 |   param {
 649 |     lr_mult: 0.0
 650 |     decay_mult: 0.0
 651 |   }
 652 |   batch_norm_param {
 653 |     use_global_stats: true
 654 |   }
 655 | }
 656 | layer {
 657 |   name: "scale_4_1_pw"
 658 |   type: "Scale"
 659 |   bottom: "conv_4_1_pw"
 660 |   top: "conv_4_1_pw"
 661 |   scale_param {
 662 |     bias_term: true
 663 |   }
 664 | }
 665 | layer {
 666 |   name: "relu_4_1_pw"
 667 |   type: "ReLU"
 668 |   bottom: "conv_4_1_pw"
 669 |   top: "conv_4_1_pw"
 670 | }
 671 | layer {
 672 |   name: "conv_4_1_dw"
 673 |   type: "Convolution"
 674 |   bottom: "conv_4_1_pw"
 675 |   top: "conv_4_1_dw"
 676 |   param {
 677 |     lr_mult: 1.0
 678 |     decay_mult: 1.0
 679 |   }
 680 |   convolution_param {
 681 |     num_output: 96
 682 |     bias_term: false
 683 |     pad: 1
 684 |     kernel_size: 3
 685 |     group: 96
 686 |     stride: 2
 687 |     weight_filler {
 688 |       type: "msra"
 689 |     }
 690 |     bias_filler {
 691 |       type: "constant"
 692 |       value: 0.0
 693 |     }
 694 |     engine: CAFFE
 695 |   }
 696 | }
 697 | layer {
 698 |   name: "bn_4_1_dw"
 699 |   type: "BatchNorm"
 700 |   bottom: "conv_4_1_dw"
 701 |   top: "conv_4_1_dw"
 702 |   param {
 703 |     lr_mult: 0.0
 704 |     decay_mult: 0.0
 705 |   }
 706 |   param {
 707 |     lr_mult: 0.0
 708 |     decay_mult: 0.0
 709 |   }
 710 |   param {
 711 |     lr_mult: 0.0
 712 |     decay_mult: 0.0
 713 |   }
 714 |   batch_norm_param {
 715 |     use_global_stats: true
 716 |   }
 717 | }
 718 | layer {
 719 |   name: "scale_4_1_dw"
 720 |   type: "Scale"
 721 |   bottom: "conv_4_1_dw"
 722 |   top: "conv_4_1_dw"
 723 |   scale_param {
 724 |     bias_term: true
 725 |   }
 726 | }
 727 | layer {
 728 |   name: "relu_4_1_dw"
 729 |   type: "ReLU"
 730 |   bottom: "conv_4_1_dw"
 731 |   top: "conv_4_1_dw"
 732 | }
 733 | layer {
 734 |   name: "conv_4_1_linear"
 735 |   type: "Convolution"
 736 |   bottom: "conv_4_1_dw"
 737 |   top: "conv_4_1_linear"
 738 |   param {
 739 |     lr_mult: 1.0
 740 |     decay_mult: 1.0
 741 |   }
 742 |   convolution_param {
 743 |     num_output: 16
 744 |     bias_term: false
 745 |     pad: 0
 746 |     kernel_size: 1
 747 |     stride: 1
 748 |     weight_filler {
 749 |       type: "msra"
 750 |     }
 751 |     bias_filler {
 752 |       type: "constant"
 753 |       value: 0.0
 754 |     }
 755 |   }
 756 | }
 757 | layer {
 758 |   name: "bn_4_1_linear"
 759 |   type: "BatchNorm"
 760 |   bottom: "conv_4_1_linear"
 761 |   top: "conv_4_1_linear"
 762 |   param {
 763 |     lr_mult: 0.0
 764 |     decay_mult: 0.0
 765 |   }
 766 |   param {
 767 |     lr_mult: 0.0
 768 |     decay_mult: 0.0
 769 |   }
 770 |   param {
 771 |     lr_mult: 0.0
 772 |     decay_mult: 0.0
 773 |   }
 774 |   batch_norm_param {
 775 |     use_global_stats: true
 776 |   }
 777 | }
 778 | layer {
 779 |   name: "scale_4_1_linear"
 780 |   type: "Scale"
 781 |   bottom: "conv_4_1_linear"
 782 |   top: "conv_4_1_linear"
 783 |   scale_param {
 784 |     bias_term: true
 785 |   }
 786 | }
 787 | layer {
 788 |   name: "conv_4_2_pw"
 789 |   type: "Convolution"
 790 |   bottom: "conv_4_1_linear"
 791 |   top: "conv_4_2_pw"
 792 |   param {
 793 |     lr_mult: 1.0
 794 |     decay_mult: 1.0
 795 |   }
 796 |   convolution_param {
 797 |     num_output: 96
 798 |     bias_term: false
 799 |     pad: 0
 800 |     kernel_size: 1
 801 |     stride: 1
 802 |     weight_filler {
 803 |       type: "msra"
 804 |     }
 805 |     bias_filler {
 806 |       type: "constant"
 807 |       value: 0.0
 808 |     }
 809 |   }
 810 | }
 811 | layer {
 812 |   name: "bn_4_2_pw"
 813 |   type: "BatchNorm"
 814 |   bottom: "conv_4_2_pw"
 815 |   top: "conv_4_2_pw"
 816 |   param {
 817 |     lr_mult: 0.0
 818 |     decay_mult: 0.0
 819 |   }
 820 |   param {
 821 |     lr_mult: 0.0
 822 |     decay_mult: 0.0
 823 |   }
 824 |   param {
 825 |     lr_mult: 0.0
 826 |     decay_mult: 0.0
 827 |   }
 828 |   batch_norm_param {
 829 |     use_global_stats: true
 830 |   }
 831 | }
 832 | layer {
 833 |   name: "scale_4_2_pw"
 834 |   type: "Scale"
 835 |   bottom: "conv_4_2_pw"
 836 |   top: "conv_4_2_pw"
 837 |   scale_param {
 838 |     bias_term: true
 839 |   }
 840 | }
 841 | layer {
 842 |   name: "relu_4_2_pw"
 843 |   type: "ReLU"
 844 |   bottom: "conv_4_2_pw"
 845 |   top: "conv_4_2_pw"
 846 | }
 847 | layer {
 848 |   name: "conv_4_2_dw"
 849 |   type: "Convolution"
 850 |   bottom: "conv_4_2_pw"
 851 |   top: "conv_4_2_dw"
 852 |   param {
 853 |     lr_mult: 1.0
 854 |     decay_mult: 1.0
 855 |   }
 856 |   convolution_param {
 857 |     num_output: 96
 858 |     bias_term: false
 859 |     pad: 1
 860 |     kernel_size: 3
 861 |     group: 96
 862 |     stride: 1
 863 |     weight_filler {
 864 |       type: "msra"
 865 |     }
 866 |     bias_filler {
 867 |       type: "constant"
 868 |       value: 0.0
 869 |     }
 870 |     engine: CAFFE
 871 |   }
 872 | }
 873 | layer {
 874 |   name: "bn_4_2_dw"
 875 |   type: "BatchNorm"
 876 |   bottom: "conv_4_2_dw"
 877 |   top: "conv_4_2_dw"
 878 |   param {
 879 |     lr_mult: 0.0
 880 |     decay_mult: 0.0
 881 |   }
 882 |   param {
 883 |     lr_mult: 0.0
 884 |     decay_mult: 0.0
 885 |   }
 886 |   param {
 887 |     lr_mult: 0.0
 888 |     decay_mult: 0.0
 889 |   }
 890 |   batch_norm_param {
 891 |     use_global_stats: true
 892 |   }
 893 | }
 894 | layer {
 895 |   name: "scale_4_2_dw"
 896 |   type: "Scale"
 897 |   bottom: "conv_4_2_dw"
 898 |   top: "conv_4_2_dw"
 899 |   scale_param {
 900 |     bias_term: true
 901 |   }
 902 | }
 903 | layer {
 904 |   name: "relu_4_2_dw"
 905 |   type: "ReLU"
 906 |   bottom: "conv_4_2_dw"
 907 |   top: "conv_4_2_dw"
 908 | }
 909 | layer {
 910 |   name: "conv_4_2_linear"
 911 |   type: "Convolution"
 912 |   bottom: "conv_4_2_dw"
 913 |   top: "conv_4_2_linear"
 914 |   param {
 915 |     lr_mult: 1.0
 916 |     decay_mult: 1.0
 917 |   }
 918 |   convolution_param {
 919 |     num_output: 16
 920 |     bias_term: false
 921 |     pad: 0
 922 |     kernel_size: 1
 923 |     stride: 1
 924 |     weight_filler {
 925 |       type: "msra"
 926 |     }
 927 |     bias_filler {
 928 |       type: "constant"
 929 |       value: 0.0
 930 |     }
 931 |   }
 932 | }
 933 | layer {
 934 |   name: "bn_4_2_linear"
 935 |   type: "BatchNorm"
 936 |   bottom: "conv_4_2_linear"
 937 |   top: "conv_4_2_linear"
 938 |   param {
 939 |     lr_mult: 0.0
 940 |     decay_mult: 0.0
 941 |   }
 942 |   param {
 943 |     lr_mult: 0.0
 944 |     decay_mult: 0.0
 945 |   }
 946 |   param {
 947 |     lr_mult: 0.0
 948 |     decay_mult: 0.0
 949 |   }
 950 |   batch_norm_param {
 951 |     use_global_stats: true
 952 |   }
 953 | }
 954 | layer {
 955 |   name: "scale_4_2_linear"
 956 |   type: "Scale"
 957 |   bottom: "conv_4_2_linear"
 958 |   top: "conv_4_2_linear"
 959 |   scale_param {
 960 |     bias_term: true
 961 |   }
 962 | }
 963 | layer {
 964 |   name: "add_4_2"
 965 |   type: "Eltwise"
 966 |   bottom: "conv_4_1_linear"
 967 |   bottom: "conv_4_2_linear"
 968 |   top: "add_4_2"
 969 | }
 970 | layer {
 971 |   name: "conv_4_3_pw"
