ResNet152网络复现(Caffe)

2019年12月08日 阅读数:403
这篇文章主要向大家介绍ResNet152网络复现(Caffe),主要内容包括基础应用、实用技巧、原理机制等方面,希望对大家有所帮助。

1、准备数据集python

1)  下载数据集linux

Imagnet网站上下载了三类图片,分别是big cat、dog、fish,其中训练集的图片数一共是4149,测试集的图片数是1003,训练集和测试集的图片数比例4:1,将训练集的图片保存在train文件夹下,测试集图片保存在val文件夹下.windows

train、val文件夹下面均有bigcat、dog、fish三个文件夹,分别存放着对应类别的图片.网络

 

2) 利用python代码,生成train.txt、val.txtapp

train.txt、val.txt分别存储着训练集和测试集图片的文件名及其类别标签(注:bigcat:0、 dog:一、fish:2),格式以下:dom

n02084071_9865.JPEG 1
n02512053_3388.JPEG 2
n02512053_6294.JPEG 2
n02512053_2413.JPEG 2
n02084071_5655.JPEG 1
n02127808_9965.JPEG 0
n02127808_8206.JPEG 0
n02127808_4887.JPEG 0
n02512053_1952.JPEG 2

     Python代码以下(能够先在windows环境下利用如下的代码生成txt):ide

 1 import os
 2 import random
 3 
 4 
 5 trainPath = 'F:\\Resnet152\\train\\' 
 6 valPath = 'F:\\Resnet152\\val\\' 
 7 
 8 train = {}
 9 val = {}
10 
11     
12 # add
13 for name in os.listdir(trainPath + "bigcat\\"):
14     train[name] = 0
15      
16 # add 
17 for name in os.listdir(trainPath + "dog\\"):
18     train[name] = 1
19     
20 # add 
21 for name in os.listdir(trainPath + "fish\\"):
22     train[name] = 2
23 
25 
26 # add
27 for name in os.listdir(valPath + "bigcat\\"):
28     val[name] = 0
29      
30 # add
31 for name in os.listdir(valPath + "dog\\"):
32     val[name] = 1
33     
34 
35 # add
36 for name in os.listdir(valPath + "fish\\"):
37     val[name] = 2
38     
39     
41 ftrain = open("F:\\Resnet152\\train\\train.txt", 'w')
42 fval = open("F:\\Resnet152\\val\\val.txt", 'w')
43 
44 trainName = []
45 valName = []
46 for (item) in train:
47     trainName.append(item)
48 
49 for item in val:
50     valName.append(item)
51 
52 random.shuffle(trainName)
53 random.shuffle(valName)
54 
55 for name in trainName:
56     label = train[name]
57     ftrain.write(name + " " + str(label) + "\n")
58 
59 for name in valName:
60     label = val[name]
61     fval.write(name + " " + str(label) + "\n")
62 
63 ftrain.close() 
64 fval.close()   

 

3) 生成数据集函数

编写lmdb.sh脚本文件,利用train.txt、val.txt生成train_lmdb、val_lmdb测试

lmdb.sh脚本代码以下网站

 1 #!/usr/bin/env sh
 2 # Create the face_48 lmdb inputs
 3 # N.B. set the path to the face_48 train + val data dirs
 4 
 5 EXAMPLE=/home/wy/ResNet152    #lmdb生成后存放目录
 6 DATA=/home/wy/ResNet152          #train.txt、val.txt存放目录
 7 TOOLS=/home/wy/caffe/build/tools #caffe安装目录
 8 
 9 TRAIN_DATA_ROOT=/home/wy/ResNet152/train/
10 VAL_DATA_ROOT=/home/wy/ResNet152/val/
11 
12 # Set RESIZE=true to resize the images to 224 x 224. Leave as false if images have
13 # already been resized using another tool.
14 RESIZE=true
15 if $RESIZE; then
16   RESIZE_HEIGHT=224   #resize图片大小
17   RESIZE_WIDTH=224
18 else
19   RESIZE_HEIGHT=0
20   RESIZE_WIDTH=0
21 fi
22 
23 if [ ! -d "$TRAIN_DATA_ROOT" ]; then
24   echo "Error: TRAIN_DATA_ROOT is not a path to a directory: $TRAIN_DATA_ROOT"
25   echo "Set the TRAIN_DATA_ROOT variable in create_face_48.sh to the path" \
26        "where the face_48 training data is stored."
27   exit 1
28 fi
29 
30 if [ ! -d "$VAL_DATA_ROOT" ]; then
31   echo "Error: VAL_DATA_ROOT is not a path to a directory: $VAL_DATA_ROOT"
32   echo "Set the VAL_DATA_ROOT variable in create_face_48.sh to the path" \
33        "where the face_48 validation data is stored."
34   exit 1
35 fi
36 
37 echo "Creating train lmdb..."
38 
39 GLOG_logtostderr=1 $TOOLS/convert_imageset \
40     --resize_height=$RESIZE_HEIGHT \
41     --resize_width=$RESIZE_WIDTH \
42     --shuffle \
43     $TRAIN_DATA_ROOT \
44     $DATA/train.txt \
45     $EXAMPLE/train_lmdb
46 
47 echo "Creating val lmdb..."
48 
49 GLOG_logtostderr=1 $TOOLS/convert_imageset \
50     --resize_height=$RESIZE_HEIGHT \
51     --resize_width=$RESIZE_WIDTH \
52     --shuffle \
53     $VAL_DATA_ROOT \
54     $DATA/val.txt \
55     $EXAMPLE/val_lmdb
56 
57 echo "Done."
58 Status API Training Shop Blog About

注:上述代码中有RESIZE_HEIGHT=22四、RESIZE_WIDTH=224,由于使用的是ResNet网络,

须要统一图片大小,若是使用的网络对图片大小无要求,或者已是224*224大小的图片,可不进行resize操做

linux系统终端中执行sh lmdb.sh便可生成lmdb文件,完成数据集的生成工做

执行结束后,会在ResNet152目录下,生成两个文件夹,分别是train_lmdb、val_lmdb

 

4) 计算图片平均值

训练网络图片时,须要对图片作减均值处理,先将图片平均值保存到文件中,在训练网络中直接引用,先介绍如何生成图片平均值文件,后面引用的时候会特别注明,命令以下:

/home/wy/caffe/build/tools/compute_image_mean /home/wy/ResNet152/train_lmdb /home/wy/ResNet152/mean.binaryproto

      利用Caffe安装目录下面的compute_image_mean.bin进行图片平均值的计算,输入的是train_lmdb,输出的为mean.binaryproto

      

     

 

2、修改网络配置文件

修改配置以前,先列出须要的文件:train_val.prototxt、deploy.prototxt、solver.prototxt(三个文件的具体内容会在博客最后给出)

1)  train_val.prototxt

文件名train_val包括train和val,能够先这样理解每进行500次训练的时候,会进行一次验证,方便输出供训练者观察状况,既然这样,就须要在这个文件里面指定train数据集和val数据集

      

       

       关于train_val.prototxt中其他部分的代码均是ResNet152的网络结构代码,无需修改

 

2)  deploy.prototxt

deploy.prototxt主要是当模型训练出来之后,利用模型对用户提交的一张图片进行分类应用的时候使用的网络文件,只须要进行一次前向传播计算输入图片所属类别的几率值,因此,此文件里面没有损失函数层的定义

 

3) solver.prototxt

文件指定训练的相关规则和参数,具体内容介绍见下图中的标注

 

3、训练网络

网络文件编辑完后,便可开始训练网络,可直接在linux终端输入如下命令

/home/wy/caffe/build/tools/caffe train --solver=/home/wy/ResNet152/solver.prototxt

前者是caffe安装目录路径,后者是solver.prototxt存放路径,回车后便可开始训练;也可建立train.sh脚本文件,文件中保存上述指令

此时,在linux终端(导航到train.sh存放的路径),输入sh  train.sh便可

训练过程当中的一张图

 

4、测试图片

用训练出的model进行图片的预测

/home/wy/caffe/build/examples/cpp_classification/classification.bin /home/wy/ResNet152/deploy.prototxt /home/wy/ResNet152/model/solver_iter_60000.caffemodel /home/wy/ResNet152/mean.binaryproto /home/wy/ResNet152/class_name.txt /home/wy/ResNet152/testImages/ISIC_0000001.jpg

在linux下输入上述命令,便可对图片’ISIC_0000001.jpg’进行预测

其中,class_name.txt文件中内容(和trainx.txt中标签要对应):

0 bigcat  
1 dog  
2 fish 

测试图片:

看下测试结果(测试的时候换了台电脑,因此发现目录不太同样):

因此,属于bigcat的几率为0.9999,属于fish的几率为0.0001,属于dog的几率为0

 

train_val.prototxt

   1 name: "ResNet-152"
   2 layer {
   3     name: "data"
   4     type: "Data"
   5     top: "data"
   6     top: "label"
   7     include {
   8         phase: TRAIN
   9     }
  10     transform_param {
  11         mirror: true
  12         crop_size: 224
  13         mean_file: "/home/wy/ResNet152/mean.binaryproto"
  14     }
  15     data_param {
  16         source: "/home/wy/ResNet152/train_lmdb"
  17         batch_size: 1
  18         backend: LMDB
  19     }
  20 }
  21 layer {
  22     name: "data"
  23     type: "Data"
  24     top: "data"
  25     top: "label"
  26     include {
  27         phase: TEST
  28     }
  29     transform_param {
  30         mirror: false
  31         crop_size: 224
  32         mean_file:"/usr/develop/repertory/ResNet152/mean.binaryproto"
  33     }
  34     data_param {
  35         source: "/usr/develop/repertory/ResNet152/val_lmdb"
  36         batch_size: 1
  37         backend: LMDB
  38     }
  39 }
  40 
  41 layer {
  42     bottom: "data"
  43     top: "conv1"
  44     name: "conv1"
  45     type: "Convolution"
  46     convolution_param {
  47         num_output: 64
  48         kernel_size: 7
  49         pad: 3
  50         stride: 2
  51         weight_filler {
  52             type: "msra"
  53         }
  54         bias_term: false
  55 
  56     }
  57 }
  58 
  59 layer {
  60     bottom: "conv1"
  61     top: "conv1"
  62     name: "bn_conv1"
  63     type: "BatchNorm"
  64     batch_norm_param {
  65         use_global_stats: false
  66     }
  67 }
  68 
  69 layer {
  70     bottom: "conv1"
  71     top: "conv1"
  72     name: "scale_conv1"
  73     type: "Scale"
  74     scale_param {
  75         bias_term: true
  76     }
  77 }
  78 
  79 layer {
  80     bottom: "conv1"
  81     top: "conv1"
  82     name: "conv1_relu"
  83     type: "ReLU"
  84 }
  85 
  86 layer {
  87     bottom: "conv1"
  88     top: "pool1"
  89     name: "pool1"
  90     type: "Pooling"
  91     pooling_param {
  92         kernel_size: 3
  93         stride: 2
  94         pool: MAX
  95     }
  96 }
  97 
  98 layer {
  99     bottom: "pool1"
 100     top: "res2a_branch1"
 101     name: "res2a_branch1"
 102     type: "Convolution"
 103     convolution_param {
 104         num_output: 256
 105         kernel_size: 1
 106         pad: 0
 107         stride: 1
 108         weight_filler {
 109             type: "msra"
 110         }
 111         bias_term: false
 112 
 113     }
 114 }
 115 
 116 layer {
 117     bottom: "res2a_branch1"
 118     top: "res2a_branch1"
 119     name: "bn2a_branch1"
 120     type: "BatchNorm"
 121     batch_norm_param {
 122         use_global_stats: false
 123     }
 124 }
 125 
 126 layer {
 127     bottom: "res2a_branch1"
 128     top: "res2a_branch1"
 129     name: "scale2a_branch1"
 130     type: "Scale"
 131     scale_param {
 132         bias_term: true
 133     }
 134 }
 135 
 136 layer {
 137     bottom: "pool1"
 138     top: "res2a_branch2a"
 139     name: "res2a_branch2a"
 140     type: "Convolution"
 141     convolution_param {
 142         num_output: 64
 143         kernel_size: 1
 144         pad: 0
 145         stride: 1
 146         weight_filler {
 147             type: "msra"
 148         }
 149         bias_term: false
 150 
 151     }
 152 }
 153 
 154 layer {
 155     bottom: "res2a_branch2a"
 156     top: "res2a_branch2a"
 157     name: "bn2a_branch2a"
 158     type: "BatchNorm"
 159     batch_norm_param {
 160         use_global_stats: false
 161     }
 162 }
 163 
 164 layer {
 165     bottom: "res2a_branch2a"
 166     top: "res2a_branch2a"
 167     name: "scale2a_branch2a"
 168     type: "Scale"
 169     scale_param {
 170         bias_term: true
 171     }
 172 }
 173 
 174 layer {
 175     bottom: "res2a_branch2a"
 176     top: "res2a_branch2a"
 177     name: "res2a_branch2a_relu"
 178     type: "ReLU"
 179 }
 180 
 181 layer {
 182     bottom: "res2a_branch2a"
 183     top: "res2a_branch2b"
 184     name: "res2a_branch2b"
 185     type: "Convolution"
 186     convolution_param {
 187         num_output: 64
 188         kernel_size: 3
 189         pad: 1
 190         stride: 1
 191         weight_filler {
 192             type: "msra"
 193         }
 194         bias_term: false
 195 
 196     }
 197 }
 198 
 199 layer {
 200     bottom: "res2a_branch2b"
 201     top: "res2a_branch2b"
 202     name: "bn2a_branch2b"
 203     type: "BatchNorm"
 204     batch_norm_param {
 205         use_global_stats: false
 206     }
 207 }
 208 
 209 layer {
 210     bottom: "res2a_branch2b"
 211     top: "res2a_branch2b"
 212     name: "scale2a_branch2b"
 213     type: "Scale"
 214     scale_param {
 215         bias_term: true
 216     }
 217 }
 218 
 219 layer {
 220     bottom: "res2a_branch2b"
 221     top: "res2a_branch2b"
 222     name: "res2a_branch2b_relu"
 223     type: "ReLU"
 224 }
 225 
 226 layer {
 227     bottom: "res2a_branch2b"
 228     top: "res2a_branch2c"
 229     name: "res2a_branch2c"
 230     type: "Convolution"
 231     convolution_param {
 232         num_output: 256
 233         kernel_size: 1
 234         pad: 0
 235         stride: 1
 236         weight_filler {
 237             type: "msra"
 238         }
 239         bias_term: false
 240 
 241     }
 242 }
 243 
 244 layer {
 245     bottom: "res2a_branch2c"
 246     top: "res2a_branch2c"
 247     name: "bn2a_branch2c"
 248     type: "BatchNorm"
 249     batch_norm_param {
 250         use_global_stats: false
 251     }
 252 }
 253 
 254 layer {
 255     bottom: "res2a_branch2c"
 256     top: "res2a_branch2c"
 257     name: "scale2a_branch2c"
 258     type: "Scale"
 259     scale_param {
 260         bias_term: true
 261     }
 262 }
 263 
 264 layer {
 265     bottom: "res2a_branch1"
 266     bottom: "res2a_branch2c"
 267     top: "res2a"
 268     name: "res2a"
 269     type: "Eltwise"
 270     eltwise_param {
 271         operation: SUM
 272     }
 273 }
 274 
 275 layer {
 276     bottom: "res2a"
 277     top: "res2a"
 278     name: "res2a_relu"
 279     type: "ReLU"
 280 }
 281 
 282 layer {
 283     bottom: "res2a"
 284     top: "res2b_branch2a"
 285     name: "res2b_branch2a"
 286     type: "Convolution"
 287     convolution_param {
 288         num_output: 64
 289         kernel_size: 1
 290         pad: 0
 291         stride: 1
 292         weight_filler {
 293             type: "msra"
 294         }
 295         bias_term: false
 296 
 297     }
 298 }
 299 
 300 layer {
 301     bottom: "res2b_branch2a"
 302     top: "res2b_branch2a"
 303     name: "bn2b_branch2a"
 304     type: "BatchNorm"
 305     batch_norm_param {
 306         use_global_stats: false
 307     }
 308 }
 309 
 310 layer {
 311     bottom: "res2b_branch2a"
 312     top: "res2b_branch2a"
 313     name: "scale2b_branch2a"
 314     type: "Scale"
 315     scale_param {
 316         bias_term: true
 317     }
 318 }
 319 
 320 layer {
 321     bottom: "res2b_branch2a"
 322     top: "res2b_branch2a"
 323     name: "res2b_branch2a_relu"
 324     type: "ReLU"
 325 }
 326 
 327 layer {
 328     bottom: "res2b_branch2a"
 329     top: "res2b_branch2b"
 330     name: "res2b_branch2b"
 331     type: "Convolution"
 332     convolution_param {
 333         num_output: 64
 334         kernel_size: 3
 335         pad: 1
 336         stride: 1
 337         weight_filler {
 338             type: "msra"
 339         }
 340         bias_term: false
 341 
 342     }
 343 }
 344 
 345 layer {
 346     bottom: "res2b_branch2b"
 347     top: "res2b_branch2b"
 348     name: "bn2b_branch2b"
 349     type: "BatchNorm"
 350     batch_norm_param {
 351         use_global_stats: false
 352     }
 353 }
 354 
 355 layer {
 356     bottom: "res2b_branch2b"
 357     top: "res2b_branch2b"
 358     name: "scale2b_branch2b"
 359     type: "Scale"
 360     scale_param {
 361         bias_term: true
 362     }
 363 }
 364 
 365 layer {
 366     bottom: "res2b_branch2b"
 367     top: "res2b_branch2b"
 368     name: "res2b_branch2b_relu"
 369     type: "ReLU"
 370 }
 371 
 372 layer {
 373     bottom: "res2b_branch2b"
 374     top: "res2b_branch2c"
 375     name: "res2b_branch2c"
 376     type: "Convolution"
 377     convolution_param {
 378         num_output: 256
 379         kernel_size: 1
 380         pad: 0
 381         stride: 1
 382         weight_filler {
 383             type: "msra"
 384         }
 385         bias_term: false
 386 
 387     }
 388 }
 389 
 390 layer {
 391     bottom: "res2b_branch2c"
 392     top: "res2b_branch2c"
 393     name: "bn2b_branch2c"
 394     type: "BatchNorm"
 395     batch_norm_param {
 396         use_global_stats: false
 397     }
 398 }
 399 
 400 layer {
 401     bottom: "res2b_branch2c"
 402     top: "res2b_branch2c"
 403     name: "scale2b_branch2c"
 404     type: "Scale"
 405     scale_param {
 406         bias_term: true
 407     }
 408 }
 409 
 410 layer {
 411     bottom: "res2a"
 412     bottom: "res2b_branch2c"
 413     top: "res2b"
 414     name: "res2b"
 415     type: "Eltwise"
 416     eltwise_param {
 417         operation: SUM
 418     }
 419 }
 420 
 421 layer {
 422     bottom: "res2b"
 423     top: "res2b"
 424     name: "res2b_relu"
 425     type: "ReLU"
 426 }
 427 
 428 layer {
 429     bottom: "res2b"
 430     top: "res2c_branch2a"
 431     name: "res2c_branch2a"
 432     type: "Convolution"
 433     convolution_param {
 434         num_output: 64
 435         kernel_size: 1
 436         pad: 0
 437         stride: 1
 438         weight_filler {
 439             type: "msra"
 440         }
 441         bias_term: false
 442 
 443     }
 444 }
 445 
 446 layer {
 447     bottom: "res2c_branch2a"
 448     top: "res2c_branch2a"
 449     name: "bn2c_branch2a"
 450     type: "BatchNorm"
 451     batch_norm_param {
 452         use_global_stats: false
 453     }
 454 }
 455 
 456 layer {
 457     bottom: "res2c_branch2a"
 458     top: "res2c_branch2a"
 459     name: "scale2c_branch2a"
 460     type: "Scale"
 461     scale_param {
 462         bias_term: true
 463     }
 464 }
 465 
 466 layer {
 467     bottom: "res2c_branch2a"
 468     top: "res2c_branch2a"
 469     name: "res2c_branch2a_relu"
 470     type: "ReLU"
 471 }
 472 
 473 layer {
 474     bottom: "res2c_branch2a"
 475     top: "res2c_branch2b"
 476     name: "res2c_branch2b"
 477     type: "Convolution"
 478     convolution_param {
 479         num_output: 64
 480         kernel_size: 3
 481         pad: 1
 482         stride: 1
 483         weight_filler {
 484             type: "msra"
 485         }
 486         bias_term: false
 487 
 488     }
 489 }
 490 
 491 layer {
 492     bottom: "res2c_branch2b"
 493     top: "res2c_branch2b"
 494     name: "bn2c_branch2b"
 495     type: "BatchNorm"
 496     batch_norm_param {
 497         use_global_stats: false
 498     }
 499 }
 500 
 501 layer {
 502     bottom: "res2c_branch2b"
 503     top: "res2c_branch2b"
 504     name: "scale2c_branch2b"
 505     type: "Scale"
 506     scale_param {
 507         bias_term: true
 508     }
 509 }
 510 
 511 layer {
 512     bottom: "res2c_branch2b"
 513     top: "res2c_branch2b"
 514     name: "res2c_branch2b_relu"
 515     type: "ReLU"
 516 }
 517 
 518 layer {
 519     bottom: "res2c_branch2b"
 520     top: "res2c_branch2c"
 521     name: "res2c_branch2c"
 522     type: "Convolution"
 523     convolution_param {
 524         num_output: 256
 525         kernel_size: 1
 526         pad: 0
 527         stride: 1
 528         weight_filler {
 529             type: "msra"
 530         }
 531         bias_term: false
 532 
 533     }
 534 }
 535 
 536 layer {
 537     bottom: "res2c_branch2c"
 538     top: "res2c_branch2c"
 539     name: "bn2c_branch2c"
 540     type: "BatchNorm"
 541     batch_norm_param {
 542         use_global_stats: false
 543     }
 544 }
 545 
 546 layer {
 547     bottom: "res2c_branch2c"
 548     top: "res2c_branch2c"
 549     name: "scale2c_branch2c"
 550     type: "Scale"
 551     scale_param {
 552         bias_term: true
 553     }
 554 }
 555 
 556 layer {
 557     bottom: "res2b"
 558     bottom: "res2c_branch2c"
 559     top: "res2c"
 560     name: "res2c"
 561     type: "Eltwise"
 562     eltwise_param {
 563         operation: SUM
 564     }
 565 }
 566 
 567 layer {
 568     bottom: "res2c"
 569     top: "res2c"
 570     name: "res2c_relu"
 571     type: "ReLU"
 572 }
 573 
 574 layer {
 575     bottom: "res2c"
 576     top: "res3a_branch1"
 577     name: "res3a_branch1"
 578     type: "Convolution"
 579     convolution_param {
 580         num_output: 512
 581         kernel_size: 1
 582         pad: 0
 583         stride: 2
 584         weight_filler {
 585             type: "msra"
 586         }
 587         bias_term: false
 588 
 589     }
 590 }