 972 |   type: "Convolution"
 973 |   bottom: "add_4_2"
 974 |   top: "conv_4_3_pw"
 975 |   param {
 976 |     lr_mult: 1.0
 977 |     decay_mult: 1.0
 978 |   }
 979 |   convolution_param {
 980 |     num_output: 96
 981 |     bias_term: false
 982 |     pad: 0
 983 |     kernel_size: 1
 984 |     stride: 1
 985 |     weight_filler {
 986 |       type: "msra"
 987 |     }
 988 |     bias_filler {
 989 |       type: "constant"
 990 |       value: 0.0
 991 |     }
 992 |   }
 993 | }
 994 | layer {
 995 |   name: "bn_4_3_pw"
 996 |   type: "BatchNorm"
 997 |   bottom: "conv_4_3_pw"
 998 |   top: "conv_4_3_pw"
 999 |   param {
1000 |     lr_mult: 0.0
1001 |     decay_mult: 0.0
1002 |   }
1003 |   param {
1004 |     lr_mult: 0.0
1005 |     decay_mult: 0.0
1006 |   }
1007 |   param {
1008 |     lr_mult: 0.0
1009 |     decay_mult: 0.0
1010 |   }
1011 |   batch_norm_param {
1012 |     use_global_stats: true
1013 |   }
1014 | }
1015 | layer {
1016 |   name: "scale_4_3_pw"
1017 |   type: "Scale"
1018 |   bottom: "conv_4_3_pw"
1019 |   top: "conv_4_3_pw"
1020 |   scale_param {
1021 |     bias_term: true
1022 |   }
1023 | }
1024 | layer {
1025 |   name: "relu_4_3_pw"
1026 |   type: "ReLU"
1027 |   bottom: "conv_4_3_pw"
1028 |   top: "conv_4_3_pw"
1029 | }
1030 | layer {
1031 |   name: "conv_4_3_dw"
1032 |   type: "Convolution"
1033 |   bottom: "conv_4_3_pw"
1034 |   top: "conv_4_3_dw"
1035 |   param {
1036 |     lr_mult: 1.0
1037 |     decay_mult: 1.0
1038 |   }
1039 |   convolution_param {
1040 |     num_output: 96
1041 |     bias_term: false
1042 |     pad: 1
1043 |     kernel_size: 3
1044 |     group: 96
1045 |     stride: 1
1046 |     weight_filler {
1047 |       type: "msra"
1048 |     }
1049 |     bias_filler {
1050 |       type: "constant"
1051 |       value: 0.0
1052 |     }
1053 |     engine: CAFFE
1054 |   }
1055 | }
1056 | layer {
1057 |   name: "bn_4_3_dw"
1058 |   type: "BatchNorm"
1059 |   bottom: "conv_4_3_dw"
1060 |   top: "conv_4_3_dw"
1061 |   param {
1062 |     lr_mult: 0.0
1063 |     decay_mult: 0.0
1064 |   }
1065 |   param {
1066 |     lr_mult: 0.0
1067 |     decay_mult: 0.0
1068 |   }
1069 |   param {
1070 |     lr_mult: 0.0
1071 |     decay_mult: 0.0
1072 |   }
1073 |   batch_norm_param {
1074 |     use_global_stats: true
1075 |   }
1076 | }
1077 | layer {
1078 |   name: "scale_4_3_dw"
1079 |   type: "Scale"
1080 |   bottom: "conv_4_3_dw"
1081 |   top: "conv_4_3_dw"
1082 |   scale_param {
1083 |     bias_term: true
1084 |   }
1085 | }
1086 | layer {
1087 |   name: "relu_4_3_dw"
1088 |   type: "ReLU"
1089 |   bottom: "conv_4_3_dw"
1090 |   top: "conv_4_3_dw"
1091 | }
1092 | layer {
1093 |   name: "conv_4_3_linear"
1094 |   type: "Convolution"
1095 |   bottom: "conv_4_3_dw"
1096 |   top: "conv_4_3_linear"
1097 |   param {
1098 |     lr_mult: 1.0
1099 |     decay_mult: 1.0
1100 |   }
1101 |   convolution_param {
1102 |     num_output: 16
1103 |     bias_term: false
1104 |     pad: 0
1105 |     kernel_size: 1
1106 |     stride: 1
1107 |     weight_filler {
1108 |       type: "msra"
1109 |     }
1110 |     bias_filler {
1111 |       type: "constant"
1112 |       value: 0.0
1113 |     }
1114 |   }
1115 | }
1116 | layer {
1117 |   name: "bn_4_3_linear"
1118 |   type: "BatchNorm"
1119 |   bottom: "conv_4_3_linear"
1120 |   top: "conv_4_3_linear"
1121 |   param {
1122 |     lr_mult: 0.0
1123 |     decay_mult: 0.0
1124 |   }
1125 |   param {
1126 |     lr_mult: 0.0
1127 |     decay_mult: 0.0
1128 |   }
1129 |   param {
1130 |     lr_mult: 0.0
1131 |     decay_mult: 0.0
1132 |   }
1133 |   batch_norm_param {
1134 |     use_global_stats: true
1135 |   }
1136 | }
1137 | layer {
1138 |   name: "scale_4_3_linear"
1139 |   type: "Scale"
1140 |   bottom: "conv_4_3_linear"
1141 |   top: "conv_4_3_linear"
1142 |   scale_param {
1143 |     bias_term: true
1144 |   }
1145 | }
1146 | layer {
1147 |   name: "add_4_3"
1148 |   type: "Eltwise"
1149 |   bottom: "add_4_2"
1150 |   bottom: "conv_4_3_linear"
1151 |   top: "add_4_3"
1152 | }
1153 | layer {
1154 |   name: "conv_5_1_pw"
1155 |   type: "Convolution"
1156 |   bottom: "add_4_3"
1157 |   top: "conv_5_1_pw"
1158 |   param {
1159 |     lr_mult: 1.0
1160 |     decay_mult: 1.0
1161 |   }
1162 |   convolution_param {
1163 |     num_output: 96
1164 |     bias_term: false
1165 |     pad: 0
1166 |     kernel_size: 1
1167 |     stride: 1
1168 |     weight_filler {
1169 |       type: "msra"
1170 |     }
1171 |     bias_filler {
1172 |       type: "constant"
1173 |       value: 0.0
1174 |     }
1175 |   }
1176 | }
1177 | layer {
1178 |   name: "bn_5_1_pw"
1179 |   type: "BatchNorm"
1180 |   bottom: "conv_5_1_pw"
1181 |   top: "conv_5_1_pw"
1182 |   param {
1183 |     lr_mult: 0.0
1184 |     decay_mult: 0.0
1185 |   }
1186 |   param {
1187 |     lr_mult: 0.0
1188 |     decay_mult: 0.0
1189 |   }
1190 |   param {
1191 |     lr_mult: 0.0
1192 |     decay_mult: 0.0
1193 |   }
1194 |   batch_norm_param {
1195 |     use_global_stats: true
1196 |   }
1197 | }
1198 | layer {
1199 |   name: "scale_5_1_pw"
1200 |   type: "Scale"
1201 |   bottom: "conv_5_1_pw"
1202 |   top: "conv_5_1_pw"
1203 |   scale_param {
1204 |     bias_term: true
1205 |   }
1206 | }
1207 | layer {
1208 |   name: "relu_5_1_pw"
1209 |   type: "ReLU"
1210 |   bottom: "conv_5_1_pw"
1211 |   top: "conv_5_1_pw"
1212 | }
1213 | layer {
1214 |   name: "conv_5_1_dw"
1215 |   type: "Convolution"
1216 |   bottom: "conv_5_1_pw"
1217 |   top: "conv_5_1_dw"
1218 |   param {
1219 |     lr_mult: 1.0
1220 |     decay_mult: 1.0
1221 |   }
1222 |   convolution_param {
1223 |     num_output: 96
1224 |     bias_term: false
1225 |     pad: 1
1226 |     kernel_size: 3
1227 |     group: 96
1228 |     stride: 2
1229 |     weight_filler {
1230 |       type: "msra"
1231 |     }
1232 |     bias_filler {
1233 |       type: "constant"
1234 |       value: 0.0
1235 |     }
1236 |     engine: CAFFE
1237 |   }
1238 | }
1239 | layer {
1240 |   name: "bn_5_1_dw"
1241 |   type: "BatchNorm"
1242 |   bottom: "conv_5_1_dw"
1243 |   top: "conv_5_1_dw"
1244 |   param {
1245 |     lr_mult: 0.0
1246 |     decay_mult: 0.0
1247 |   }
1248 |   param {
1249 |     lr_mult: 0.0
1250 |     decay_mult: 0.0
1251 |   }
1252 |   param {
1253 |     lr_mult: 0.0
1254 |     decay_mult: 0.0
1255 |   }
1256 |   batch_norm_param {
1257 |     use_global_stats: true
1258 |   }
1259 | }
1260 | layer {
1261 |   name: "scale_5_1_dw"
1262 |   type: "Scale"
1263 |   bottom: "conv_5_1_dw"
1264 |   top: "conv_5_1_dw"
1265 |   scale_param {
1266 |     bias_term: true
1267 |   }
1268 | }
1269 | layer {
1270 |   name: "relu_5_1_dw"
1271 |   type: "ReLU"
1272 |   bottom: "conv_5_1_dw"
1273 |   top: "conv_5_1_dw"
1274 | }
1275 | layer {
1276 |   name: "conv_5_1_linear"
1277 |   type: "Convolution"
1278 |   bottom: "conv_5_1_dw"
1279 |   top: "conv_5_1_linear"
1280 |   param {
1281 |     lr_mult: 1.0
1282 |     decay_mult: 1.0
1283 |   }
1284 |   convolution_param {
1285 |     num_output: 32
1286 |     bias_term: false
1287 |     pad: 0
1288 |     kernel_size: 1
1289 |     stride: 1
1290 |     weight_filler {
1291 |       type: "msra"
1292 |     }
1293 |     bias_filler {
1294 |       type: "constant"
1295 |       value: 0.0