 591 
 592 layer {
 593     bottom: "res3a_branch1"
 594     top: "res3a_branch1"
 595     name: "bn3a_branch1"
 596     type: "BatchNorm"
 597     batch_norm_param {
 598         use_global_stats: false
 599     }
 600 }
 601 
 602 layer {
 603     bottom: "res3a_branch1"
 604     top: "res3a_branch1"
 605     name: "scale3a_branch1"
 606     type: "Scale"
 607     scale_param {
 608         bias_term: true
 609     }
 610 }
 611 
 612 layer {
 613     bottom: "res2c"
 614     top: "res3a_branch2a"
 615     name: "res3a_branch2a"
 616     type: "Convolution"
 617     convolution_param {
 618         num_output: 128
 619         kernel_size: 1
 620         pad: 0
 621         stride: 2
 622         weight_filler {
 623             type: "msra"
 624         }
 625         bias_term: false
 626 
 627     }
 628 }
 629 
 630 layer {
 631     bottom: "res3a_branch2a"
 632     top: "res3a_branch2a"
 633     name: "bn3a_branch2a"
 634     type: "BatchNorm"
 635     batch_norm_param {
 636         use_global_stats: false
 637     }
 638 }
 639 
 640 layer {
 641     bottom: "res3a_branch2a"
 642     top: "res3a_branch2a"
 643     name: "scale3a_branch2a"
 644     type: "Scale"
 645     scale_param {
 646         bias_term: true
 647     }
 648 }
 649 
 650 layer {
 651     bottom: "res3a_branch2a"
 652     top: "res3a_branch2a"
 653     name: "res3a_branch2a_relu"
 654     type: "ReLU"
 655 }
 656 
 657 layer {
 658     bottom: "res3a_branch2a"
 659     top: "res3a_branch2b"
 660     name: "res3a_branch2b"
 661     type: "Convolution"
 662     convolution_param {
 663         num_output: 128
 664         kernel_size: 3
 665         pad: 1
 666         stride: 1
 667         weight_filler {
 668             type: "msra"
 669         }
 670         bias_term: false
 671 
 672     }
 673 }
 674 
 675 layer {
 676     bottom: "res3a_branch2b"
 677     top: "res3a_branch2b"
 678     name: "bn3a_branch2b"
 679     type: "BatchNorm"
 680     batch_norm_param {
 681         use_global_stats: false
 682     }
 683 }
 684 
 685 layer {
 686     bottom: "res3a_branch2b"
 687     top: "res3a_branch2b"
 688     name: "scale3a_branch2b"
 689     type: "Scale"
 690     scale_param {
 691         bias_term: true
 692     }
 693 }
 694 
 695 layer {
 696     bottom: "res3a_branch2b"
 697     top: "res3a_branch2b"
 698     name: "res3a_branch2b_relu"
 699     type: "ReLU"
 700 }
 701 
 702 layer {
 703     bottom: "res3a_branch2b"
 704     top: "res3a_branch2c"
 705     name: "res3a_branch2c"
 706     type: "Convolution"
 707     convolution_param {
 708         num_output: 512
 709         kernel_size: 1
 710         pad: 0
 711         stride: 1
 712         weight_filler {
 713             type: "msra"
 714         }
 715         bias_term: false
 716 
 717     }
 718 }
 719 
 720 layer {
 721     bottom: "res3a_branch2c"
 722     top: "res3a_branch2c"
 723     name: "bn3a_branch2c"
 724     type: "BatchNorm"
 725     batch_norm_param {
 726         use_global_stats: false
 727     }
 728 }
 729 
 730 layer {
 731     bottom: "res3a_branch2c"
 732     top: "res3a_branch2c"
 733     name: "scale3a_branch2c"
 734     type: "Scale"
 735     scale_param {
 736         bias_term: true
 737     }
 738 }
 739 
 740 layer {
 741     bottom: "res3a_branch1"
 742     bottom: "res3a_branch2c"
 743     top: "res3a"
 744     name: "res3a"
 745     type: "Eltwise"
 746     eltwise_param {
 747         operation: SUM
 748     }
 749 }
 750 
 751 layer {
 752     bottom: "res3a"
 753     top: "res3a"
 754     name: "res3a_relu"
 755     type: "ReLU"
 756 }
 757 
 758 layer {
 759     bottom: "res3a"
 760     top: "res3b1_branch2a"
 761     name: "res3b1_branch2a"
 762     type: "Convolution"
 763     convolution_param {
 764         num_output: 128
 765         kernel_size: 1
 766         pad: 0
 767         stride: 1
 768         weight_filler {
 769             type: "msra"
 770         }
 771         bias_term: false
 772 
 773     }
 774 }
 775 
 776 layer {
 777     bottom: "res3b1_branch2a"
 778     top: "res3b1_branch2a"
 779     name: "bn3b1_branch2a"
 780     type: "BatchNorm"
 781     batch_norm_param {
 782         use_global_stats: false
 783     }
 784 }
 785 
 786 layer {
 787     bottom: "res3b1_branch2a"
 788     top: "res3b1_branch2a"
 789     name: "scale3b1_branch2a"
 790     type: "Scale"
 791     scale_param {
 792         bias_term: true
 793     }
 794 }
 795 
 796 layer {
 797     bottom: "res3b1_branch2a"
 798     top: "res3b1_branch2a"
 799     name: "res3b1_branch2a_relu"
 800     type: "ReLU"
 801 }
 802 
 803 layer {
 804     bottom: "res3b1_branch2a"
 805     top: "res3b1_branch2b"
 806     name: "res3b1_branch2b"
 807     type: "Convolution"
 808     convolution_param {
 809         num_output: 128
 810         kernel_size: 3
 811         pad: 1
 812         stride: 1
 813         weight_filler {
 814             type: "msra"
 815         }
 816         bias_term: false
 817 
 818     }
 819 }
 820 
 821 layer {
 822     bottom: "res3b1_branch2b"
 823     top: "res3b1_branch2b"
 824     name: "bn3b1_branch2b"
 825     type: "BatchNorm"
 826     batch_norm_param {
 827         use_global_stats: false
 828     }
 829 }
 830 
 831 layer {
 832     bottom: "res3b1_branch2b"
 833     top: "res3b1_branch2b"
 834     name: "scale3b1_branch2b"
 835     type: "Scale"
 836     scale_param {
 837         bias_term: true
 838     }
 839 }
 840 
 841 layer {
 842     bottom: "res3b1_branch2b"
 843     top: "res3b1_branch2b"
 844     name: "res3b1_branch2b_relu"
 845     type: "ReLU"
 846 }
 847 
 848 layer {
 849     bottom: "res3b1_branch2b"
 850     top: "res3b1_branch2c"
 851     name: "res3b1_branch2c"
 852     type: "Convolution"
 853     convolution_param {
 854         num_output: 512
 855         kernel_size: 1
 856         pad: 0
 857         stride: 1
 858         weight_filler {
 859             type: "msra"
 860         }
 861         bias_term: false
 862 
 863     }
 864 }
 865 
 866 layer {
 867     bottom: "res3b1_branch2c"
 868     top: "res3b1_branch2c"
 869     name: "bn3b1_branch2c"
 870     type: "BatchNorm"
 871     batch_norm_param {
 872         use_global_stats: false
 873     }
 874 }
 875 
 876 layer {
 877     bottom: "res3b1_branch2c"
 878     top: "res3b1_branch2c"
 879     name: "scale3b1_branch2c"
 880     type: "Scale"
 881     scale_param {
 882         bias_term: true
 883     }
 884 }
 885 
 886 layer {
 887     bottom: "res3a"
 888     bottom: "res3b1_branch2c"
 889     top: "res3b1"
 890     name: "res3b1"
 891     type: "Eltwise"
 892     eltwise_param {
 893         operation: SUM
 894     }
 895 }
 896 
 897 layer {
 898     bottom: "res3b1"
 899     top: "res3b1"
 900     name: "res3b1_relu"
 901     type: "ReLU"
 902 }
 903 
 904 layer {
 905     bottom: "res3b1"
 906     top: "res3b2_branch2a"
 907     name: "res3b2_branch2a"
 908     type: "Convolution"
 909     convolution_param {
 910         num_output: 128
 911         kernel_size: 1
 912         pad: 0
 913         stride: 1
 914         weight_filler {
 915             type: "msra"
 916         }
 917         bias_term: false
 918 
 919     }
 920 }
 921 
 922 layer {
 923     bottom: "res3b2_branch2a"
 924     top: "res3b2_branch2a"
 925     name: "bn3b2_branch2a"
 926     type: "BatchNorm"
 927     batch_norm_param {
 928         use_global_stats: false
 929     }
 930 }
 931 
 932 layer {
 933     bottom: "res3b2_branch2a"
 934     top: "res3b2_branch2a"
 935     name: "scale3b2_branch2a"
 936     type: "Scale"
 937     scale_param {
 938         bias_term: true
 939     }
 940 }
 941 
 942 layer {
 943     bottom: "res3b2_branch2a"
 944     top: "res3b2_branch2a"
 945     name: "res3b2_branch2a_relu"
 946     type: "ReLU"
 947 }
 948 
 949 layer {
 950     bottom: "res3b2_branch2a"
 951     top: "res3b2_branch2b"
 952     name: "res3b2_branch2b"
 953     type: "Convolution"
 954     convolution_param {
 955         num_output: 128
 956         kernel_size: 3
 957         pad: 1
 958         stride: 1
 959         weight_filler {
 960             type: "msra"
 961         }
 962         bias_term: false
 963 
 964     }
 965 }
 966 
 967 layer {
 968     bottom: "res3b2_branch2b"
 969     top: "res3b2_branch2b"
 970     name: "bn3b2_branch2b"
 971     type: "BatchNorm"
 972     batch_norm_param {
 973         use_global_stats: false
 974     }
 975 }
 976 
 977 layer {
 978     bottom: "res3b2_branch2b"
 979     top: "res3b2_branch2b"
 980     name: "scale3b2_branch2b"
 981     type: "Scale"
 982     scale_param {
 983         bias_term: true
 984     }
 985 }
 986 
 987 layer {
 988     bottom: "res3b2_branch2b"
 989     top: "res3b2_branch2b"
 990     name: "res3b2_branch2b_relu"
 991     type: "ReLU"
 992 }
 993 
 994 layer {
 995     bottom: "res3b2_branch2b"
 996     top: "res3b2_branch2c"
 997     name: "res3b2_branch2c"
 998     type: "Convolution"
 999     convolution_param {
1000         num_output: 512
1001         kernel_size: 1
1002         pad: 0
1003         stride: 1
1004         weight_filler {
1005             type: "msra"
1006         }
1007         bias_term: false
1008 
1009     }
1010 }
1011 
1012 layer {
1013     bottom: "res3b2_branch2c"
1014     top: "res3b2_branch2c"
1015     name: "bn3b2_branch2c"
1016     type: "BatchNorm"
1017     batch_norm_param {
1018         use_global_stats: false
1019     }
1020 }
1021 
1022 layer {
1023     bottom: "res3b2_branch2c"
1024     top: "res3b2_branch2c"
1025     name: "scale3b2_branch2c"
1026     type: "Scale"
1027     scale_param {
1028         bias_term: true
1029     }
1030 }
1031 
1032 layer {
1033     bottom: "res3b1"
1034     bottom: "res3b2_branch2c"
1035     top: "res3b2"
1036     name: "res3b2"
1037     type: "Eltwise"
1038     eltwise_param {
1039         operation: SUM
1040     }
1041 }
1042 
1043 layer {
1044     bottom: "res3b2"
1045     top: "res3b2"
1046     name: "res3b2_relu"
1047     type: "ReLU"
1048 }
1049 
1050 layer {
1051     bottom: "res3b2"
1052     top: "res3b3_branch2a"
1053     name: "res3b3_branch2a"
1054     type: "Convolution"
1055     convolution_param {
1056         num_output: 128
1057         kernel_size: 1
1058         pad: 0
1059         stride: 1
1060         weight_filler {
1061             type: "msra"
1062         }
1063         bias_term: false
1064 
1065     }
1066 }
1067 
1068 layer {
1069     bottom: "res3b3_branch2a"
1070     top: "res3b3_branch2a"
1071     name: "bn3b3_branch2a"
1072     type: "BatchNorm"
1073     batch_norm_param {
1074         use_global_stats: false
1075     }
1076 }
1077 
1078 layer {
1079     bottom: "res3b3_branch2a"
1080     top: "res3b3_branch2a"
1081     name: "scale3b3_branch2a"
1082     type: "Scale"
1083     scale_param {
1084         bias_term: true
1085     }
1086 }
1087 
1088 layer {
1089     bottom: "res3b3_branch2a"
1090     top: "res3b3_branch2a"
1091     name: "res3b3_branch2a_relu"
1092     type: "ReLU"
1093 }
1094 
1095 layer {
1096     bottom: "res3b3_branch2a"
1097     top: "res3b3_branch2b"
1098     name: "res3b3_branch2b"
1099     type: "Convolution"
1100     convolution_param {
1101         num_output: 128
1102         kernel_size: 3
1103         pad: 1
1104         stride: 1
1105         weight_filler {
1106             type: "msra"
1107         }
1108         bias_term: false
1109 
1110     }
1111 }
1112 
1113 layer {
1114     bottom: "res3b3_branch2b"
1115     top: "res3b3_branch2b"
1116     name: "bn3b3_branch2b"
1117     type: "BatchNorm"
1118     batch_norm_param {
1119         use_global_stats: false
1120     }
1121 }
1122 
1123 layer {
1124     bottom: "res3b3_branch2b"
1125     top: "res3b3_branch2b"
1126     name: "scale3b3_branch2b"
1127     type: "Scale"
1128     scale_param {
1129         bias_term: true
1130     }
1131 }
1132 
1133 layer {
1134     bottom: "res3b3_branch2b"
1135     top: "res3b3_branch2b"
1136     name: "res3b3_branch2b_relu"
1137     type: "ReLU"
1138 }
1139 
1140 layer {
1141     bottom: "res3b3_branch2b"
1142     top: "res3b3_branch2c"
1143     name: "res3b3_branch2c"
1144     type: "Convolution"
1145     convolution_param {
1146         num_output: 512
1147         kernel_size: 1
1148         pad: 0
1149         stride: 1
1150         weight_filler {
1151             type: "msra"
1152         }
1153         bias_term: false
1154 
1155     }
1156 }
1157 
1158 layer {
1159     bottom: "res3b3_branch2c"
1160     top: "res3b3_branch2c"
1161     name: "bn3b3_branch2c"
1162     type: "BatchNorm"
1163     batch_norm_param {
1164         use_global_stats: false
1165     }
1166 }
1167 
1168 layer {
1169     bottom: "res3b3_branch2c"
1170     top: "res3b3_branch2c"
1171     name: "scale3b3_branch2c"
1172     type: "Scale"
1173     scale_param {