1296 |     }
1297 |   }
1298 | }
1299 | layer {
1300 |   name: "bn_5_1_linear"
1301 |   type: "BatchNorm"
1302 |   bottom: "conv_5_1_linear"
1303 |   top: "conv_5_1_linear"
1304 |   param {
1305 |     lr_mult: 0.0
1306 |     decay_mult: 0.0
1307 |   }
1308 |   param {
1309 |     lr_mult: 0.0
1310 |     decay_mult: 0.0
1311 |   }
1312 |   param {
1313 |     lr_mult: 0.0
1314 |     decay_mult: 0.0
1315 |   }
1316 |   batch_norm_param {
1317 |     use_global_stats: true
1318 |   }
1319 | }
1320 | layer {
1321 |   name: "scale_5_1_linear"
1322 |   type: "Scale"
1323 |   bottom: "conv_5_1_linear"
1324 |   top: "conv_5_1_linear"
1325 |   scale_param {
1326 |     bias_term: true
1327 |   }
1328 | }
1329 | layer {
1330 |   name: "conv_5_2_pw"
1331 |   type: "Convolution"
1332 |   bottom: "conv_5_1_linear"
1333 |   top: "conv_5_2_pw"
1334 |   param {
1335 |     lr_mult: 1.0
1336 |     decay_mult: 1.0
1337 |   }
1338 |   convolution_param {
1339 |     num_output: 192
1340 |     bias_term: false
1341 |     pad: 0
1342 |     kernel_size: 1
1343 |     stride: 1
1344 |     weight_filler {
1345 |       type: "msra"
1346 |     }
1347 |     bias_filler {
1348 |       type: "constant"
1349 |       value: 0.0
1350 |     }
1351 |   }
1352 | }
1353 | layer {
1354 |   name: "bn_5_2_pw"
1355 |   type: "BatchNorm"
1356 |   bottom: "conv_5_2_pw"
1357 |   top: "conv_5_2_pw"
1358 |   param {
1359 |     lr_mult: 0.0
1360 |     decay_mult: 0.0
1361 |   }
1362 |   param {
1363 |     lr_mult: 0.0
1364 |     decay_mult: 0.0
1365 |   }
1366 |   param {
1367 |     lr_mult: 0.0
1368 |     decay_mult: 0.0
1369 |   }
1370 |   batch_norm_param {
1371 |     use_global_stats: true
1372 |   }
1373 | }
1374 | layer {
1375 |   name: "scale_5_2_pw"
1376 |   type: "Scale"
1377 |   bottom: "conv_5_2_pw"
1378 |   top: "conv_5_2_pw"
1379 |   scale_param {
1380 |     bias_term: true
1381 |   }
1382 | }
1383 | layer {
1384 |   name: "relu_5_2_pw"
1385 |   type: "ReLU"
1386 |   bottom: "conv_5_2_pw"
1387 |   top: "conv_5_2_pw"
1388 | }
1389 | layer {
1390 |   name: "conv_5_2_dw"
1391 |   type: "Convolution"
1392 |   bottom: "conv_5_2_pw"
1393 |   top: "conv_5_2_dw"
1394 |   param {
1395 |     lr_mult: 1.0
1396 |     decay_mult: 1.0
1397 |   }
1398 |   convolution_param {
1399 |     num_output: 192
1400 |     bias_term: false
1401 |     pad: 1
1402 |     kernel_size: 3
1403 |     group: 192
1404 |     stride: 1
1405 |     weight_filler {
1406 |       type: "msra"
1407 |     }
1408 |     bias_filler {
1409 |       type: "constant"
1410 |       value: 0.0
1411 |     }
1412 |     engine: CAFFE
1413 |   }
1414 | }
1415 | layer {
1416 |   name: "bn_5_2_dw"
1417 |   type: "BatchNorm"
1418 |   bottom: "conv_5_2_dw"
1419 |   top: "conv_5_2_dw"
1420 |   param {
1421 |     lr_mult: 0.0
1422 |     decay_mult: 0.0
1423 |   }
1424 |   param {
1425 |     lr_mult: 0.0
1426 |     decay_mult: 0.0
1427 |   }
1428 |   param {
1429 |     lr_mult: 0.0
1430 |     decay_mult: 0.0
1431 |   }
1432 |   batch_norm_param {
1433 |     use_global_stats: true
1434 |   }
1435 | }
1436 | layer {
1437 |   name: "scale_5_2_dw"
1438 |   type: "Scale"
1439 |   bottom: "conv_5_2_dw"
1440 |   top: "conv_5_2_dw"
1441 |   scale_param {
1442 |     bias_term: true
1443 |   }
1444 | }
1445 | layer {
1446 |   name: "relu_5_2_dw"
1447 |   type: "ReLU"
1448 |   bottom: "conv_5_2_dw"
1449 |   top: "conv_5_2_dw"
1450 | }
1451 | layer {
1452 |   name: "conv_5_2_linear"
1453 |   type: "Convolution"
1454 |   bottom: "conv_5_2_dw"
1455 |   top: "conv_5_2_linear"
1456 |   param {
1457 |     lr_mult: 1.0
1458 |     decay_mult: 1.0
1459 |   }
1460 |   convolution_param {
1461 |     num_output: 32
1462 |     bias_term: false
1463 |     pad: 0
1464 |     kernel_size: 1
1465 |     stride: 1
1466 |     weight_filler {
1467 |       type: "msra"
1468 |     }
1469 |     bias_filler {
1470 |       type: "constant"
1471 |       value: 0.0
1472 |     }
1473 |   }
1474 | }
1475 | layer {
1476 |   name: "bn_5_2_linear"
1477 |   type: "BatchNorm"
1478 |   bottom: "conv_5_2_linear"
1479 |   top: "conv_5_2_linear"
1480 |   param {
1481 |     lr_mult: 0.0
1482 |     decay_mult: 0.0
1483 |   }
1484 |   param {
1485 |     lr_mult: 0.0
1486 |     decay_mult: 0.0
1487 |   }
1488 |   param {
1489 |     lr_mult: 0.0
1490 |     decay_mult: 0.0
1491 |   }
1492 |   batch_norm_param {
1493 |     use_global_stats: true
1494 |   }
1495 | }
1496 | layer {
1497 |   name: "scale_5_2_linear"
1498 |   type: "Scale"
1499 |   bottom: "conv_5_2_linear"
1500 |   top: "conv_5_2_linear"
1501 |   scale_param {
1502 |     bias_term: true
1503 |   }
1504 | }
1505 | layer {
1506 |   name: "add_5_2"
1507 |   type: "Eltwise"
1508 |   bottom: "conv_5_1_linear"
1509 |   bottom: "conv_5_2_linear"
1510 |   top: "add_5_2"
1511 | }
1512 | layer {
1513 |   name: "conv_5_3_pw"
1514 |   type: "Convolution"
1515 |   bottom: "add_5_2"
1516 |   top: "conv_5_3_pw"
1517 |   param {
1518 |     lr_mult: 1.0
1519 |     decay_mult: 1.0
1520 |   }
1521 |   convolution_param {
1522 |     num_output: 192
1523 |     bias_term: false
1524 |     pad: 0
1525 |     kernel_size: 1
1526 |     stride: 1
1527 |     weight_filler {
1528 |       type: "msra"
1529 |     }
1530 |     bias_filler {
1531 |       type: "constant"
1532 |       value: 0.0
1533 |     }
1534 |   }
1535 | }
1536 | layer {
1537 |   name: "bn_5_3_pw"
1538 |   type: "BatchNorm"
1539 |   bottom: "conv_5_3_pw"
1540 |   top: "conv_5_3_pw"
1541 |   param {
1542 |     lr_mult: 0.0
1543 |     decay_mult: 0.0
1544 |   }
1545 |   param {
1546 |     lr_mult: 0.0
1547 |     decay_mult: 0.0
1548 |   }
1549 |   param {
1550 |     lr_mult: 0.0
1551 |     decay_mult: 0.0
1552 |   }
1553 |   batch_norm_param {
1554 |     use_global_stats: true
1555 |   }
1556 | }
1557 | layer {
1558 |   name: "scale_5_3_pw"
1559 |   type: "Scale"
1560 |   bottom: "conv_5_3_pw"
1561 |   top: "conv_5_3_pw"
1562 |   scale_param {
1563 |     bias_term: true
1564 |   }
1565 | }
1566 | layer {
1567 |   name: "relu_5_3_pw"
1568 |   type: "ReLU"
1569 |   bottom: "conv_5_3_pw"
1570 |   top: "conv_5_3_pw"
1571 | }
1572 | layer {
1573 |   name: "conv_5_3_dw"
1574 |   type: "Convolution"
1575 |   bottom: "conv_5_3_pw"
1576 |   top: "conv_5_3_dw"
1577 |   param {
1578 |     lr_mult: 1.0
1579 |     decay_mult: 1.0
1580 |   }
1581 |   convolution_param {
1582 |     num_output: 192
1583 |     bias_term: false
1584 |     pad: 1
1585 |     kernel_size: 3
1586 |     group: 192
1587 |     stride: 1
1588 |     weight_filler {
1589 |       type: "msra"
1590 |     }
1591 |     bias_filler {
1592 |       type: "constant"
1593 |       value: 0.0
1594 |     }
1595 |     engine: CAFFE
1596 |   }
1597 | }
1598 | layer {
1599 |   name: "bn_5_3_dw"
1600 |   type: "BatchNorm"
1601 |   bottom: "conv_5_3_dw"
1602 |   top: "conv_5_3_dw"
1603 |   param {
1604 |     lr_mult: 0.0
1605 |     decay_mult: 0.0
1606 |   }
1607 |   param {
1608 |     lr_mult: 0.0
1609 |     decay_mult: 0.0
1610 |   }
1611 |   param {
1612 |     lr_mult: 0.0
1613 |     decay_mult: 0.0
1614 |   }
1615 |   batch_norm_param {
1616 |     use_global_stats: true
1617 |   }
1618 | }
1619 | layer {
1620 |   name: "scale_5_3_dw"
1621 |   type: "Scale"
1622 |   bottom: "conv_5_3_dw"
1623 |   top: "conv_5_3_dw"
1624 |   scale_param {
1625 |     bias_term: true
1626 |   }
1627 | }
1628 | layer {
1629 |   name: "relu_5_3_dw"
1630 |   type: "ReLU"
1631 |   bottom: "conv_5_3_dw"
1632 |   top: "conv_5_3_dw"
1633 | }
1634 | layer {
1635 |   name: "conv_5_3_linear"
1636 |   type: "Convolution"
1637 |   bottom: "conv_5_3_dw"
1638 |   top: "conv_5_3_linear"
1639 |   param {
1640 |     lr_mult: 1.0
1641 |     decay_mult: 1.0
1642 |   }
1643 |   convolution_param {
1644 |     num_output: 32
1645 |     bias_term: false
1646 |     pad: 0
1647 |     kernel_size: 1
1648 |     stride: 1
1649 |     weight_filler {
1650 |       type: "msra"
1651 |     }
1652 |     bias_filler {
1653 |       type: "constant"
1654 |       value: 0.0
1655 |     }
1656 |   }
1657 | }
1658 | layer {
1659 |   name: "bn_5_3_linear"
1660 |   type: "BatchNorm"
1661 |   bottom: "conv_5_3_linear"
1662 |   top: "conv_5_3_linear"
1663 |   param {
1664 |     lr_mult: 0.0
1665 |     decay_mult: 0.0
1666 |   }
1667 |   param {
1668 |     lr_mult: 0.0
1669 |     decay_mult: 0.0
1670 |   }
1671 |   param {
1672 |     lr_mult: 0.0
1673 |     decay_mult: 0.0
1674 |   }
1675 |   batch_norm_param {
1676 |     use_global_stats: true
1677 |   }
1678 | }
1679 | layer {
1680 |   name: "scale_5_3_linear"
1681 |   type: "Scale"
1682 |   bottom: "conv_5_3_linear"
1683 |   top: "conv_5_3_linear"
1684 |   scale_param {
1685 |     bias_term: true
1686 |   }
1687 | }
1688 | layer {
1689 |   name: "add_5_3"
1690 |   type: "Eltwise"
1691 |   bottom: "add_5_2"
1692 |   bottom: "conv_5_3_linear"
1693 |   top: "add_5_3"
1694 | }
1695 | layer {
1696 |   name: "conv_5_4_pw"
1697 |   type: "Convolution"
1698 |   bottom: "add_5_3"
1699 |   top: "conv_5_4_pw"
1700 |   param {
1701 |     lr_mult: 1.0
1702 |     decay_mult: 1.0
1703 |   }
1704 |   convolution_param {
1705 |     num_output: 192
1706 |     bias_term: false
1707 |     pad: 0
1708 |     kernel_size: 1
1709 |     stride: 1
1710 |     weight_filler {
1711 |       type: "msra"
1712 |     }
1713 |     bias_filler {
1714 |       type: "constant"
1715 |       value: 0.0
1716 |     }
1717 |   }
1718 | }
1719 | layer {
1720 |   name: "bn_5_4_pw"
1721 |   type: "BatchNorm"
1722 |   bottom: "conv_5_4_pw"
1723 |   top: "conv_5_4_pw"
1724 |   param {
1725 |     lr_mult: 0.0
1726 |     decay_mult: 0.0
1727 |   }
1728 |   param {
1729 |     lr_mult: 0.0
1730 |     decay_mult: 0.0
1731 |   }
1732 |   param {
1733 |     lr_mult: 0.0
1734 |     decay_mult: 0.0
1735 |   }
1736 |   batch_norm_param {
1737 |     use_global_stats: true
1738 |   }
1739 | }
1740 | layer {
1741 |   name: "scale_5_4_pw"
1742 |   type: "Scale"
1743 |   bottom: "conv_5_4_pw"
1744 |   top: "conv_5_4_pw"
1745 |   scale_param {
1746 |     bias_term: true
1747 |   }
1748 | }
1749 | layer {
1750 |   name: "relu_5_4_pw"
1751 |   type: "ReLU"
1752 |   bottom: "conv_5_4_pw"
1753 |   top: "conv_5_4_pw"
1754 | }
1755 | layer {
1756 |   name: "conv_5_4_dw"
1757 |   type: "Convolution"
1758 |   bottom: "conv_5_4_pw"
1759 |   top: "conv_5_4_dw"
1760 |   param {
1761 |     lr_mult: 1.0
1762 |     decay_mult: 1.0
1763 |   }
1764 |   convolution_param {
1765 |     num_output: 192
1766 |     bias_term: false
1767 |     pad: 1
1768 |     kernel_size: 3
1769 |     group: 192
1770 |     stride: 1
1771 |     weight_filler {
1772 |       type: "msra"
1773 |     }
1774 |     bias_filler {
1775 |       type: "constant"
1776 |       value: 0.0
1777 |     }
1778 |     engine: CAFFE
1779 |   }
1780 | }
1781 | layer {
1782 |   name: "bn_5_4_dw"
1783 |   type: "BatchNorm"
1784 |   bottom: "conv_5_4_dw"
1785 |   top: "conv_5_4_dw"
1786 |   param {
1787 |     lr_mult: 0.0
1788 |     decay_mult: 0.0
1789 |   }
1790 |   param {
1791 |     lr_mult: 0.0
1792 |     decay_mult: 0.0
1793 |   }
1794 |   param {
1795 |     lr_mult: 0.0
1796 |     decay_mult: 0.0
1797 |   }
1798 |   batch_norm_param {
1799 |     use_global_stats: true
1800 |   }
1801 | }
1802 | layer {
1803 |   name: "scale_5_4_dw"
1804 |   type: "Scale"
1805 |   bottom: "conv_5_4_dw"
1806 |   top: "conv_5_4_dw"
1807 |   scale_param {
1808 |     bias_term: true
1809 |   }
1810 | }
1811 | layer {
1812 |   name: "relu_5_4_dw"
1813 |   type: "ReLU"
1814 |   bottom: "conv_5_4_dw"
1815 |   top: "conv_5_4_dw"
1816 | }
1817 | layer {
1818 |   name: "conv_5_4_linear"
1819 |   type: "Convolution"
1820 |   bottom: "conv_5_4_dw"
1821 |   top: "conv_5_4_linear"
1822 |   param {
1823 |     lr_mult: 1.0
1824 |     decay_mult: 1.0
1825 |   }
1826 |   convolution_param {
1827 |     num_output: 32
1828 |     bias_term: false
1829 |     pad: 0
1830 |     kernel_size: 1
1831 |     stride: 1
1832 |     weight_filler {
1833 |       type: "msra"
1834 |     }
1835 |     bias_filler {
1836 |       type: "constant"
1837 |       value: 0.0
1838 |     }
1839 |   }
1840 | }
1841 | layer {
1842 |   name: "bn_5_4_linear"
1843 |   type: "BatchNorm"
1844 |   bottom: "conv_5_4_linear"
1845 |   top: "conv_5_4_linear"
1846 |   param {
1847 |     lr_mult: 0.0
1848 |     decay_mult: 0.0
1849 |   }
1850 |   param {
1851 |     lr_mult: 0.0
1852 |     decay_mult: 0.0
1853 |   }
1854 |   param {
1855 |     lr_mult: 0.0
1856 |     decay_mult: 0.0
1857 |   }
1858 |   batch_norm_param {
1859 |     use_global_stats: true
1860 |   }
1861 | }
1862 | layer {
1863 |   name: "scale_5_4_linear"
1864 |   type: "Scale"
1865 |   bottom: "conv_5_4_linear"
1866 |   top: "conv_5_4_linear"
1867 |   scale_param {
1868 |     bias_term: true
1869 |   }
1870 | }
1871 | layer {
1872 |   name: "add_5_4"
1873 |   type: "Eltwise"
1874 |   bottom: "add_5_3"
1875 |   bottom: "conv_5_4_linear"
1876 |   top: "add_5_4"
1877 | }
1878 | layer {
1879 |   name: "conv_6_1_pw"
1880 |   type: "Convolution"
1881 |   bottom: "add_5_4"
1882 |   top: "conv_6_1_pw"
1883 |   param {
1884 |     lr_mult: 1.0
1885 |     decay_mult: 1.0
1886 |   }
1887 |   convolution_param {
1888 |     num_output: 192
1889 |     bias_term: false
1890 |     pad: 0
1891 |     kernel_size: 1
1892 |     stride: 1
1893 |     weight_filler {
1894 |       type: "msra"
1895 |     }
1896 |     bias_filler {
1897 |       type: "constant"
1898 |       value: 0.0
1899 |     }
1900 |   }
1901 | }
1902 | layer {
1903 |   name: "bn_6_1_pw"
1904 |   type: "BatchNorm"
1905 |   bottom: "conv_6_1_pw"
1906 |   top: "conv_6_1_pw"
1907 |   param {
1908 |     lr_mult: 0.0
1909 |     decay_mult: 0.0
1910 |   }
1911 |   param {
1912 |     lr_mult: 0.0
1913 |     decay_mult: 0.0
1914 |   }
1915 |   param {
1916 |     lr_mult: 0.0
1917 |     decay_mult: 0.0
1918 |   }
1919 |   batch_norm_param {
1920 |     use_global_stats: true
1921 |   }
1922 | }
1923 | layer {
1924 |   name: "scale_6_1_pw"
1925 |   type: "Scale"
1926 |   bottom: "conv_6_1_pw"
1927 |   top: "conv_6_1_pw"
1928 |   scale_param {
1929 |     bias_term: true
1930 |   }
1931 | }
1932 | layer {
1933 |   name: "relu_6_1_pw"
1934 |   type: "ReLU"
1935 |   bottom: "conv_6_1_pw"
1936 |   top: "conv_6_1_pw"
1937 | }
1938 | layer {
1939 |   name: "conv_6_1_dw"
1940 |   type: "Convolution"
1941 |   bottom: "conv_6_1_pw"
1942 |   top: "conv_6_1_dw"
1943 |   param {