1174         bias_term: true
1175     }
1176 }
1177 
1178 layer {
1179     bottom: "res3b2"
1180     bottom: "res3b3_branch2c"
1181     top: "res3b3"
1182     name: "res3b3"
1183     type: "Eltwise"
1184     eltwise_param {
1185         operation: SUM
1186     }
1187 }
1188 
1189 layer {
1190     bottom: "res3b3"
1191     top: "res3b3"
1192     name: "res3b3_relu"
1193     type: "ReLU"
1194 }
1195 
1196 layer {
1197     bottom: "res3b3"
1198     top: "res3b4_branch2a"
1199     name: "res3b4_branch2a"
1200     type: "Convolution"
1201     convolution_param {
1202         num_output: 128
1203         kernel_size: 1
1204         pad: 0
1205         stride: 1
1206         weight_filler {
1207             type: "msra"
1208         }
1209         bias_term: false
1210 
1211     }
1212 }
1213 
1214 layer {
1215     bottom: "res3b4_branch2a"
1216     top: "res3b4_branch2a"
1217     name: "bn3b4_branch2a"
1218     type: "BatchNorm"
1219     batch_norm_param {
1220         use_global_stats: false
1221     }
1222 }
1223 
1224 layer {
1225     bottom: "res3b4_branch2a"
1226     top: "res3b4_branch2a"
1227     name: "scale3b4_branch2a"
1228     type: "Scale"
1229     scale_param {
1230         bias_term: true
1231     }
1232 }
1233 
1234 layer {
1235     bottom: "res3b4_branch2a"
1236     top: "res3b4_branch2a"
1237     name: "res3b4_branch2a_relu"
1238     type: "ReLU"
1239 }
1240 
1241 layer {
1242     bottom: "res3b4_branch2a"
1243     top: "res3b4_branch2b"
1244     name: "res3b4_branch2b"
1245     type: "Convolution"
1246     convolution_param {
1247         num_output: 128
1248         kernel_size: 3
1249         pad: 1
1250         stride: 1
1251         weight_filler {
1252             type: "msra"
1253         }
1254         bias_term: false
1255 
1256     }
1257 }
1258 
1259 layer {
1260     bottom: "res3b4_branch2b"
1261     top: "res3b4_branch2b"
1262     name: "bn3b4_branch2b"
1263     type: "BatchNorm"
1264     batch_norm_param {
1265         use_global_stats: false
1266     }
1267 }
1268 
1269 layer {
1270     bottom: "res3b4_branch2b"
1271     top: "res3b4_branch2b"
1272     name: "scale3b4_branch2b"
1273     type: "Scale"
1274     scale_param {
1275         bias_term: true
1276     }
1277 }
1278 
1279 layer {
1280     bottom: "res3b4_branch2b"
1281     top: "res3b4_branch2b"
1282     name: "res3b4_branch2b_relu"
1283     type: "ReLU"
1284 }
1285 
1286 layer {
1287     bottom: "res3b4_branch2b"
1288     top: "res3b4_branch2c"
1289     name: "res3b4_branch2c"
1290     type: "Convolution"
1291     convolution_param {
1292         num_output: 512
1293         kernel_size: 1
1294         pad: 0
1295         stride: 1
1296         weight_filler {
1297             type: "msra"
1298         }
1299         bias_term: false
1300 
1301     }
1302 }
1303 
1304 layer {
1305     bottom: "res3b4_branch2c"
1306     top: "res3b4_branch2c"
1307     name: "bn3b4_branch2c"
1308     type: "BatchNorm"
1309     batch_norm_param {
1310         use_global_stats: false
1311     }
1312 }
1313 
1314 layer {
1315     bottom: "res3b4_branch2c"
1316     top: "res3b4_branch2c"
1317     name: "scale3b4_branch2c"
1318     type: "Scale"
1319     scale_param {
1320         bias_term: true
1321     }
1322 }
1323 
1324 layer {
1325     bottom: "res3b3"
1326     bottom: "res3b4_branch2c"
1327     top: "res3b4"
1328     name: "res3b4"
1329     type: "Eltwise"
1330     eltwise_param {
1331         operation: SUM
1332     }
1333 }
1334 
1335 layer {
1336     bottom: "res3b4"
1337     top: "res3b4"
1338     name: "res3b4_relu"
1339     type: "ReLU"
1340 }
1341 
1342 layer {
1343     bottom: "res3b4"
1344     top: "res3b5_branch2a"
1345     name: "res3b5_branch2a"
1346     type: "Convolution"
1347     convolution_param {
1348         num_output: 128
1349         kernel_size: 1
1350         pad: 0
1351         stride: 1
1352         weight_filler {
1353             type: "msra"
1354         }
1355         bias_term: false
1356 
1357     }
1358 }
1359 
1360 layer {
1361     bottom: "res3b5_branch2a"
1362     top: "res3b5_branch2a"
1363     name: "bn3b5_branch2a"
1364     type: "BatchNorm"
1365     batch_norm_param {
1366         use_global_stats: false
1367     }
1368 }
1369 
1370 layer {
1371     bottom: "res3b5_branch2a"
1372     top: "res3b5_branch2a"
1373     name: "scale3b5_branch2a"
1374     type: "Scale"
1375     scale_param {
1376         bias_term: true
1377     }
1378 }
1379 
1380 layer {
1381     bottom: "res3b5_branch2a"
1382     top: "res3b5_branch2a"
1383     name: "res3b5_branch2a_relu"
1384     type: "ReLU"
1385 }
1386 
1387 layer {
1388     bottom: "res3b5_branch2a"
1389     top: "res3b5_branch2b"
1390     name: "res3b5_branch2b"
1391     type: "Convolution"
1392     convolution_param {
1393         num_output: 128
1394         kernel_size: 3
1395         pad: 1
1396         stride: 1
1397         weight_filler {
1398             type: "msra"
1399         }
1400         bias_term: false
1401 
1402     }
1403 }
1404 
1405 layer {
1406     bottom: "res3b5_branch2b"
1407     top: "res3b5_branch2b"
1408     name: "bn3b5_branch2b"
1409     type: "BatchNorm"
1410     batch_norm_param {
1411         use_global_stats: false
1412     }
1413 }
1414 
1415 layer {
1416     bottom: "res3b5_branch2b"
1417     top: "res3b5_branch2b"
1418     name: "scale3b5_branch2b"
1419     type: "Scale"
1420     scale_param {
1421         bias_term: true
1422     }
1423 }
1424 
1425 layer {
1426     bottom: "res3b5_branch2b"
1427     top: "res3b5_branch2b"
1428     name: "res3b5_branch2b_relu"
1429     type: "ReLU"
1430 }
1431 
1432 layer {
1433     bottom: "res3b5_branch2b"
1434     top: "res3b5_branch2c"
1435     name: "res3b5_branch2c"
1436     type: "Convolution"
1437     convolution_param {
1438         num_output: 512
1439         kernel_size: 1
1440         pad: 0
1441         stride: 1
1442         weight_filler {
1443             type: "msra"
1444         }
1445         bias_term: false
1446 
1447     }
1448 }
1449 
1450 layer {
1451     bottom: "res3b5_branch2c"
1452     top: "res3b5_branch2c"
1453     name: "bn3b5_branch2c"
1454     type: "BatchNorm"
1455     batch_norm_param {
1456         use_global_stats: false
1457     }
1458 }
1459 
1460 layer {
1461     bottom: "res3b5_branch2c"
1462     top: "res3b5_branch2c"
1463     name: "scale3b5_branch2c"
1464     type: "Scale"
1465     scale_param {
1466         bias_term: true
1467     }
1468 }
1469 
1470 layer {
1471     bottom: "res3b4"
1472     bottom: "res3b5_branch2c"
1473     top: "res3b5"
1474     name: "res3b5"
1475     type: "Eltwise"
1476     eltwise_param {
1477         operation: SUM
1478     }
1479 }
1480 
1481 layer {
1482     bottom: "res3b5"
1483     top: "res3b5"
1484     name: "res3b5_relu"
1485     type: "ReLU"
1486 }
1487 
1488 layer {
1489     bottom: "res3b5"
1490     top: "res3b6_branch2a"
1491     name: "res3b6_branch2a"
1492     type: "Convolution"
1493     convolution_param {
1494         num_output: 128
1495         kernel_size: 1
1496         pad: 0
1497         stride: 1
1498         weight_filler {
1499             type: "msra"
1500         }
1501         bias_term: false
1502 
1503     }
1504 }
1505 
1506 layer {
1507     bottom: "res3b6_branch2a"
1508     top: "res3b6_branch2a"
1509     name: "bn3b6_branch2a"
1510     type: "BatchNorm"
1511     batch_norm_param {
1512         use_global_stats: false
1513     }
1514 }
1515 
1516 layer {
1517     bottom: "res3b6_branch2a"
1518     top: "res3b6_branch2a"
1519     name: "scale3b6_branch2a"
1520     type: "Scale"
1521     scale_param {
1522         bias_term: true
1523     }
1524 }
1525 
1526 layer {
1527     bottom: "res3b6_branch2a"
1528     top: "res3b6_branch2a"
1529     name: "res3b6_branch2a_relu"
1530     type: "ReLU"
1531 }
1532 
1533 layer {
1534     bottom: "res3b6_branch2a"
1535     top: "res3b6_branch2b"
1536     name: "res3b6_branch2b"
1537     type: "Convolution"
1538     convolution_param {
1539         num_output: 128
1540         kernel_size: 3
1541         pad: 1
1542         stride: 1
1543         weight_filler {
1544             type: "msra"
1545         }
1546         bias_term: false
1547 
1548     }
1549 }
1550 
1551 layer {
1552     bottom: "res3b6_branch2b"
1553     top: "res3b6_branch2b"
1554     name: "bn3b6_branch2b"
1555     type: "BatchNorm"
1556     batch_norm_param {
1557         use_global_stats: false
1558     }
1559 }
1560 
1561 layer {
1562     bottom: "res3b6_branch2b"
1563     top: "res3b6_branch2b"
1564     name: "scale3b6_branch2b"
1565     type: "Scale"
1566     scale_param {
1567         bias_term: true
1568     }
1569 }
1570 
1571 layer {
1572     bottom: "res3b6_branch2b"
1573     top: "res3b6_branch2b"
1574     name: "res3b6_branch2b_relu"
1575     type: "ReLU"
1576 }
1577 
1578 layer {
1579     bottom: "res3b6_branch2b"
1580     top: "res3b6_branch2c"
1581     name: "res3b6_branch2c"
1582     type: "Convolution"
1583     convolution_param {
1584         num_output: 512
1585         kernel_size: 1
1586         pad: 0
1587         stride: 1
1588         weight_filler {
1589             type: "msra"
1590         }
1591         bias_term: false
1592 
1593     }
1594 }
1595 
1596 layer {
1597     bottom: "res3b6_branch2c"
1598     top: "res3b6_branch2c"
1599     name: "bn3b6_branch2c"
1600     type: "BatchNorm"
1601     batch_norm_param {
1602         use_global_stats: false
1603     }
1604 }
1605 
1606 layer {
1607     bottom: "res3b6_branch2c"
1608     top: "res3b6_branch2c"
1609     name: "scale3b6_branch2c"
1610     type: "Scale"
1611     scale_param {
1612         bias_term: true
1613     }
1614 }
1615 
1616 layer {
1617     bottom: "res3b5"
1618     bottom: "res3b6_branch2c"
1619     top: "res3b6"
1620     name: "res3b6"
1621     type: "Eltwise"
1622     eltwise_param {
1623         operation: SUM
1624     }
1625 }
1626 
1627 layer {
1628     bottom: "res3b6"
1629     top: "res3b6"
1630     name: "res3b6_relu"
1631     type: "ReLU"
1632 }
1633 
1634 layer {
1635     bottom: "res3b6"
1636     top: "res3b7_branch2a"
1637     name: "res3b7_branch2a"
1638     type: "Convolution"
1639     convolution_param {
1640         num_output: 128
1641         kernel_size: 1
1642         pad: 0
1643         stride: 1
1644         weight_filler {
1645             type: "msra"
1646         }
1647         bias_term: false
1648 
1649     }
1650 }
1651 
1652 layer {
1653     bottom: "res3b7_branch2a"
1654     top: "res3b7_branch2a"
1655     name: "bn3b7_branch2a"
1656     type: "BatchNorm"
1657     batch_norm_param {
1658         use_global_stats: false
1659     }
1660 }
1661 
1662 layer {
1663     bottom: "res3b7_branch2a"
1664     top: "res3b7_branch2a"
1665     name: "scale3b7_branch2a"
1666     type: "Scale"
1667     scale_param {
1668         bias_term: true
1669     }
1670 }
1671 
1672 layer {
1673     bottom: "res3b7_branch2a"
1674     top: "res3b7_branch2a"
1675     name: "res3b7_branch2a_relu"
1676     type: "ReLU"
1677 }
1678 
1679 layer {
1680     bottom: "res3b7_branch2a"
1681     top: "res3b7_branch2b"
1682     name: "res3b7_branch2b"
1683     type: "Convolution"
1684     convolution_param {
1685         num_output: 128
1686         kernel_size: 3
1687         pad: 1
1688         stride: 1
1689         weight_filler {
1690             type: "msra"
1691         }
1692         bias_term: false
1693 
1694     }
1695 }
1696 
1697 layer {
1698     bottom: "res3b7_branch2b"
1699     top: "res3b7_branch2b"
1700     name: "bn3b7_branch2b"
1701     type: "BatchNorm"
1702     batch_norm_param {
1703         use_global_stats: false
1704     }
1705 }
1706 
1707 layer {
1708     bottom: "res3b7_branch2b"
1709     top: "res3b7_branch2b"
1710     name: "scale3b7_branch2b"
1711     type: "Scale"
1712     scale_param {
1713         bias_term: true
1714     }
1715 }
1716 
1717 layer {
1718     bottom: "res3b7_branch2b"
1719     top: "res3b7_branch2b"
1720     name: "res3b7_branch2b_relu"
1721     type: "ReLU"
1722 }
1723 
1724 layer {
1725     bottom: "res3b7_branch2b"
1726     top: "res3b7_branch2c"
1727     name: "res3b7_branch2c"
1728     type: "Convolution"
1729     convolution_param {
1730         num_output: 512
1731         kernel_size: 1
1732         pad: 0
1733         stride: 1
1734         weight_filler {
1735             type: "msra"
1736         }
1737         bias_term: false
1738 
1739     }
1740 }
1741 
1742 layer {
1743     bottom: "res3b7_branch2c"
1744     top: "res3b7_branch2c"
1745     name: "bn3b7_branch2c"
1746     type: "BatchNorm"
1747     batch_norm_param {
1748         use_global_stats: false
1749     }
1750 }
1751 
1752 layer {
1753     bottom: "res3b7_branch2c"
1754     top: "res3b7_branch2c"
1755     name: "scale3b7_branch2c"
1756     type: "Scale"
1757     scale_param {