1944 |     lr_mult: 1.0
1945 |     decay_mult: 1.0
1946 |   }
1947 |   convolution_param {
1948 |     num_output: 192
1949 |     bias_term: false
1950 |     pad: 1
1951 |     kernel_size: 3
1952 |     group: 192
1953 |     stride: 1
1954 |     weight_filler {
1955 |       type: "msra"
1956 |     }
1957 |     bias_filler {
1958 |       type: "constant"
1959 |       value: 0.0
1960 |     }
1961 |     engine: CAFFE
1962 |   }
1963 | }
1964 | layer {
1965 |   name: "bn_6_1_dw"
1966 |   type: "BatchNorm"
1967 |   bottom: "conv_6_1_dw"
1968 |   top: "conv_6_1_dw"
1969 |   param {
1970 |     lr_mult: 0.0
1971 |     decay_mult: 0.0
1972 |   }
1973 |   param {
1974 |     lr_mult: 0.0
1975 |     decay_mult: 0.0
1976 |   }
1977 |   param {
1978 |     lr_mult: 0.0
1979 |     decay_mult: 0.0
1980 |   }
1981 |   batch_norm_param {
1982 |     use_global_stats: true
1983 |   }
1984 | }
1985 | layer {
1986 |   name: "scale_6_1_dw"
1987 |   type: "Scale"
1988 |   bottom: "conv_6_1_dw"
1989 |   top: "conv_6_1_dw"
1990 |   scale_param {
1991 |     bias_term: true
1992 |   }
1993 | }
1994 | layer {
1995 |   name: "relu_6_1_dw"
1996 |   type: "ReLU"
1997 |   bottom: "conv_6_1_dw"
1998 |   top: "conv_6_1_dw"
1999 | }
2000 | layer {
2001 |   name: "conv_6_1_linear"
2002 |   type: "Convolution"
2003 |   bottom: "conv_6_1_dw"
2004 |   top: "conv_6_1_linear"
2005 |   param {
2006 |     lr_mult: 1.0
2007 |     decay_mult: 1.0
2008 |   }
2009 |   convolution_param {
2010 |     num_output: 48
2011 |     bias_term: false
2012 |     pad: 0
2013 |     kernel_size: 1
2014 |     stride: 1
2015 |     weight_filler {
2016 |       type: "msra"
2017 |     }
2018 |     bias_filler {
2019 |       type: "constant"
2020 |       value: 0.0
2021 |     }
2022 |   }
2023 | }
2024 | layer {
2025 |   name: "bn_6_1_linear"
2026 |   type: "BatchNorm"
2027 |   bottom: "conv_6_1_linear"
2028 |   top: "conv_6_1_linear"
2029 |   param {
2030 |     lr_mult: 0.0
2031 |     decay_mult: 0.0
2032 |   }
2033 |   param {
2034 |     lr_mult: 0.0
2035 |     decay_mult: 0.0
2036 |   }
2037 |   param {
2038 |     lr_mult: 0.0
2039 |     decay_mult: 0.0
2040 |   }
2041 |   batch_norm_param {
2042 |     use_global_stats: true
2043 |   }
2044 | }
2045 | layer {
2046 |   name: "scale_6_1_linear"
2047 |   type: "Scale"
2048 |   bottom: "conv_6_1_linear"
2049 |   top: "conv_6_1_linear"
2050 |   scale_param {
2051 |     bias_term: true
2052 |   }
2053 | }
2054 | layer {
2055 |   name: "conv_6_2_pw"
2056 |   type: "Convolution"
2057 |   bottom: "conv_6_1_linear"
2058 |   top: "conv_6_2_pw"
2059 |   param {
2060 |     lr_mult: 1.0
2061 |     decay_mult: 1.0
2062 |   }
2063 |   convolution_param {
2064 |     num_output: 288
2065 |     bias_term: false
2066 |     pad: 0
2067 |     kernel_size: 1
2068 |     stride: 1
2069 |     weight_filler {
2070 |       type: "msra"
2071 |     }
2072 |     bias_filler {
2073 |       type: "constant"
2074 |       value: 0.0
2075 |     }
2076 |   }
2077 | }
2078 | layer {
2079 |   name: "bn_6_2_pw"
2080 |   type: "BatchNorm"
2081 |   bottom: "conv_6_2_pw"
2082 |   top: "conv_6_2_pw"
2083 |   param {
2084 |     lr_mult: 0.0
2085 |     decay_mult: 0.0
2086 |   }
2087 |   param {
2088 |     lr_mult: 0.0
2089 |     decay_mult: 0.0
2090 |   }
2091 |   param {
2092 |     lr_mult: 0.0
2093 |     decay_mult: 0.0
2094 |   }
2095 |   batch_norm_param {
2096 |     use_global_stats: true
2097 |   }
2098 | }
2099 | layer {
2100 |   name: "scale_6_2_pw"
2101 |   type: "Scale"
2102 |   bottom: "conv_6_2_pw"
2103 |   top: "conv_6_2_pw"
2104 |   scale_param {
2105 |     bias_term: true
2106 |   }
2107 | }
2108 | layer {
2109 |   name: "relu_6_2_pw"
2110 |   type: "ReLU"
2111 |   bottom: "conv_6_2_pw"
2112 |   top: "conv_6_2_pw"
2113 | }
2114 | layer {
2115 |   name: "conv_6_2_dw"
2116 |   type: "Convolution"
2117 |   bottom: "conv_6_2_pw"
2118 |   top: "conv_6_2_dw"
2119 |   param {
2120 |     lr_mult: 1.0
2121 |     decay_mult: 1.0
2122 |   }
2123 |   convolution_param {
2124 |     num_output: 288
2125 |     bias_term: false
2126 |     pad: 1
2127 |     kernel_size: 3
2128 |     group: 288
2129 |     stride: 1
2130 |     weight_filler {
2131 |       type: "msra"
2132 |     }
2133 |     bias_filler {
2134 |       type: "constant"
2135 |       value: 0.0
2136 |     }
2137 |     engine: CAFFE
2138 |   }
2139 | }
2140 | layer {
2141 |   name: "bn_6_2_dw"
2142 |   type: "BatchNorm"
2143 |   bottom: "conv_6_2_dw"
2144 |   top: "conv_6_2_dw"
2145 |   param {
2146 |     lr_mult: 0.0
2147 |     decay_mult: 0.0
2148 |   }
2149 |   param {
2150 |     lr_mult: 0.0
2151 |     decay_mult: 0.0
2152 |   }
2153 |   param {
2154 |     lr_mult: 0.0
2155 |     decay_mult: 0.0
2156 |   }
2157 |   batch_norm_param {
2158 |     use_global_stats: true
2159 |   }
2160 | }
2161 | layer {
2162 |   name: "scale_6_2_dw"
2163 |   type: "Scale"
2164 |   bottom: "conv_6_2_dw"
2165 |   top: "conv_6_2_dw"
2166 |   scale_param {
2167 |     bias_term: true
2168 |   }
2169 | }
2170 | layer {
2171 |   name: "relu_6_2_dw"
2172 |   type: "ReLU"
2173 |   bottom: "conv_6_2_dw"
2174 |   top: "conv_6_2_dw"
2175 | }
2176 | layer {
2177 |   name: "conv_6_2_linear"
2178 |   type: "Convolution"
2179 |   bottom: "conv_6_2_dw"
2180 |   top: "conv_6_2_linear"
2181 |   param {
2182 |     lr_mult: 1.0
2183 |     decay_mult: 1.0
2184 |   }
2185 |   convolution_param {
2186 |     num_output: 48
2187 |     bias_term: false
2188 |     pad: 0
2189 |     kernel_size: 1
2190 |     stride: 1
2191 |     weight_filler {
2192 |       type: "msra"
2193 |     }
2194 |     bias_filler {
2195 |       type: "constant"
2196 |       value: 0.0
2197 |     }
2198 |   }
2199 | }
2200 | layer {
2201 |   name: "bn_6_2_linear"
2202 |   type: "BatchNorm"
2203 |   bottom: "conv_6_2_linear"
2204 |   top: "conv_6_2_linear"
2205 |   param {
2206 |     lr_mult: 0.0
2207 |     decay_mult: 0.0
2208 |   }
2209 |   param {
2210 |     lr_mult: 0.0
2211 |     decay_mult: 0.0
2212 |   }
2213 |   param {
2214 |     lr_mult: 0.0
2215 |     decay_mult: 0.0
2216 |   }
2217 |   batch_norm_param {
2218 |     use_global_stats: true
2219 |   }
2220 | }
2221 | layer {
2222 |   name: "scale_6_2_linear"
2223 |   type: "Scale"
2224 |   bottom: "conv_6_2_linear"
2225 |   top: "conv_6_2_linear"
2226 |   scale_param {
2227 |     bias_term: true
2228 |   }
2229 | }
2230 | layer {
2231 |   name: "add_6_2"
2232 |   type: "Eltwise"
2233 |   bottom: "conv_6_1_linear"
2234 |   bottom: "conv_6_2_linear"
2235 |   top: "add_6_2"
2236 | }
2237 | layer {
2238 |   name: "conv_6_3_pw"
2239 |   type: "Convolution"
2240 |   bottom: "add_6_2"
2241 |   top: "conv_6_3_pw"
2242 |   param {
2243 |     lr_mult: 1.0
2244 |     decay_mult: 1.0
2245 |   }
2246 |   convolution_param {
2247 |     num_output: 288
2248 |     bias_term: false
2249 |     pad: 0
2250 |     kernel_size: 1
2251 |     stride: 1
2252 |     weight_filler {
2253 |       type: "msra"
2254 |     }
2255 |     bias_filler {
2256 |       type: "constant"
2257 |       value: 0.0
2258 |     }
2259 |   }
2260 | }
2261 | layer {
2262 |   name: "bn_6_3_pw"
2263 |   type: "BatchNorm"
2264 |   bottom: "conv_6_3_pw"
2265 |   top: "conv_6_3_pw"
2266 |   param {