1758         bias_term: true
1759     }
1760 }
1761 
1762 layer {
1763     bottom: "res3b6"
1764     bottom: "res3b7_branch2c"
1765     top: "res3b7"
1766     name: "res3b7"
1767     type: "Eltwise"
1768     eltwise_param {
1769         operation: SUM
1770     }
1771 }
1772 
1773 layer {
1774     bottom: "res3b7"
1775     top: "res3b7"
1776     name: "res3b7_relu"
1777     type: "ReLU"
1778 }
1779 
1780 layer {
1781     bottom: "res3b7"
1782     top: "res4a_branch1"
1783     name: "res4a_branch1"
1784     type: "Convolution"
1785     convolution_param {
1786         num_output: 1024
1787         kernel_size: 1
1788         pad: 0
1789         stride: 2
1790         weight_filler {
1791             type: "msra"
1792         }
1793         bias_term: false
1794 
1795     }
1796 }
1797 
1798 layer {
1799     bottom: "res4a_branch1"
1800     top: "res4a_branch1"
1801     name: "bn4a_branch1"
1802     type: "BatchNorm"
1803     batch_norm_param {
1804         use_global_stats: false
1805     }
1806 }
1807 
1808 layer {
1809     bottom: "res4a_branch1"
1810     top: "res4a_branch1"
1811     name: "scale4a_branch1"
1812     type: "Scale"
1813     scale_param {
1814         bias_term: true
1815     }
1816 }
1817 
1818 layer {
1819     bottom: "res3b7"
1820     top: "res4a_branch2a"
1821     name: "res4a_branch2a"
1822     type: "Convolution"
1823     convolution_param {
1824         num_output: 256
1825         kernel_size: 1
1826         pad: 0
1827         stride: 2
1828         weight_filler {
1829             type: "msra"
1830         }
1831         bias_term: false
1832 
1833     }
1834 }
1835 
1836 layer {
1837     bottom: "res4a_branch2a"
1838     top: "res4a_branch2a"
1839     name: "bn4a_branch2a"
1840     type: "BatchNorm"
1841     batch_norm_param {
1842         use_global_stats: false
1843     }
1844 }
1845 
1846 layer {
1847     bottom: "res4a_branch2a"
1848     top: "res4a_branch2a"
1849     name: "scale4a_branch2a"
1850     type: "Scale"
1851     scale_param {
1852         bias_term: true
1853     }
1854 }
1855 
1856 layer {
1857     bottom: "res4a_branch2a"
1858     top: "res4a_branch2a"
1859     name: "res4a_branch2a_relu"
1860     type: "ReLU"
1861 }
1862 
1863 layer {
1864     bottom: "res4a_branch2a"
1865     top: "res4a_branch2b"
1866     name: "res4a_branch2b"
1867     type: "Convolution"
1868     convolution_param {
1869         num_output: 256
1870         kernel_size: 3
1871         pad: 1
1872         stride: 1
1873         weight_filler {
1874             type: "msra"
1875         }
1876         bias_term: false
1877 
1878     }
1879 }
1880 
1881 layer {
1882     bottom: "res4a_branch2b"
1883     top: "res4a_branch2b"
1884     name: "bn4a_branch2b"
1885     type: "BatchNorm"
1886     batch_norm_param {
1887         use_global_stats: false
1888     }
1889 }
1890 
1891 layer {
1892     bottom: "res4a_branch2b"
1893     top: "res4a_branch2b"
1894     name: "scale4a_branch2b"
1895     type: "Scale"
1896     scale_param {
1897         bias_term: true
1898     }
1899 }
1900 
1901 layer {
1902     bottom: "res4a_branch2b"
1903     top: "res4a_branch2b"
1904     name: "res4a_branch2b_relu"
1905     type: "ReLU"
1906 }
1907 
1908 layer {
1909     bottom: "res4a_branch2b"
1910     top: "res4a_branch2c"
1911     name: "res4a_branch2c"
1912     type: "Convolution"
1913     convolution_param {
1914         num_output: 1024
1915         kernel_size: 1
1916         pad: 0
1917         stride: 1
1918         weight_filler {
1919             type: "msra"
1920         }
1921         bias_term: false
1922 
1923     }
1924 }
1925 
1926 layer {
1927     bottom: "res4a_branch2c"
1928     top: "res4a_branch2c"
1929     name: "bn4a_branch2c"
1930     type: "BatchNorm"
1931     batch_norm_param {
1932         use_global_stats: false
1933     }
1934 }
1935 
1936 layer {
1937     bottom: "res4a_branch2c"
1938     top: "res4a_branch2c"
1939     name: "scale4a_branch2c"
1940     type: "Scale"
1941     scale_param {
1942         bias_term: true
1943     }
1944 }
1945 
1946 layer {
1947     bottom: "res4a_branch1"
1948     bottom: "res4a_branch2c"
1949     top: "res4a"
1950     name: "res4a"
1951     type: "Eltwise"
1952     eltwise_param {
1953         operation: SUM
1954     }
1955 }
1956 
1957 layer {
1958     bottom: "res4a"
1959     top: "res4a"
1960     name: "res4a_relu"
1961     type: "ReLU"
1962 }
1963 
1964 layer {
1965     bottom: "res4a"
1966     top: "res4b1_branch2a"
1967     name: "res4b1_branch2a"
1968     type: "Convolution"
1969     convolution_param {
1970         num_output: 256
1971         kernel_size: 1
1972         pad: 0
1973         stride: 1
1974         weight_filler {
1975             type: "msra"
1976         }
1977         bias_term: false
1978 
1979     }
1980 }
1981 
1982 layer {
1983     bottom: "res4b1_branch2a"
1984     top: "res4b1_branch2a"
1985     name: "bn4b1_branch2a"
1986     type: "BatchNorm"
1987     batch_norm_param {
1988         use_global_stats: false
1989     }
1990 }
1991 
1992 layer {
1993     bottom: "res4b1_branch2a"
1994     top: "res4b1_branch2a"
1995     name: "scale4b1_branch2a"
1996     type: "Scale"
1997     scale_param {
1998         bias_term: true
1999     }
2000 }
2001 
2002 layer {
2003     bottom: "res4b1_branch2a"
2004     top: "res4b1_branch2a"
2005     name: "res4b1_branch2a_relu"
2006     type: "ReLU"
2007 }
2008 
2009 layer {
2010     bottom: "res4b1_branch2a"
2011     top: "res4b1_branch2b"
2012     name: "res4b1_branch2b"
2013     type: "Convolution"
2014     convolution_param {
2015         num_output: 256
2016         kernel_size: 3
2017         pad: 1
2018         stride: 1
2019         weight_filler {
2020             type: "msra"
2021         }
2022         bias_term: false
2023 
2024     }
2025 }
2026 
2027 layer {
2028     bottom: "res4b1_branch2b"
2029     top: "res4b1_branch2b"
2030     name: "bn4b1_branch2b"
2031     type: "BatchNorm"
2032     batch_norm_param {
2033         use_global_stats: false
2034     }
2035 }
2036 
2037 layer {
2038     bottom: "res4b1_branch2b"
2039     top: "res4b1_branch2b"
2040     name: "scale4b1_branch2b"
2041     type: "Scale"
2042     scale_param {
2043         bias_term: true
2044     }
2045 }
2046 
2047 layer {
2048     bottom: "res4b1_branch2b"
2049     top: "res4b1_branch2b"
2050     name: "res4b1_branch2b_relu"
2051     type: "ReLU"
2052 }
2053 
2054 layer {
2055     bottom: "res4b1_branch2b"
2056     top: "res4b1_branch2c"
2057     name: "res4b1_branch2c"
2058     type: "Convolution"
2059     convolution_param {
2060         num_output: 1024
2061         kernel_size: 1
2062         pad: 0
2063         stride: 1
2064         weight_filler {
2065             type: "msra"
2066         }
2067         bias_term: false
2068 
2069     }
2070 }
2071 
2072 layer {
2073     bottom: "res4b1_branch2c"
2074     top: "res4b1_branch2c"
2075     name: "bn4b1_branch2c"
2076     type: "BatchNorm"
2077     batch_norm_param {
2078         use_global_stats: false
2079     }
2080 }
2081 
2082 layer {
2083     bottom: "res4b1_branch2c"
2084     top: "res4b1_branch2c"
2085     name: "scale4b1_branch2c"
2086     type: "Scale"
2087     scale_param {
2088         bias_term: true
2089     }
2090 }
2091 
2092 layer {
2093     bottom: "res4a"
2094     bottom: "res4b1_branch2c"
2095     top: "res4b1"
2096     name: "res4b1"
2097     type: "Eltwise"
2098     eltwise_param {
2099         operation: SUM
2100     }
2101 }
2102 
2103 layer {
2104     bottom: "res4b1"
2105     top: "res4b1"
2106     name: "res4b1_relu"
2107     type: "ReLU"
2108 }
2109 
2110 layer {
2111     bottom: "res4b1"
2112     top: "res4b2_branch2a"
2113     name: "res4b2_branch2a"
2114     type: "Convolution"
2115     convolution_param {
2116         num_output: 256
2117         kernel_size: 1
2118         pad: 0
2119         stride: 1
2120         weight_filler {
2121             type: "msra"
2122         }
2123         bias_term: false
2124 
2125     }
2126 }
2127 
2128 layer {
2129     bottom: "res4b2_branch2a"
2130     top: "res4b2_branch2a"
2131     name: "bn4b2_branch2a"
2132     type: "BatchNorm"
2133     batch_norm_param {
2134         use_global_stats: false
2135     }
2136 }
2137 
2138 layer {
2139     bottom: "res4b2_branch2a"
2140     top: "res4b2_branch2a"
2141     name: "scale4b2_branch2a"
2142     type: "Scale"
2143     scale_param {
2144         bias_term: true
2145     }
2146 }
2147 
2148 layer {
2149     bottom: "res4b2_branch2a"
2150     top: "res4b2_branch2a"
2151     name: "res4b2_branch2a_relu"
2152     type: "ReLU"
2153 }
2154 
2155 layer {
2156     bottom: "res4b2_branch2a"
2157     top: "res4b2_branch2b"
2158     name: "res4b2_branch2b"
2159     type: "Convolution"
2160     convolution_param {
2161         num_output: 256
2162         kernel_size: 3
2163         pad: 1
2164         stride: 1
2165         weight_filler {
2166             type: "msra"
2167         }
2168         bias_term: false
2169 
2170     }
2171 }
2172 
2173 layer {
2174     bottom: "res4b2_branch2b"
2175     top: "res4b2_branch2b"
2176     name: "bn4b2_branch2b"
2177     type: "BatchNorm"
2178     batch_norm_param {
2179         use_global_stats: false
2180     }
2181 }
2182 
2183 layer {
2184     bottom: "res4b2_branch2b"
2185     top: "res4b2_branch2b"
2186     name: "scale4b2_branch2b"
2187     type: "Scale"
2188     scale_param {
2189         bias_term: true
2190     }
2191 }
2192 
2193 layer {
2194     bottom: "res4b2_branch2b"
2195     top: "res4b2_branch2b"
2196     name: "res4b2_branch2b_relu"
2197     type: "ReLU"
2198 }
2199 
2200 layer {
2201     bottom: "res4b2_branch2b"
2202     top: "res4b2_branch2c"
2203     name: "res4b2_branch2c"
2204     type: "Convolution"
2205     convolution_param {
2206         num_output: 1024
2207         kernel_size: 1
2208         pad: 0
2209         stride: 1
2210         weight_filler {
2211             type: "msra"
2212         }
2213         bias_term: false
2214 
2215     }
2216 }
2217 
2218 layer {
2219     bottom: "res4b2_branch2c"
2220     top: "res4b2_branch2c"
2221     name: "bn4b2_branch2c"
2222     type: "BatchNorm"
2223     batch_norm_param {
2224         use_global_stats: false
2225     }
2226 }
2227 
2228 layer {
2229     bottom: "res4b2_branch2c"
2230     top: "res4b2_branch2c"
2231     name: "scale4b2_branch2c"
2232     type: "Scale"
2233     scale_param {
2234         bias_term: true
2235     }
2236 }
2237 
2238 layer {
2239     bottom: "res4b1"
2240     bottom: "res4b2_branch2c"
2241     top: "res4b2"
2242     name: "res4b2"
2243     type: "Eltwise"
2244     eltwise_param {
2245         operation: SUM
2246     }
2247 }
2248 
2249 layer {
2250     bottom: "res4b2"
2251     top: "res4b2"
2252     name: "res4b2_relu"
2253     type: "ReLU"
2254 }
2255 
2256 layer {
2257     bottom: "res4b2"
2258     top: "res4b3_branch2a"
2259     name: "res4b3_branch2a"
2260     type: "Convolution"
2261     convolution_param {
2262         num_output: 256
2263         kernel_size: 1
2264         pad: 0
2265         stride: 1
2266         weight_filler {
2267             type: "msra"
2268         }
2269         bias_term: false
2270 
2271     }
2272 }
2273 
2274 layer {
2275     bottom: "res4b3_branch2a"
2276     top: "res4b3_branch2a"
2277     name: "bn4b3_branch2a"
2278     type: "BatchNorm"
2279     batch_norm_param {
2280         use_global_stats: false
2281     }
2282 }
2283 
2284 layer {
2285     bottom: "res4b3_branch2a"
2286     top: "res4b3_branch2a"
2287     name: "scale4b3_branch2a"
2288     type: "Scale"
2289     scale_param {
2290         bias_term: true
2291     }
2292 }
2293 
2294 layer {
2295     bottom: "res4b3_branch2a"
2296     top: "res4b3_branch2a"
2297     name: "res4b3_branch2a_relu"
2298     type: "ReLU"
2299 }
2300 
2301 layer {
2302     bottom: "res4b3_branch2a"
2303     top: "res4b3_branch2b"
2304     name: "res4b3_branch2b"
2305     type: "Convolution"
2306     convolution_param {
2307         num_output: 256
2308         kernel_size: 3
2309         pad: 1
2310         stride: 1
2311         weight_filler {
2312             type: "msra"
2313         }
2314         bias_term: false
2315 
2316     }
2317 }
2318 
2319 layer {
2320     bottom: "res4b3_branch2b"
2321     top: "res4b3_branch2b"
2322     name: "bn4b3_branch2b"
2323     type: "BatchNorm"
2324     batch_norm_param {
2325         use_global_stats: false
2326     }
2327 }
2328 
2329 layer {
2330     bottom: "res4b3_branch2b"
2331     top: "res4b3_branch2b"
2332     name: "scale4b3_branch2b"
2333     type: "Scale"
2334     scale_param {
2335         bias_term: true
2336     }
2337 }
2338 
2339 layer {
2340     bottom: "res4b3_branch2b"
2341     top: "res4b3_branch2b"
2342     name: "res4b3_branch2b_relu"
2343     type: "ReLU"
2344 }
2345 
2346 layer {
2347     bottom: "res4b3_branch2b"
2348     top: "res4b3_branch2c"
2349     name: "res4b3_branch2c"
2350     type: "Convolution"
2351     convolution_param {
2352         num_output: 1024
2353         kernel_size: 1
2354         pad: 0
2355         stride: 1
2356         weight_filler {
2357             type: "msra"
2358         }
2359         bias_term: false
2360 
2361     }
2362 }
2363 
2364 layer {
2365     bottom: "res4b3_branch2c"
2366     top: "res4b3_branch2c"
2367     name: "bn4b3_branch2c"
2368     type: "BatchNorm"
2369     batch_norm_param {
2370         use_global_stats: false
2371     }
2372 }
2373 
2374 layer {
2375     bottom: "res4b3_branch2c"
2376     top: "res4b3_branch2c"
2377     name: "scale4b3_branch2c"
2378     type: "Scale"
2379     scale_param {
2380         bias_term: true
2381     }
2382 }
2383 
2384 layer {
2385     bottom: "res4b2"
2386     bottom: "res4b3_branch2c"
2387     top: "res4b3"
2388     name: "res4b3"
2389     type: "Eltwise"
2390     eltwise_param {
2391         operation: SUM
2392     }
2393 }
2394 
2395 layer {
2396     bottom: "res4b3"
2397     top: "res4b3"
2398     name: "res4b3_relu"
2399     type: "ReLU"
2400 }
2401 
2402 layer {
2403     bottom: "res4b3"
2404     top: "res4b4_branch2a"
2405     name: "res4b4_branch2a"
2406     type: "Convolution"
2407     convolution_param {
2408         num_output: 256
2409         kernel_size: 1
2410         pad: 0
2411         stride: 1
2412         weight_filler {
2413             type: "msra"
2414         }
2415         bias_term: false
2416 
2417     }
2418 }
2419 
2420 layer {
2421     bottom: "res4b4_branch2a"
2422     top: "res4b4_branch2a"
2423     name: "bn4b4_branch2a"
2424     type: "BatchNorm"
2425     batch_norm_param {
2426         use_global_stats: false
2427     }
2428 }
2429 
2430 layer {
2431     bottom: "res4b4_branch2a"
2432     top: "res4b4_branch2a"
2433     name: "scale4b4_branch2a"
2434     type: "Scale"
2435     scale_param {
2436         bias_term: true
2437     }
2438 }
2439 
2440 layer {
2441     bottom: "res4b4_branch2a"
2442     top: "res4b4_branch2a"
2443     name: "res4b4_branch2a_relu"
2444     type: "ReLU"
2445 }
2446 
2447 layer {
2448     bottom: "res4b4_branch2a"
2449     top: "res4b4_branch2b"
2450     name: "res4b4_branch2b"
2451     type: "Convolution"
2452     convolution_param {
2453         num_output: 256
2454         kernel_size: 3
2455         pad: 1
2456         stride: 1
2457         weight_filler {
2458             type: "msra"
2459         }
2460         bias_term: false
2461 
2462     }
2463 }
2464 
2465 layer {
2466     bottom: "res4b4_branch2b"
2467     top: "res4b4_branch2b"
2468     name: "bn4b4_branch2b"
2469     type: "BatchNorm"
2470     batch_norm_param {
2471         use_global_stats: false
2472     }
2473 }
2474 
2475 layer {
2476     bottom: "res4b4_branch2b"
2477     top: "res4b4_branch2b"
2478     name: "scale4b4_branch2b"
2479     type: "Scale"
2480     scale_param {
2481         bias_term: true
2482     }
2483 }
2484 
2485 layer {
2486     bottom: "res4b4_branch2b"
2487     top: "res4b4_branch2b"
2488     name: "res4b4_branch2b_relu"
2489     type: "ReLU"
2490 }
2491 
2492 layer {
2493     bottom: "res4b4_branch2b"
2494     top: "res4b4_branch2c"
2495     name: "res4b4_branch2c"
2496     type: "Convolution"
2497     convolution_param {
2498         num_output: 1024
2499         kernel_size: 1
2500         pad: 0
2501         stride: 1
2502         weight_filler {
2503             type: "msra"
2504         }
2505         bias_term: false
2506 
2507     }
2508 }
2509 
2510 layer {
2511     bottom: "res4b4_branch2c"
2512     top: "res4b4_branch2c"
2513     name: "bn4b4_branch2c"
2514     type: "BatchNorm"
2515     batch_norm_param {
2516         use_global_stats: false
2517     }
2518 }
2519 
2520 layer {
2521     bottom: "res4b4_branch2c"
2522     top: "res4b4_branch2c"
2523     name: "scale4b4_branch2c"
2524     type: "Scale"
2525     scale_param {
2526         bias_term: true
2527     }
2528 }
2529 
2530 layer {
2531     bottom: "res4b3"
2532     bottom: "res4b4_branch2c"
2533     top: "res4b4"
2534     name: "res4b4"
2535     type: "Eltwise"
2536     eltwise_param {
2537         operation: SUM
2538     }
2539 }
2540 
2541 layer {
2542     bottom: "res4b4"
2543     top: "res4b4"
2544     name: "res4b4_relu"
2545     type: "ReLU"
2546 }
2547 
2548 layer {
2549     bottom: "res4b4"
2550     top: "res4b5_branch2a"
2551     name: "res4b5_branch2a"
2552     type: "Convolution"
2553     convolution_param {
2554         num_output: 256
2555         kernel_size: 1
2556         pad: 0
2557         stride: 1
2558         weight_filler {
2559             type: "msra"
2560         }
2561         bias_term: false
2562 
2563     }
2564 }
2565 
2566 layer {
2567     bottom: "res4b5_branch2a"
2568     top: "res4b5_branch2a"
2569     name: "bn4b5_branch2a"
2570     type: "BatchNorm"
2571     batch_norm_param {
2572         use_global_stats: false
2573     }
2574 }
2575 
2576 layer {
2577     bottom: "res4b5_branch2a"
2578     top: "res4b5_branch2a"
2579     name: "scale4b5_branch2a"
2580     type: "Scale"
2581     scale_param {
2582         bias_term: true
2583     }
2584 }
2585 
2586 layer {
2587     bottom: "res4b5_branch2a"
2588     top: "res4b5_branch2a"
2589     name: "res4b5_branch2a_relu"
2590     type: "ReLU"
2591 }
2592 
2593 layer {
2594     bottom: "res4b5_branch2a"
2595     top: "res4b5_branch2b"
2596     name: "res4b5_branch2b"
2597     type: "Convolution"
2598     convolution_param {
2599         num_output: 256
2600         kernel_size: 3
2601         pad: 1
2602         stride: 1
2603         weight_filler {
2604             type: "msra"
2605         }
2606         bias_term: false
2607 
2608     }
2609 }
2610 
2611 layer {
2612     bottom: "res4b5_branch2b"
2613     top: "res4b5_branch2b"
2614     name: "bn4b5_branch2b"
2615     type: "BatchNorm"
2616     batch_norm_param {
2617         use_global_stats: false
2618     }
2619 }
2620 
2621 layer {
2622     bottom: "res4b5_branch2b"
2623     top: "res4b5_branch2b"
2624     name: "scale4b5_branch2b"
2625     type: "Scale"
2626     scale_param {
2627         bias_term: true
2628     }
2629 }
2630 
2631 layer {
2632     bottom: "res4b5_branch2b"
2633     top: "res4b5_branch2b"
2634     name: "res4b5_branch2b_relu"
2635     type: "ReLU"
2636 }
2637 
2638 layer {
2639     bottom: "res4b5_branch2b"
2640     top: "res4b5_branch2c"
2641     name: "res4b5_branch2c"
2642     type: "Convolution"
2643     convolution_param {
2644         num_output: 1024
2645         kernel_size: 1
2646         pad: 0
2647         stride: 1
2648         weight_filler {
2649             type: "msra"
2650         }
2651         bias_term: false
2652 
2653     }
2654 }
2655 
2656 layer {
2657     bottom: "res4b5_branch2c"
2658     top: "res4b5_branch2c"
2659     name: "bn4b5_branch2c"
2660     type: "BatchNorm"
2661     batch_norm_param {
2662         use_global_stats: false
2663     }
2664 }
2665 
2666 layer {
2667     bottom: "res4b5_branch2c"
2668     top: "res4b5_branch2c"
2669     name: "scale4b5_branch2c"
2670     type: "Scale"
2671     scale_param {
2672         bias_term: true
2673     }
2674 }
2675 
2676 layer {
2677     bottom: "res4b4"
2678     bottom: "res4b5_branch2c"
2679     top: "res4b5"
2680     name: "res4b5"
2681     type: "Eltwise"
2682     eltwise_param {
2683         operation: SUM
2684     }
2685 }
2686 
2687 layer {
2688     bottom: "res4b5"
2689     top: "res4b5"
2690     name: "res4b5_relu"
2691     type: "ReLU"
2692 }
2693 
2694 layer {
2695     bottom: "res4b5"
2696     top: "res4b6_branch2a"
2697     name: "res4b6_branch2a"
2698     type: "Convolution"
2699     convolution_param {
2700         num_output: 256
2701         kernel_size: 1
2702         pad: 0
2703         stride: 1
2704         weight_filler {
2705             type: "msra"
2706         }
2707         bias_term: false
2708 
2709     }
2710 }
2711 
2712 layer {
2713     bottom: "res4b6_branch2a"
2714     top: "res4b6_branch2a"
2715     name: "bn4b6_branch2a"
2716     type: "BatchNorm"
2717     batch_norm_param {
2718         use_global_stats: false
2719     }
2720 }
2721 
2722 layer {
2723     bottom: "res4b6_branch2a"
2724     top: "res4b6_branch2a"
2725     name: "scale4b6_branch2a"
2726     type: "Scale"
2727     scale_param {
2728         bias_term: true
2729     }
2730 }
2731 
2732 layer {
2733     bottom: "res4b6_branch2a"
2734     top: "res4b6_branch2a"
2735     name: "res4b6_branch2a_relu"
2736     type: "ReLU"
2737 }
2738 
2739 layer {
2740     bottom: "res4b6_branch2a"
2741     top: "res4b6_branch2b"
2742     name: "res4b6_branch2b"
2743     type: "Convolution"
2744     convolution_param {
2745         num_output: 256
2746         kernel_size: 3
2747         pad: 1
2748         stride: 1
2749         weight_filler {
2750             type: "msra"
2751         }
2752         bias_term: false
2753 
2754     }
2755 }
2756 
2757 layer {
2758     bottom: "res4b6_branch2b"
2759     top: "res4b6_branch2b"
2760     name: "bn4b6_branch2b"
2761     type: "BatchNorm"
2762     batch_norm_param {
2763         use_global_stats: false
2764     }
2765 }
2766 
2767 layer {
2768     bottom: "res4b6_branch2b"
2769     top: "res4b6_branch2b"
2770     name: "scale4b6_branch2b"
2771     type: "Scale"
2772     scale_param {
2773         bias_term: true
2774     }
2775 }
2776 
2777 layer {
2778     bottom: "res4b6_branch2b"
2779     top: "res4b6_branch2b"
2780     name: "res4b6_branch2b_relu"
2781     type: "ReLU"
2782 }
2783 
2784 layer {
2785     bottom: "res4b6_branch2b"
2786     top: "res4b6_branch2c"
2787     name: "res4b6_branch2c"
2788     type: "Convolution"
2789     convolution_param {
2790         num_output: 1024
2791         kernel_size: 1
2792         pad: 0
2793         stride: 1
2794         weight_filler {
2795             type: "msra"
2796         }
2797         bias_term: false
2798 
2799     }
2800 }
2801 
2802 layer {
2803     bottom: "res4b6_branch2c"
2804     top: "res4b6_branch2c"
2805     name: "bn4b6_branch2c"
2806     type: "BatchNorm"
2807     batch_norm_param {
2808         use_global_stats: false
2809     }
2810 }
2811 
2812 layer {
2813     bottom: "res4b6_branch2c"
2814     top: "res4b6_branch2c"
2815     name: "scale4b6_branch2c"
2816     type: "Scale"
2817     scale_param {
2818         bias_term: true
2819     }
2820 }
2821 
2822 layer {
2823     bottom: "res4b5"
2824     bottom: "res4b6_branch2c"
2825     top: "res4b6"
2826     name: "res4b6"
2827     type: "Eltwise"
2828     eltwise_param {
2829         operation: SUM
2830     }
2831 }
2832 
2833 layer {
2834     bottom: "res4b6"
2835     top: "res4b6"
2836     name: "res4b6_relu"
2837     type: "ReLU"
2838 }
2839 
2840 layer {
2841     bottom: "res4b6"
2842     top: "res4b7_branch2a"
2843     name: "res4b7_branch2a"
2844     type: "Convolution"
2845     convolution_param {
2846         num_output: 256
2847         kernel_size: 1
2848         pad: 0
2849         stride: 1
2850         weight_filler {
2851             type: "msra"
2852         }
2853         bias_term: false
2854 
2855     }
2856 }
2857 
2858 layer {
2859     bottom: "res4b7_branch2a"
2860     top: "res4b7_branch2a"
2861     name: "bn4b7_branch2a"
2862     type: "BatchNorm"
2863     batch_norm_param {
2864         use_global_stats: false
2865     }
2866 }
2867 
2868 layer {
2869     bottom: "res4b7_branch2a"
2870     top: "res4b7_branch2a"
2871     name: "scale4b7_branch2a"
2872     type: "Scale"
2873     scale_param {
2874         bias_term: true
2875     }
2876 }
2877 
2878 layer {
2879     bottom: "res4b7_branch2a"
2880     top: "res4b7_branch2a"
2881     name: "res4b7_branch2a_relu"
2882     type: "ReLU"
2883 }
2884 
2885 layer {
2886     bottom: "res4b7_branch2a"
2887     top: "res4b7_branch2b"
2888     name: "res4b7_branch2b"
2889     type: "Convolution"
2890     convolution_param {
2891         num_output: 256
2892         kernel_size: 3
2893         pad: 1
2894         stride: 1
2895         weight_filler {
2896             type: "msra"
2897         }
2898         bias_term: false
2899 
2900     }
2901 }
2902 
2903 layer {
2904     bottom: "res4b7_branch2b"
2905     top: "res4b7_branch2b"
2906     name: "bn4b7_branch2b"
2907     type: "BatchNorm"
2908     batch_norm_param {
2909         use_global_stats: false
2910     }
2911 }
2912 
2913 layer {
2914     bottom: "res4b7_branch2b"
2915     top: "res4b7_branch2b"
2916     name: "scale4b7_branch2b"
2917     type: "Scale"
2918     scale_param {
2919         bias_term: true
2920     }
2921 }
2922 
2923 layer {
2924     bottom: "res4b7_branch2b"
2925     top: "res4b7_branch2b"
2926     name: "res4b7_branch2b_relu"
2927     type: "ReLU"
2928 }
2929 
2930 layer {