2267 |     lr_mult: 0.0
2268 |     decay_mult: 0.0
2269 |   }
2270 |   param {
2271 |     lr_mult: 0.0
2272 |     decay_mult: 0.0
2273 |   }
2274 |   param {
2275 |     lr_mult: 0.0
2276 |     decay_mult: 0.0
2277 |   }
2278 |   batch_norm_param {
2279 |     use_global_stats: true
2280 |   }
2281 | }
2282 | layer {
2283 |   name: "scale_6_3_pw"
2284 |   type: "Scale"
2285 |   bottom: "conv_6_3_pw"
2286 |   top: "conv_6_3_pw"
2287 |   scale_param {
2288 |     bias_term: true
2289 |   }
2290 | }
2291 | layer {
2292 |   name: "relu_6_3_pw"
2293 |   type: "ReLU"
2294 |   bottom: "conv_6_3_pw"
2295 |   top: "conv_6_3_pw"
2296 | }
2297 | layer {
2298 |   name: "conv_6_3_dw"
2299 |   type: "Convolution"
2300 |   bottom: "conv_6_3_pw"
2301 |   top: "conv_6_3_dw"
2302 |   param {
2303 |     lr_mult: 1.0
2304 |     decay_mult: 1.0
2305 |   }
2306 |   convolution_param {
2307 |     num_output: 288
2308 |     bias_term: false
2309 |     pad: 1
2310 |     kernel_size: 3
2311 |     group: 288
2312 |     stride: 1
2313 |     weight_filler {
2314 |       type: "msra"
2315 |     }
2316 |     bias_filler {
2317 |       type: "constant"
2318 |       value: 0.0
2319 |     }
2320 |     engine: CAFFE
2321 |   }
2322 | }
2323 | layer {
2324 |   name: "bn_6_3_dw"
2325 |   type: "BatchNorm"
2326 |   bottom: "conv_6_3_dw"
2327 |   top: "conv_6_3_dw"
2328 |   param {
2329 |     lr_mult: 0.0
2330 |     decay_mult: 0.0
2331 |   }
2332 |   param {
2333 |     lr_mult: 0.0
2334 |     decay_mult: 0.0
2335 |   }
2336 |   param {
2337 |     lr_mult: 0.0
2338 |     decay_mult: 0.0
2339 |   }
2340 |   batch_norm_param {
2341 |     use_global_stats: true
2342 |   }
2343 | }
2344 | layer {
2345 |   name: "scale_6_3_dw"
2346 |   type: "Scale"
2347 |   bottom: "conv_6_3_dw"
2348 |   top: "conv_6_3_dw"
2349 |   scale_param {
2350 |     bias_term: true
2351 |   }
2352 | }
2353 | layer {
2354 |   name: "relu_6_3_dw"
2355 |   type: "ReLU"
2356 |   bottom: "conv_6_3_dw"
2357 |   top: "conv_6_3_dw"
2358 | }
2359 | layer {
2360 |   name: "conv_6_3_linear"
2361 |   type: "Convolution"
2362 |   bottom: "conv_6_3_dw"
2363 |   top: "conv_6_3_linear"
2364 |   param {
2365 |     lr_mult: 1.0
2366 |     decay_mult: 1.0
2367 |   }
2368 |   convolution_param {
2369 |     num_output: 48
2370 |     bias_term: false
2371 |     pad: 0
2372 |     kernel_size: 1
2373 |     stride: 1
2374 |     weight_filler {
2375 |       type: "msra"
2376 |     }
2377 |     bias_filler {
2378 |       type: "constant"
2379 |       value: 0.0
2380 |     }
2381 |   }
2382 | }
2383 | layer {
2384 |   name: "bn_6_3_linear"
2385 |   type: "BatchNorm"
2386 |   bottom: "conv_6_3_linear"
2387 |   top: "conv_6_3_linear"
2388 |   param {
2389 |     lr_mult: 0.0
2390 |     decay_mult: 0.0
2391 |   }
2392 |   param {
2393 |     lr_mult: 0.0
2394 |     decay_mult: 0.0
2395 |   }
2396 |   param {
2397 |     lr_mult: 0.0
2398 |     decay_mult: 0.0
2399 |   }
2400 |   batch_norm_param {
2401 |     use_global_stats: true
2402 |   }
2403 | }
2404 | layer {
2405 |   name: "scale_6_3_linear"
2406 |   type: "Scale"
2407 |   bottom: "conv_6_3_linear"
2408 |   top: "conv_6_3_linear"
2409 |   scale_param {
2410 |     bias_term: true
2411 |   }
2412 | }
2413 | layer {
2414 |   name: "add_6_3"
2415 |   type: "Eltwise"
2416 |   bottom: "add_6_2"
2417 |   bottom: "conv_6_3_linear"
2418 |   top: "add_6_3"
2419 | }
2420 | layer {
2421 |   name: "conv_7_1_pw"
2422 |   type: "Convolution"
2423 |   bottom: "add_6_3"
2424 |   top: "conv_7_1_pw"
2425 |   param {
2426 |     lr_mult: 1.0
2427 |     decay_mult: 1.0
2428 |   }
2429 |   convolution_param {
2430 |     num_output: 288
2431 |     bias_term: false
2432 |     pad: 0
2433 |     kernel_size: 1
2434 |     stride: 1
2435 |     weight_filler {
2436 |       type: "msra"
2437 |     }
2438 |     bias_filler {
2439 |       type: "constant"
2440 |       value: 0.0
2441 |     }
2442 |   }
2443 | }
2444 | layer {
2445 |   name: "bn_7_1_pw"
2446 |   type: "BatchNorm"
2447 |   bottom: "conv_7_1_pw"
2448 |   top: "conv_7_1_pw"
2449 |   param {
2450 |     lr_mult: 0.0
2451 |     decay_mult: 0.0
2452 |   }
2453 |   param {
2454 |     lr_mult: 0.0
2455 |     decay_mult: 0.0
2456 |   }
2457 |   param {
2458 |     lr_mult: 0.0
2459 |     decay_mult: 0.0
2460 |   }
2461 |   batch_norm_param {
2462 |     use_global_stats: true
2463 |   }
2464 | }
2465 | layer {
2466 |   name: "scale_7_1_pw"
2467 |   type: "Scale"
2468 |   bottom: "conv_7_1_pw"
2469 |   top: "conv_7_1_pw"
2470 |   scale_param {
2471 |     bias_term: true
2472 |   }
2473 | }
2474 | layer {
2475 |   name: "relu_7_1_pw"
2476 |   type: "ReLU"
2477 |   bottom: "conv_7_1_pw"
2478 |   top: "conv_7_1_pw"
2479 | }
2480 | layer {
2481 |   name: "conv_7_1_dw"
2482 |   type: "Convolution"
2483 |   bottom: "conv_7_1_pw"
2484 |   top: "conv_7_1_dw"
2485 |   param {
2486 |     lr_mult: 1.0
2487 |     decay_mult: 1.0
2488 |   }
2489 |   convolution_param {
2490 |     num_output: 288
2491 |     bias_term: false
2492 |     pad: 1
2493 |     kernel_size: 3
2494 |     group: 288
2495 |     stride: 2
2496 |     weight_filler {
2497 |       type: "msra"
2498 |     }
2499 |     bias_filler {
2500 |       type: "constant"
2501 |       value: 0.0
2502 |     }
2503 |     engine: CAFFE
2504 |   }
2505 | }
2506 | layer {
2507 |   name: "bn_7_1_dw"
2508 |   type: "BatchNorm"
2509 |   bottom: "conv_7_1_dw"
2510 |   top: "conv_7_1_dw"
2511 |   param {
2512 |     lr_mult: 0.0
2513 |     decay_mult: 0.0
2514 |   }
2515 |   param {
2516 |     lr_mult: 0.0
2517 |     decay_mult: 0.0
2518 |   }
2519 |   param {
2520 |     lr_mult: 0.0
2521 |     decay_mult: 0.0
2522 |   }
2523 |   batch_norm_param {
2524 |     use_global_stats: true
2525 |   }
2526 | }
2527 | layer {
2528 |   name: "scale_7_1_dw"
2529 |   type: "Scale"
2530 |   bottom: "conv_7_1_dw"
2531 |   top: "conv_7_1_dw"
2532 |   scale_param {
2533 |     bias_term: true
2534 |   }
2535 | }
2536 | layer {
2537 |   name: "relu_7_1_dw"
2538 |   type: "ReLU"
2539 |   bottom: "conv_7_1_dw"
2540 |   top: "conv_7_1_dw"
2541 | }
2542 | layer {
2543 |   name: "conv_7_1_linear"
2544 |   type: "Convolution"
2545 |   bottom: "conv_7_1_dw"
2546 |   top: "conv_7_1_linear"
2547 |   param {
2548 |     lr_mult: 1.0
2549 |     decay_mult: 1.0
2550 |   }
2551 |   convolution_param {
2552 |     num_output: 80
2553 |     bias_term: false
2554 |     pad: 0
2555 |     kernel_size: 1
2556 |     stride: 1
2557 |     weight_filler {
2558 |       type: "msra"
2559 |     }
2560 |     bias_filler {
2561 |       type: "constant"
2562 |       value: 0.0
2563 |     }
2564 |   }
2565 | }
2566 | layer {
2567 |   name: "bn_7_1_linear"
2568 |   type: "BatchNorm"
2569 |   bottom: "conv_7_1_linear"
2570 |   top: "conv_7_1_linear"
2571 |   param {
2572 |     lr_mult: 0.0
2573 |     decay_mult: 0.0
2574 |   }
2575 |   param {
2576 |     lr_mult: 0.0
2577 |     decay_mult: 0.0
2578 |   }
2579 |   param {
2580 |     lr_mult: 0.0
2581 |     decay_mult: 0.0
2582 |   }
2583 |   batch_norm_param {
2584 |     use_global_stats: true
2585 |   }
2586 | }
2587 | layer {
2588 |   name: "scale_7_1_linear"
2589 |   type: "Scale"
2590 |   bottom: "conv_7_1_linear"
2591 |   top: "conv_7_1_linear"