2931     bottom: "res4b7_branch2b"
2932     top: "res4b7_branch2c"
2933     name: "res4b7_branch2c"
2934     type: "Convolution"
2935     convolution_param {
2936         num_output: 1024
2937         kernel_size: 1
2938         pad: 0
2939         stride: 1
2940         weight_filler {
2941             type: "msra"
2942         }
2943         bias_term: false
2944 
2945     }
2946 }
2947 
2948 layer {
2949     bottom: "res4b7_branch2c"
2950     top: "res4b7_branch2c"
2951     name: "bn4b7_branch2c"
2952     type: "BatchNorm"
2953     batch_norm_param {
2954         use_global_stats: false
2955     }
2956 }
2957 
2958 layer {
2959     bottom: "res4b7_branch2c"
2960     top: "res4b7_branch2c"
2961     name: "scale4b7_branch2c"
2962     type: "Scale"
2963     scale_param {
2964         bias_term: true
2965     }
2966 }
2967 
2968 layer {
2969     bottom: "res4b6"
2970     bottom: "res4b7_branch2c"
2971     top: "res4b7"
2972     name: "res4b7"
2973     type: "Eltwise"
2974     eltwise_param {
2975         operation: SUM
2976     }
2977 }
2978 
2979 layer {
2980     bottom: "res4b7"
2981     top: "res4b7"
2982     name: "res4b7_relu"
2983     type: "ReLU"
2984 }
2985 
2986 layer {
2987     bottom: "res4b7"
2988     top: "res4b8_branch2a"
2989     name: "res4b8_branch2a"
2990     type: "Convolution"
2991     convolution_param {
2992         num_output: 256
2993         kernel_size: 1
2994         pad: 0
2995         stride: 1
2996         weight_filler {
2997             type: "msra"
2998         }
2999         bias_term: false
3000 
3001     }
3002 }
3003 
3004 layer {
3005     bottom: "res4b8_branch2a"
3006     top: "res4b8_branch2a"
3007     name: "bn4b8_branch2a"
3008     type: "BatchNorm"
3009     batch_norm_param {
3010         use_global_stats: false
3011     }
3012 }
3013 
3014 layer {
3015     bottom: "res4b8_branch2a"
3016     top: "res4b8_branch2a"
3017     name: "scale4b8_branch2a"
3018     type: "Scale"
3019     scale_param {
3020         bias_term: true
3021     }
3022 }
3023 
3024 layer {
3025     bottom: "res4b8_branch2a"
3026     top: "res4b8_branch2a"
3027     name: "res4b8_branch2a_relu"
3028     type: "ReLU"
3029 }
3030 
3031 layer {
3032     bottom: "res4b8_branch2a"
3033     top: "res4b8_branch2b"
3034     name: "res4b8_branch2b"
3035     type: "Convolution"
3036     convolution_param {
3037         num_output: 256
3038         kernel_size: 3
3039         pad: 1
3040         stride: 1
3041         weight_filler {
3042             type: "msra"
3043         }
3044         bias_term: false
3045 
3046     }
3047 }
3048 
3049 layer {
3050     bottom: "res4b8_branch2b"
3051     top: "res4b8_branch2b"
3052     name: "bn4b8_branch2b"
3053     type: "BatchNorm"
3054     batch_norm_param {
3055         use_global_stats: false
3056     }
3057 }
3058 
3059 layer {
3060     bottom: "res4b8_branch2b"
3061     top: "res4b8_branch2b"
3062     name: "scale4b8_branch2b"
3063     type: "Scale"
3064     scale_param {
3065         bias_term: true
3066     }
3067 }
3068 
3069 layer {
3070     bottom: "res4b8_branch2b"
3071     top: "res4b8_branch2b"
3072     name: "res4b8_branch2b_relu"
3073     type: "ReLU"
3074 }
3075 
3076 layer {
3077     bottom: "res4b8_branch2b"
3078     top: "res4b8_branch2c"
3079     name: "res4b8_branch2c"
3080     type: "Convolution"
3081     convolution_param {
3082         num_output: 1024
3083         kernel_size: 1
3084         pad: 0
3085         stride: 1
3086         weight_filler {
3087             type: "msra"
3088         }
3089         bias_term: false
3090 
3091     }
3092 }
3093 
3094 layer {
3095     bottom: "res4b8_branch2c"
3096     top: "res4b8_branch2c"
3097     name: "bn4b8_branch2c"
3098     type: "BatchNorm"
3099     batch_norm_param {
3100         use_global_stats: false
3101     }
3102 }
3103 
3104 layer {
3105     bottom: "res4b8_branch2c"
3106     top: "res4b8_branch2c"
3107     name: "scale4b8_branch2c"
3108     type: "Scale"
3109     scale_param {
3110         bias_term: true
3111     }
3112 }
3113 
3114 layer {
3115     bottom: "res4b7"
3116     bottom: "res4b8_branch2c"
3117     top: "res4b8"
3118     name: "res4b8"
3119     type: "Eltwise"
3120     eltwise_param {
3121         operation: SUM
3122     }
3123 }
3124 
3125 layer {
3126     bottom: "res4b8"
3127     top: "res4b8"
3128     name: "res4b8_relu"
3129     type: "ReLU"
3130 }
3131 
3132 layer {
3133     bottom: "res4b8"
3134     top: "res4b9_branch2a"
3135     name: "res4b9_branch2a"
3136     type: "Convolution"
3137     convolution_param {
3138         num_output: 256
3139         kernel_size: 1
3140         pad: 0
3141         stride: 1
3142         weight_filler {
3143             type: "msra"
3144         }
3145         bias_term: false
3146 
3147     }
3148 }
3149 
3150 layer {
3151     bottom: "res4b9_branch2a"
3152     top: "res4b9_branch2a"
3153     name: "bn4b9_branch2a"
3154     type: "BatchNorm"
3155     batch_norm_param {
3156         use_global_stats: false
3157     }
3158 }
3159 
3160 layer {
3161     bottom: "res4b9_branch2a"
3162     top: "res4b9_branch2a"
3163     name: "scale4b9_branch2a"
3164     type: "Scale"
3165     scale_param {
3166         bias_term: true
3167     }
3168 }
3169 
3170 layer {
3171     bottom: "res4b9_branch2a"
3172     top: "res4b9_branch2a"
3173     name: "res4b9_branch2a_relu"
3174     type: "ReLU"
3175 }
3176 
3177 layer {
3178     bottom: "res4b9_branch2a"
3179     top: "res4b9_branch2b"
3180     name: "res4b9_branch2b"
3181     type: "Convolution"
3182     convolution_param {
3183         num_output: 256
3184         kernel_size: 3
3185         pad: 1
3186         stride: 1
3187         weight_filler {
3188             type: "msra"
3189         }
3190         bias_term: false
3191 
3192     }
3193 }
3194 
3195 layer {
3196     bottom: "res4b9_branch2b"
3197     top: "res4b9_branch2b"
3198     name: "bn4b9_branch2b"
3199     type: "BatchNorm"
3200     batch_norm_param {
3201         use_global_stats: false
3202     }
3203 }
3204 
3205 layer {
3206     bottom: "res4b9_branch2b"
3207     top: "res4b9_branch2b"
3208     name: "scale4b9_branch2b"
3209     type: "Scale"
3210     scale_param {
3211         bias_term: true
3212     }
3213 }
3214 
3215 layer {
3216     bottom: "res4b9_branch2b"
3217     top: "res4b9_branch2b"
3218     name: "res4b9_branch2b_relu"
3219     type: "ReLU"
3220 }
3221 
3222 layer {
3223     bottom: "res4b9_branch2b"
3224     top: "res4b9_branch2c"
3225     name: "res4b9_branch2c"
3226     type: "Convolution"
3227     convolution_param {
3228         num_output: 1024
3229         kernel_size: 1
3230         pad: 0
3231         stride: 1
3232         weight_filler {
3233             type: "msra"
3234         }
3235         bias_term: false
3236 
3237     }
3238 }
3239 
3240 layer {
3241     bottom: "res4b9_branch2c"
3242     top: "res4b9_branch2c"
3243     name: "bn4b9_branch2c"
3244     type: "BatchNorm"
3245     batch_norm_param {
3246         use_global_stats: false
3247     }
3248 }
3249 
3250 layer {
3251     bottom: "res4b9_branch2c"
3252     top: "res4b9_branch2c"
3253     name: "scale4b9_branch2c"
3254     type: "Scale"
3255     scale_param {
3256         bias_term: true
3257     }
3258 }
3259 
3260 layer {
3261     bottom: "res4b8"
3262     bottom: "res4b9_branch2c"
3263     top: "res4b9"
3264     name: "res4b9"
3265     type: "Eltwise"
3266     eltwise_param {
3267         operation: SUM
3268     }
3269 }
3270 
3271 layer {
3272     bottom: "res4b9"
3273     top: "res4b9"
3274     name: "res4b9_relu"
3275     type: "ReLU"
3276 }
3277 
3278 layer {
3279     bottom: "res4b9"
3280     top: "res4b10_branch2a"
3281     name: "res4b10_branch2a"
3282     type: "Convolution"
3283     convolution_param {
3284         num_output: 256
3285         kernel_size: 1
3286         pad: 0
3287         stride: 1
3288         weight_filler {
3289             type: "msra"
3290         }
3291         bias_term: false
3292 
3293     }
3294 }
3295 
3296 layer {
3297     bottom: "res4b10_branch2a"
3298     top: "res4b10_branch2a"
3299     name: "bn4b10_branch2a"
3300     type: "BatchNorm"
3301     batch_norm_param {
3302         use_global_stats: false
3303     }
3304 }
3305 
3306 layer {
3307     bottom: "res4b10_branch2a"
3308     top: "res4b10_branch2a"
3309     name: "scale4b10_branch2a"
3310     type: "Scale"
3311     scale_param {
3312         bias_term: true
3313     }
3314 }
3315 
3316 layer {
3317     bottom: "res4b10_branch2a"
3318     top: "res4b10_branch2a"
3319     name: "res4b10_branch2a_relu"
3320     type: "ReLU"
3321 }
3322 
3323 layer {
3324     bottom: "res4b10_branch2a"
3325     top: "res4b10_branch2b"
3326     name: "res4b10_branch2b"
3327     type: "Convolution"
3328     convolution_param {
3329         num_output: 256
3330         kernel_size: 3
3331         pad: 1
3332         stride: 1
3333         weight_filler {
3334             type: "msra"
3335         }
3336         bias_term: false
3337 
3338     }
3339 }
3340 
3341 layer {
3342     bottom: "res4b10_branch2b"
3343     top: "res4b10_branch2b"
3344     name: "bn4b10_branch2b"
3345     type: "BatchNorm"
3346     batch_norm_param {
3347         use_global_stats: false
3348     }
3349 }
3350 
3351 layer {
3352     bottom: "res4b10_branch2b"
3353     top: "res4b10_branch2b"
3354     name: "scale4b10_branch2b"
3355     type: "Scale"
3356     scale_param {
3357         bias_term: true
3358     }
3359 }
3360 
3361 layer {
3362     bottom: "res4b10_branch2b"
3363     top: "res4b10_branch2b"
3364     name: "res4b10_branch2b_relu"
3365     type: "ReLU"
3366 }
3367 
3368 layer {
3369     bottom: "res4b10_branch2b"
3370     top: "res4b10_branch2c"
3371     name: "res4b10_branch2c"
3372     type: "Convolution"
3373     convolution_param {
3374         num_output: 1024
3375         kernel_size: 1
3376         pad: 0
3377         stride: 1
3378         weight_filler {
3379             type: "msra"
3380         }
3381         bias_term: false
3382 
3383     }
3384 }
3385 
3386 layer {
3387     bottom: "res4b10_branch2c"
3388     top: "res4b10_branch2c"
3389     name: "bn4b10_branch2c"
3390     type: "BatchNorm"
3391     batch_norm_param {
3392         use_global_stats: false
3393     }
3394 }
3395 
3396 layer {
3397     bottom: "res4b10_branch2c"
3398     top: "res4b10_branch2c"
3399     name: "scale4b10_branch2c"
3400     type: "Scale"
3401     scale_param {
3402         bias_term: true
3403     }
3404 }
3405 
3406 layer {
3407     bottom: "res4b9"
3408     bottom: "res4b10_branch2c"
3409     top: "res4b10"
3410     name: "res4b10"
3411     type: "Eltwise"
3412     eltwise_param {
3413         operation: SUM
3414     }
3415 }
3416 
3417 layer {
3418     bottom: "res4b10"
3419     top: "res4b10"
3420     name: "res4b10_relu"
3421     type: "ReLU"
3422 }
3423 
3424 layer {
3425     bottom: "res4b10"
3426     top: "res4b11_branch2a"
3427     name: "res4b11_branch2a"
3428     type: "Convolution"
3429     convolution_param {
3430         num_output: 256
3431         kernel_size: 1
3432         pad: 0
3433         stride: 1
3434         weight_filler {
3435             type: "msra"
3436         }
3437         bias_term: false
3438 
3439     }
3440 }
3441 
3442 layer {
3443     bottom: "res4b11_branch2a"
3444     top: "res4b11_branch2a"
3445     name: "bn4b11_branch2a"
3446     type: "BatchNorm"
3447     batch_norm_param {
3448         use_global_stats: false
3449     }
3450 }
3451 
3452 layer {
3453     bottom: "res4b11_branch2a"
3454     top: "res4b11_branch2a"
3455     name: "scale4b11_branch2a"
3456     type: "Scale"
3457     scale_param {
3458         bias_term: true
3459     }
3460 }
3461 
3462 layer {
3463     bottom: "res4b11_branch2a"
3464     top: "res4b11_branch2a"
3465     name: "res4b11_branch2a_relu"
3466     type: "ReLU"
3467 }
3468 
3469 layer {
3470     bottom: "res4b11_branch2a"
3471     top: "res4b11_branch2b"
3472     name: "res4b11_branch2b"
3473     type: "Convolution"
3474     convolution_param {
3475         num_output: 256
3476         kernel_size: 3
3477         pad: 1
3478         stride: 1
3479         weight_filler {
3480             type: "msra"
3481         }
3482         bias_term: false
3483 
3484     }
3485 }
3486 
3487 layer {
3488     bottom: "res4b11_branch2b"
3489     top: "res4b11_branch2b"
3490     name: "bn4b11_branch2b"
3491     type: "BatchNorm"
3492     batch_norm_param {
3493         use_global_stats: false
3494     }
3495 }
3496 
3497 layer {
3498     bottom: "res4b11_branch2b"
3499     top: "res4b11_branch2b"
3500     name: "scale4b11_branch2b"
3501     type: "Scale"
3502     scale_param {
3503         bias_term: true
3504     }
3505 }
3506 
3507 layer {
3508     bottom: "res4b11_branch2b"
3509     top: "res4b11_branch2b"
3510     name: "res4b11_branch2b_relu"