2592 |   scale_param {
2593 |     bias_term: true
2594 |   }
2595 | }
2596 | layer {
2597 |   name: "conv_7_2_pw"
2598 |   type: "Convolution"
2599 |   bottom: "conv_7_1_linear"
2600 |   top: "conv_7_2_pw"
2601 |   param {
2602 |     lr_mult: 1.0
2603 |     decay_mult: 1.0
2604 |   }
2605 |   convolution_param {
2606 |     num_output: 480
2607 |     bias_term: false
2608 |     pad: 0
2609 |     kernel_size: 1
2610 |     stride: 1
2611 |     weight_filler {
2612 |       type: "msra"
2613 |     }
2614 |     bias_filler {
2615 |       type: "constant"
2616 |       value: 0.0
2617 |     }
2618 |   }
2619 | }
2620 | layer {
2621 |   name: "bn_7_2_pw"
2622 |   type: "BatchNorm"
2623 |   bottom: "conv_7_2_pw"
2624 |   top: "conv_7_2_pw"
2625 |   param {
2626 |     lr_mult: 0.0
2627 |     decay_mult: 0.0
2628 |   }
2629 |   param {
2630 |     lr_mult: 0.0
2631 |     decay_mult: 0.0
2632 |   }
2633 |   param {
2634 |     lr_mult: 0.0
2635 |     decay_mult: 0.0
2636 |   }
2637 |   batch_norm_param {
2638 |     use_global_stats: true
2639 |   }
2640 | }
2641 | layer {
2642 |   name: "scale_7_2_pw"
2643 |   type: "Scale"
2644 |   bottom: "conv_7_2_pw"
2645 |   top: "conv_7_2_pw"
2646 |   scale_param {
2647 |     bias_term: true
2648 |   }
2649 | }
2650 | layer {
2651 |   name: "relu_7_2_pw"
2652 |   type: "ReLU"
2653 |   bottom: "conv_7_2_pw"
2654 |   top: "conv_7_2_pw"
2655 | }
2656 | layer {
2657 |   name: "conv_7_2_dw"
2658 |   type: "Convolution"
2659 |   bottom: "conv_7_2_pw"
2660 |   top: "conv_7_2_dw"
2661 |   param {
2662 |     lr_mult: 1.0
2663 |     decay_mult: 1.0
2664 |   }
2665 |   convolution_param {
2666 |     num_output: 480
2667 |     bias_term: false
2668 |     pad: 1
2669 |     kernel_size: 3
2670 |     group: 480
2671 |     stride: 1
2672 |     weight_filler {
2673 |       type: "msra"
2674 |     }
2675 |     bias_filler {
2676 |       type: "constant"
2677 |       value: 0.0
2678 |     }
2679 |     engine: CAFFE
2680 |   }
2681 | }
2682 | layer {
2683 |   name: "bn_7_2_dw"
2684 |   type: "BatchNorm"
2685 |   bottom: "conv_7_2_dw"
2686 |   top: "conv_7_2_dw"
2687 |   param {
2688 |     lr_mult: 0.0
2689 |     decay_mult: 0.0
2690 |   }
2691 |   param {
2692 |     lr_mult: 0.0
2693 |     decay_mult: 0.0
2694 |   }
2695 |   param {
2696 |     lr_mult: 0.0
2697 |     decay_mult: 0.0
2698 |   }
2699 |   batch_norm_param {
2700 |     use_global_stats: true
2701 |   }
2702 | }
2703 | layer {
2704 |   name: "scale_7_2_dw"
2705 |   type: "Scale"
2706 |   bottom: "conv_7_2_dw"
2707 |   top: "conv_7_2_dw"
2708 |   scale_param {
2709 |     bias_term: true
2710 |   }
2711 | }
2712 | layer {
2713 |   name: "relu_7_2_dw"
2714 |   type: "ReLU"
2715 |   bottom: "conv_7_2_dw"
2716 |   top: "conv_7_2_dw"
2717 | }
2718 | layer {
2719 |   name: "conv_7_2_linear"
2720 |   type: "Convolution"
2721 |   bottom: "conv_7_2_dw"
2722 |   top: "conv_7_2_linear"
2723 |   param {
2724 |     lr_mult: 1.0
2725 |     decay_mult: 1.0
2726 |   }
2727 |   convolution_param {
2728 |     num_output: 80
2729 |     bias_term: false
2730 |     pad: 0
2731 |     kernel_size: 1
2732 |     stride: 1
2733 |     weight_filler {
2734 |       type: "msra"
2735 |     }
2736 |     bias_filler {
2737 |       type: "constant"
2738 |       value: 0.0
2739 |     }
2740 |   }
2741 | }
2742 | layer {
2743 |   name: "bn_7_2_linear"
2744 |   type: "BatchNorm"
2745 |   bottom: "conv_7_2_linear"
2746 |   top: "conv_7_2_linear"
2747 |   param {
2748 |     lr_mult: 0.0
2749 |     decay_mult: 0.0
2750 |   }
2751 |   param {
2752 |     lr_mult: 0.0
2753 |     decay_mult: 0.0
2754 |   }
2755 |   param {
2756 |     lr_mult: 0.0
2757 |     decay_mult: 0.0
2758 |   }
2759 |   batch_norm_param {
2760 |     use_global_stats: true
2761 |   }
2762 | }
2763 | layer {
2764 |   name: "scale_7_2_linear"
2765 |   type: "Scale"
2766 |   bottom: "conv_7_2_linear"
2767 |   top: "conv_7_2_linear"
2768 |   scale_param {
2769 |     bias_term: true
2770 |   }
2771 | }
2772 | layer {
2773 |   name: "add_7_2"
2774 |   type: "Eltwise"
2775 |   bottom: "conv_7_1_linear"
2776 |   bottom: "conv_7_2_linear"
2777 |   top: "add_7_2"
2778 | }
2779 | layer {
2780 |   name: "conv_7_3_pw"
2781 |   type: "Convolution"
2782 |   bottom: "add_7_2"
2783 |   top: "conv_7_3_pw"
2784 |   param {
2785 |     lr_mult: 1.0
2786 |     decay_mult: 1.0
2787 |   }
2788 |   convolution_param {
2789 |     num_output: 480
2790 |     bias_term: false
2791 |     pad: 0
2792 |     kernel_size: 1
2793 |     stride: 1
2794 |     weight_filler {
2795 |       type: "msra"
2796 |     }
2797 |     bias_filler {
2798 |       type: "constant"
2799 |       value: 0.0
2800 |     }
2801 |   }
2802 | }
2803 | layer {
2804 |   name: "bn_7_3_pw"
2805 |   type: "BatchNorm"
2806 |   bottom: "conv_7_3_pw"
2807 |   top: "conv_7_3_pw"
2808 |   param {
2809 |     lr_mult: 0.0
2810 |     decay_mult: 0.0
2811 |   }
2812 |   param {
2813 |     lr_mult: 0.0
2814 |     decay_mult: 0.0
2815 |   }
2816 |   param {
2817 |     lr_mult: 0.0
2818 |     decay_mult: 0.0
2819 |   }
2820 |   batch_norm_param {
2821 |     use_global_stats: true
2822 |   }
2823 | }
2824 | layer {
2825 |   name: "scale_7_3_pw"
2826 |   type: "Scale"
2827 |   bottom: "conv_7_3_pw"
2828 |   top: "conv_7_3_pw"
2829 |   scale_param {
2830 |     bias_term: true
2831 |   }
2832 | }
2833 | layer {
2834 |   name: "relu_7_3_pw"
2835 |   type: "ReLU"
2836 |   bottom: "conv_7_3_pw"
2837 |   top: "conv_7_3_pw"
2838 | }
2839 | layer {
2840 |   name: "conv_7_3_dw"
2841 |   type: "Convolution"
2842 |   bottom: "conv_7_3_pw"
2843 |   top: "conv_7_3_dw"
2844 |   param {
2845 |     lr_mult: 1.0
2846 |     decay_mult: 1.0
2847 |   }
2848 |   convolution_param {
2849 |     num_output: 480
2850 |     bias_term: false
2851 |     pad: 1
2852 |     kernel_size: 3
2853 |     group: 480
2854 |     stride: 1
2855 |     weight_filler {
2856 |       type: "msra"
2857 |     }
2858 |     bias_filler {
2859 |       type: "constant"
2860 |       value: 0.0
2861 |     }
2862 |     engine: CAFFE
2863 |   }
2864 | }
2865 | layer {
2866 |   name: "bn_7_3_dw"
2867 |   type: "BatchNorm"
2868 |   bottom: "conv_7_3_dw"
2869 |   top: "conv_7_3_dw"
2870 |   param {
2871 |     lr_mult: 0.0
2872 |     decay_mult: 0.0
2873 |   }
2874 |   param {
2875 |     lr_mult: 0.0
2876 |     decay_mult: 0.0
2877 |   }
2878 |   param {
2879 |     lr_mult: 0.0
2880 |     decay_mult: 0.0
2881 |   }
2882 |   batch_norm_param {
2883 |     use_global_stats: true
2884 |   }
2885 | }
2886 | layer {
2887 |   name: "scale_7_3_dw"
2888 |   type: "Scale"
2889 |   bottom: "conv_7_3_dw"
2890 |   top: "conv_7_3_dw"
2891 |   scale_param {
2892 |     bias_term: true
2893 |   }
2894 | }
2895 | layer {
2896 |   name: "relu_7_3_dw"
2897 |   type: "ReLU"
2898 |   bottom: "conv_7_3_dw"
2899 |   top: "conv_7_3_dw"
2900 | }
2901 | layer {
2902 |   name: "conv_7_3_linear"
2903 |   type: "Convolution"
2904 |   bottom: "conv_7_3_dw"
2905 |   top: "conv_7_3_linear"
2906 |   param {
2907 |     lr_mult: 1.0
2908 |     decay_mult: 1.0
2909 |   }
2910 |   convolution_param {
2911 |     num_output: 80
2912 |     bias_term: false
2913 |     pad: 0
2914 |     kernel_size: 1
2915 |     stride: 1
2916 |     weight_filler {
2917 |       type: "msra"
2918 |     }
2919 |     bias_filler {
2920 |       type: "constant"
2921 |       value: 0.0
2922 |     }
2923 |   }
2924 | }
2925 | layer {
2926 |   name: "bn_7_3_linear"
2927 |   type: "BatchNorm"
2928 |   bottom: "conv_7_3_linear"
2929 |   top: "conv_7_3_linear"
2930 |   param {
2931 |     lr_mult: 0.0
2932 |     decay_mult: 0.0
2933 |   }
2934 |   param {
2935 |     lr_mult: 0.0
2936 |     decay_mult: 0.0
2937 |   }
2938 |   param {
2939 |     lr_mult: 0.0
2940 |     decay_mult: 0.0
2941 |   }
2942 |   batch_norm_param {
2943 |     use_global_stats: true
2944 |   }
2945 | }
2946 | layer {
2947 |   name: "scale_7_3_linear"
2948 |   type: "Scale"
2949 |   bottom: "conv_7_3_linear"
2950 |   top: "conv_7_3_linear"
2951 |   scale_param {
2952 |     bias_term: true
2953 |   }
2954 | }
2955 | layer {
2956 |   name: "add_7_3"
2957 |   type: "Eltwise"
2958 |   bottom: "add_7_2"
2959 |   bottom: "conv_7_3_linear"
2960 |   top: "add_7_3"
2961 | }
2962 | layer {
2963 |   name: "conv_8_1_pw"
2964 |   type: "Convolution"
2965 |   bottom: "add_7_3"
2966 |   top: "conv_8_1_pw"
2967 |   param {
2968 |     lr_mult: 1.0
2969 |     decay_mult: 1.0
2970 |   }
2971 |   convolution_param {
2972 |     num_output: 480
2973 |     bias_term: false
2974 |     pad: 0
2975 |     kernel_size: 1
2976 |     stride: 1
2977 |     weight_filler {
2978 |       type: "msra"
2979 |     }
2980 |     bias_filler {
2981 |       type: "constant"
2982 |       value: 0.0
2983 |     }
2984 |   }
2985 | }
2986 | layer {
2987 |   name: "bn_8_1_pw"
2988 |   type: "BatchNorm"
2989 |   bottom: "conv_8_1_pw"
2990 |   top: "conv_8_1_pw"
2991 |   param {
2992 |     lr_mult: 0.0
2993 |     decay_mult: 0.0
2994 |   }
2995 |   param {
2996 |     lr_mult: 0.0
2997 |     decay_mult: 0.0
2998 |   }
2999 |   param {
3000 |     lr_mult: 0.0
3001 |     decay_mult: 0.0
3002 |   }
3003 |   batch_norm_param {
3004 |     use_global_stats: true
3005 |   }
3006 | }
3007 | layer {
3008 |   name: "scale_8_1_pw"
3009 |   type: "Scale"
3010 |   bottom: "conv_8_1_pw"
3011 |   top: "conv_8_1_pw"
3012 |   scale_param {
3013 |     bias_term: true
3014 |   }
3015 | }
3016 | layer {
3017 |   name: "relu_8_1_pw"
3018 |   type: "ReLU"
3019 |   bottom: "conv_8_1_pw"
3020 |   top: "conv_8_1_pw"
3021 | }
3022 | layer {
3023 |   name: "conv_8_1_dw"
3024 |   type: "Convolution"
3025 |   bottom: "conv_8_1_pw"
3026 |   top: "conv_8_1_dw"
3027 |   param {
3028 |     lr_mult: 1.0
3029 |     decay_mult: 1.0
3030 |   }
3031 |   convolution_param {
3032 |     num_output: 480
3033 |     bias_term: false
3034 |     pad: 1
3035 |     kernel_size: 3
3036 |     group: 480
3037 |     stride: 1
3038 |     weight_filler {
3039 |       type: "msra"
3040 |     }
3041 |     bias_filler {
3042 |       type: "constant"
3043 |       value: 0.0
3044 |     }
3045 |     engine: CAFFE
3046 |   }
3047 | }
3048 | layer {
3049 |   name: "bn_8_1_dw"
3050 |   type: "BatchNorm"
3051 |   bottom: "conv_8_1_dw"
3052 |   top: "conv_8_1_dw"
3053 |   param {
3054 |     lr_mult: 0.0
3055 |     decay_mult: 0.0
3056 |   }
3057 |   param {
3058 |     lr_mult: 0.0
3059 |     decay_mult: 0.0
3060 |   }
3061 |   param {
3062 |     lr_mult: 0.0
3063 |     decay_mult: 0.0
3064 |   }
3065 |   batch_norm_param {
3066 |     use_global_stats: true
3067 |   }
3068 | }
3069 | layer {
3070 |   name: "scale_8_1_dw"
3071 |   type: "Scale"
3072 |   bottom: "conv_8_1_dw"
3073 |   top: "conv_8_1_dw"
3074 |   scale_param {
3075 |     bias_term: true
3076 |   }
3077 | }
3078 | layer {
3079 |   name: "relu_8_1_dw"
3080 |   type: "ReLU"
3081 |   bottom: "conv_8_1_dw"
3082 |   top: "conv_8_1_dw"
3083 | }
3084 | layer {
3085 |   name: "conv_8_1_linear"
3086 |   type: "Convolution"
3087 |   bottom: "conv_8_1_dw"
3088 |   top: "conv_8_1_linear"
3089 |   param {
3090 |     lr_mult: 1.0
3091 |     decay_mult: 1.0
3092 |   }
3093 |   convolution_param {
3094 |     num_output: 160
3095 |     bias_term: false
3096 |     pad: 0
3097 |     kernel_size: 1
3098 |     stride: 1
3099 |     weight_filler {
3100 |       type: "msra"
3101 |     }
3102 |     bias_filler {
3103 |       type: "constant"
3104 |       value: 0.0
3105 |     }
3106 |   }
3107 | }
3108 | layer {
3109 |   name: "bn_8_1_linear"
3110 |   type: "BatchNorm"
3111 |   bottom: "conv_8_1_linear"
3112 |   top: "conv_8_1_linear"
3113 |   param {
3114 |     lr_mult: 0.0
3115 |     decay_mult: 0.0
3116 |   }
3117 |   param {
3118 |     lr_mult: 0.0
3119 |     decay_mult: 0.0
3120 |   }
3121 |   param {
3122 |     lr_mult: 0.0
3123 |     decay_mult: 0.0
3124 |   }
3125 |   batch_norm_param {
3126 |     use_global_stats: true
3127 |   }
3128 | }
3129 | layer {
3130 |   name: "scale_8_1_linear"
3131 |   type: "Scale"
3132 |   bottom: "conv_8_1_linear"
3133 |   top: "conv_8_1_linear"
3134 |   scale_param {
3135 |     bias_term: true
3136 |   }
3137 | }
3138 | layer {
3139 |   name: "conv_conv9"
3140 |   type: "Convolution"
3141 |   bottom: "conv_8_1_linear"
3142 |   top: "conv_conv9"
3143 |   param {
3144 |     lr_mult: 1.0
3145 |     decay_mult: 1.0
3146 |   }
3147 |   convolution_param {
3148 |     num_output: 1280
3149 |     bias_term: true
3150 |     pad: 0
3151 |     kernel_size: 1
3152 |     stride: 1
3153 |     weight_filler {
3154 |       type: "msra"
3155 |     }
3156 |     bias_filler {
3157 |       type: "constant"
3158 |       value: 0.0
3159 |     }
3160 |   }
3161 | }
3162 | layer {
3163 |   name: "bn_conv9"
3164 |   type: "BatchNorm"
3165 |   bottom: "conv_conv9"
3166 |   top: "conv_conv9"
3167 |   param {
3168 |     lr_mult: 0.0
3169 |     decay_mult: 0.0
3170 |   }
3171 |   param {
3172 |     lr_mult: 0.0
3173 |     decay_mult: 0.0
3174 |   }
3175 |   param {
3176 |     lr_mult: 0.0
3177 |     decay_mult: 0.0
3178 |   }
3179 |   batch_norm_param {
3180 |     use_global_stats: true
3181 |   }
3182 | }
3183 | layer {
3184 |   name: "scale_conv9"
3185 |   type: "Scale"
3186 |   bottom: "conv_conv9"
3187 |   top: "conv_conv9"
3188 |   scale_param {
3189 |     bias_term: true
3190 |   }
3191 | }
3192 | layer {
3193 |   name: "relu_conv9"
3194 |   type: "ReLU"
3195 |   bottom: "conv_conv9"
3196 |   top: "conv_conv9"
3197 | }
3198 | layer {
3199 |   name: "pool9"
3200 |   type: "Pooling"
3201 |   bottom: "conv_conv9"
3202 |   top: "pool9"
3203 |   pooling_param {
3204 |     pool: AVE
3205 |     stride: 1
3206 |     global_pooling: true
3207 |   }
3208 | }
3209 | layer {
3210 |   name: "conv10"
3211 |   type: "Convolution"
3212 |   bottom: "pool9"
3213 |   top: "conv10"
3214 |   param {
3215 |     lr_mult: 1.0
3216 |     decay_mult: 1.0
3217 |   }
3218 |   convolution_param {
3219 |     num_output: 1000
3220 |     pad: 0
3221 |     kernel_size: 1
3222 |     stride: 1
3223 |     weight_filler {
3224 |       type: "msra"
3225 |     }
3226 |     bias_filler {
3227 |       type: "constant"
3228 |       value: 0.0
3229 |     }
3230 |   }
3231 | }
3232 | layer {
3233 |   name: "loss"
3234 |   type: "SoftmaxWithLoss"
3235 |   bottom: "conv10"
3236 |   bottom: "label"
3237 |   top: "loss"
3238 | }
3239 | 
3240 | 


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