3511     type: "ReLU"
3512 }
3513 
3514 layer {
3515     bottom: "res4b11_branch2b"
3516     top: "res4b11_branch2c"
3517     name: "res4b11_branch2c"
3518     type: "Convolution"
3519     convolution_param {
3520         num_output: 1024
3521         kernel_size: 1
3522         pad: 0
3523         stride: 1
3524         weight_filler {
3525             type: "msra"
3526         }
3527         bias_term: false
3528 
3529     }
3530 }
3531 
3532 layer {
3533     bottom: "res4b11_branch2c"
3534     top: "res4b11_branch2c"
3535     name: "bn4b11_branch2c"
3536     type: "BatchNorm"
3537     batch_norm_param {
3538         use_global_stats: false
3539     }
3540 }
3541 
3542 layer {
3543     bottom: "res4b11_branch2c"
3544     top: "res4b11_branch2c"
3545     name: "scale4b11_branch2c"
3546     type: "Scale"
3547     scale_param {
3548         bias_term: true
3549     }
3550 }
3551 
3552 layer {
3553     bottom: "res4b10"
3554     bottom: "res4b11_branch2c"
3555     top: "res4b11"
3556     name: "res4b11"
3557     type: "Eltwise"
3558     eltwise_param {
3559         operation: SUM
3560     }
3561 }
3562 
3563 layer {
3564     bottom: "res4b11"
3565     top: "res4b11"
3566     name: "res4b11_relu"
3567     type: "ReLU"
3568 }
3569 
3570 layer {
3571     bottom: "res4b11"
3572     top: "res4b12_branch2a"
3573     name: "res4b12_branch2a"
3574     type: "Convolution"
3575     convolution_param {
3576         num_output: 256
3577         kernel_size: 1
3578         pad: 0
3579         stride: 1
3580         weight_filler {
3581             type: "msra"
3582         }
3583         bias_term: false
3584 
3585     }
3586 }
3587 
3588 layer {
3589     bottom: "res4b12_branch2a"
3590     top: "res4b12_branch2a"
3591     name: "bn4b12_branch2a"
3592     type: "BatchNorm"
3593     batch_norm_param {
3594         use_global_stats: false
3595     }
3596 }
3597 
3598 layer {
3599     bottom: "res4b12_branch2a"
3600     top: "res4b12_branch2a"
3601     name: "scale4b12_branch2a"
3602     type: "Scale"
3603     scale_param {
3604         bias_term: true
3605     }
3606 }
3607 
3608 layer {
3609     bottom: "res4b12_branch2a"
3610     top: "res4b12_branch2a"
3611     name: "res4b12_branch2a_relu"
3612     type: "ReLU"
3613 }
3614 
3615 layer {
3616     bottom: "res4b12_branch2a"
3617     top: "res4b12_branch2b"
3618     name: "res4b12_branch2b"
3619     type: "Convolution"
3620     convolution_param {
3621         num_output: 256
3622         kernel_size: 3
3623         pad: 1
3624         stride: 1
3625         weight_filler {
3626             type: "msra"
3627         }
3628         bias_term: false
3629 
3630     }
3631 }
3632 
3633 layer {
3634     bottom: "res4b12_branch2b"
3635     top: "res4b12_branch2b"
3636     name: "bn4b12_branch2b"
3637     type: "BatchNorm"
3638     batch_norm_param {
3639         use_global_stats: false
3640     }
3641 }
3642 
3643 layer {
3644     bottom: "res4b12_branch2b"
3645     top: "res4b12_branch2b"
3646     name: "scale4b12_branch2b"
3647     type: "Scale"
3648     scale_param {
3649         bias_term: true
3650     }
3651 }
3652 
3653 layer {
3654     bottom: "res4b12_branch2b"
3655     top: "res4b12_branch2b"
3656     name: "res4b12_branch2b_relu"
3657     type: "ReLU"
3658 }
3659 
3660 layer {
3661     bottom: "res4b12_branch2b"
3662     top: "res4b12_branch2c"
3663     name: "res4b12_branch2c"
3664     type: "Convolution"
3665     convolution_param {
3666         num_output: 1024
3667         kernel_size: 1
3668         pad: 0
3669         stride: 1
3670         weight_filler {
3671             type: "msra"
3672         }
3673         bias_term: false
3674 
3675     }
3676 }
3677 
3678 layer {
3679     bottom: "res4b12_branch2c"
3680     top: "res4b12_branch2c"
3681     name: "bn4b12_branch2c"
3682     type: "BatchNorm"
3683     batch_norm_param {
3684         use_global_stats: false
3685     }
3686 }
3687 
3688 layer {
3689     bottom: "res4b12_branch2c"
3690     top: "res4b12_branch2c"
3691     name: "scale4b12_branch2c"
3692     type: "Scale"
3693     scale_param {
3694         bias_term: true
3695     }
3696 }
3697 
3698 layer {
3699     bottom: "res4b11"
3700     bottom: "res4b12_branch2c"
3701     top: "res4b12"
3702     name: "res4b12"
3703     type: "Eltwise"
3704     eltwise_param {
3705         operation: SUM
3706     }
3707 }
3708 
3709 layer {
3710     bottom: "res4b12"
3711     top: "res4b12"
3712     name: "res4b12_relu"
3713     type: "ReLU"
3714 }
3715 
3716 layer {
3717     bottom: "res4b12"
3718     top: "res4b13_branch2a"
3719     name: "res4b13_branch2a"
3720     type: "Convolution"
3721     convolution_param {
3722         num_output: 256
3723         kernel_size: 1
3724         pad: 0
3725         stride: 1
3726         weight_filler {
3727             type: "msra"
3728         }
3729         bias_term: false
3730 
3731     }
3732 }
3733 
3734 layer {
3735     bottom: "res4b13_branch2a"
3736     top: "res4b13_branch2a"
3737     name: "bn4b13_branch2a"
3738     type: "BatchNorm"
3739     batch_norm_param {
3740         use_global_stats: false
3741     }
3742 }
3743 
3744 layer {
3745     bottom: "res4b13_branch2a"
3746     top: "res4b13_branch2a"
3747     name: "scale4b13_branch2a"
3748     type: "Scale"
3749     scale_param {
3750         bias_term: true
3751     }
3752 }
3753 
3754 layer {
3755     bottom: "res4b13_branch2a"
3756     top: "res4b13_branch2a"
3757     name: "res4b13_branch2a_relu"
3758     type: "ReLU"
3759 }
3760 
3761 layer {
3762     bottom: "res4b13_branch2a"
3763     top: "res4b13_branch2b"
3764     name: "res4b13_branch2b"
3765     type: "Convolution"
3766     convolution_param {
3767         num_output: 256
3768         kernel_size: 3
3769         pad: 1
3770         stride: 1
3771         weight_filler {
3772             type: "msra"
3773         }
3774         bias_term: false
3775 
3776     }
3777 }
3778 
3779 layer {
3780     bottom: "res4b13_branch2b"
3781     top: "res4b13_branch2b"
3782     name: "bn4b13_branch2b"
3783     type: "BatchNorm"
3784     batch_norm_param {
3785         use_global_stats: false
3786     }
3787 }
3788 
3789 layer {
3790     bottom: "res4b13_branch2b"
3791     top: "res4b13_branch2b"
3792     name: "scale4b13_branch2b"
3793     type: "Scale"
3794     scale_param {
3795         bias_term: true
3796     }
3797 }
3798 
3799 layer {
3800     bottom: "res4b13_branch2b"
3801     top: "res4b13_branch2b"
3802     name: "res4b13_branch2b_relu"
3803     type: "ReLU"
3804 }
3805 
3806 layer {
3807     bottom: "res4b13_branch2b"
3808     top: "res4b13_branch2c"
3809     name: "res4b13_branch2c"
3810     type: "Convolution"
3811     convolution_param {
3812         num_output: 1024
3813         kernel_size: 1
3814         pad: 0
3815         stride: 1
3816         weight_filler {
3817             type: "msra"
3818         }
3819         bias_term: false
3820 
3821     }
3822 }
3823 
3824 layer {
3825     bottom: "res4b13_branch2c"
3826     top: "res4b13_branch2c"
3827     name: "bn4b13_branch2c"
3828     type: "BatchNorm"
3829     batch_norm_param {
3830         use_global_stats: false
3831     }
3832 }
3833 
3834 layer {
3835     bottom: "res4b13_branch2c"
3836     top: "res4b13_branch2c"
3837     name: "scale4b13_branch2c"
3838     type: "Scale"
3839     scale_param {
3840         bias_term: true
3841     }
3842 }
3843 
3844 layer {
3845     bottom: "res4b12"
3846     bottom: "res4b13_branch2c"
3847     top: "res4b13"
3848     name: "res4b13"
3849     type: "Eltwise"
3850     eltwise_param {
3851         operation: SUM
3852     }
3853 }
3854 
3855 layer {
3856     bottom: "res4b13"
3857     top: "res4b13"
3858     name: "res4b13_relu"
3859     type: "ReLU"
3860 }
3861 
3862 layer {
3863     bottom: "res4b13"
3864     top: "res4b14_branch2a"
3865     name: "res4b14_branch2a"
3866     type: "Convolution"
3867     convolution_param {
3868         num_output: 256
3869         kernel_size: 1
3870         pad: 0
3871         stride: 1
3872         weight_filler {
3873             type: "msra"
3874         }
3875         bias_term: false
3876 
3877     }
3878 }
3879 
3880 layer {
3881     bottom: "res4b14_branch2a"
3882     top: "res4b14_branch2a"
3883     name: "bn4b14_branch2a"
3884     type: "BatchNorm"
3885     batch_norm_param {
3886         use_global_stats: false
3887     }
3888 }
3889 
3890 layer {
3891     bottom: "res4b14_branch2a"
3892     top: "res4b14_branch2a"
3893     name: "scale4b14_branch2a"
3894     type: "Scale"
3895     scale_param {
3896         bias_term: true
3897     }
3898 }
3899 
3900 layer {
3901     bottom: "res4b14_branch2a"
3902     top: "res4b14_branch2a"
3903     name: "res4b14_branch2a_relu"
3904     type: "ReLU"
3905 }
3906 
3907 layer {
3908     bottom: "res4b14_branch2a"
3909     top: "res4b14_branch2b"
3910     name: "res4b14_branch2b"
3911     type: "Convolution"
3912     convolution_param {
3913         num_output: 256
3914         kernel_size: 3
3915         pad: 1
3916         stride: 1
3917         weight_filler {
3918             type: "msra"
3919         }
3920         bias_term: false
3921 
3922     }
3923 }
3924 
3925 layer {
3926     bottom: "res4b14_branch2b"
3927     top: "res4b14_branch2b"
3928     name: "bn4b14_branch2b"
3929     type: "BatchNorm"
3930     batch_norm_param {
3931         use_global_stats: false
3932     }
3933 }
3934 
3935 layer {
3936     bottom: "res4b14_branch2b"
3937     top: "res4b14_branch2b"
3938     name: "scale4b14_branch2b"
3939     type: "Scale"
3940     scale_param {
3941         bias_term: true
3942     }
3943 }
3944 
3945 layer {
3946     bottom: "res4b14_branch2b"
3947     top: "res4b14_branch2b"
3948     name: "res4b14_branch2b_relu"
3949     type: "ReLU"
3950 }
3951 
3952 layer {
3953     bottom: "res4b14_branch2b"
3954     top: "res4b14_branch2c"
3955     name: "res4b14_branch2c"
3956     type: "Convolution"
3957     convolution_param {
3958         num_output: 1024
3959         kernel_size: 1
3960         pad: 0
3961         stride: 1
3962         weight_filler {
3963             type: "msra"
3964         }
3965         bias_term: false
3966 
3967     }
3968 }
3969 
3970 layer {
3971     bottom: "res4b14_branch2c"
3972     top: "res4b14_branch2c"
3973     name: "bn4b14_branch2c"
3974     type: "BatchNorm"
3975     batch_norm_param {
3976         use_global_stats: false
3977     }
3978 }
3979 
3980 layer {
3981     bottom: "res4b14_branch2c"
3982     top: "res4b14_branch2c"
3983     name: "scale4b14_branch2c"
3984     type: "Scale"
3985     scale_param {
3986         bias_term: true
3987     }
3988 }
3989 
3990 layer {
3991     bottom: "res4b13"
3992     bottom: "res4b14_branch2c"
3993     top: "res4b14"
3994     name: "res4b14"
3995     type: "Eltwise"
3996     eltwise_param {
3997         operation: SUM
3998     }
3999 }
4000 
4001 layer {
4002     bottom: "res4b14"
4003     top: "res4b14"
4004     name: "res4b14_relu"
4005     type: "ReLU"
4006 }
4007 
4008 layer {
4009     bottom: "res4b14"
4010     top: "res4b15_branch2a"
4011     name: "res4b15_branch2a"
4012     type: "Convolution"
4013     convolution_param {
4014         num_output: 256
4015         kernel_size: 1
4016         pad: 0
4017         stride: 1
4018         weight_filler {
4019             type: "msra"
4020         }
4021         bias_term: false
4022 
4023     }
4024 }
4025 
4026 layer {
4027     bottom: "res4b15_branch2a"
4028     top: "res4b15_branch2a"
4029     name: "bn4b15_branch2a"
4030     type: "BatchNorm"
4031     batch_norm_param {
4032         use_global_stats: false
4033     }
4034 }
4035 
4036 layer {
4037     bottom: "res4b15_branch2a"
4038     top: "res4b15_branch2a"
4039     name: "scale4b15_branch2a"
4040     type: "Scale"
4041     scale_param {
4042         bias_term: true
4043     }
4044 }
4045 
4046 layer {
4047     bottom: "res4b15_branch2a"
4048     top: "res4b15_branch2a"
4049     name: "res4b15_branch2a_relu"
4050     type: "ReLU"
4051 }
4052 
4053 layer {
4054     bottom: "res4b15_branch2a"
4055     top: "res4b15_branch2b"
4056     name: "res4b15_branch2b"
4057     type: "Convolution"
4058     convolution_param {
4059         num_output: 256
4060         kernel_size: 3
4061         pad: 1
4062         stride: 1
4063         weight_filler {
4064             type: "msra"
4065         }
4066         bias_term: false
4067 
4068     }
4069 }
4070 
4071 layer {
4072     bottom: "res4b15_branch2b"
4073     top: "res4b15_branch2b"
4074     name: "bn4b15_branch2b"
4075     type: "BatchNorm"
4076     batch_norm_param {
4077         use_global_stats: false
4078     }
4079 }
4080 
4081 layer {
4082     bottom: "res4b15_branch2b"
4083     top: "res4b15_branch2b"
4084     name: "scale4b15_branch2b"
4085     type: "Scale"
4086     scale_param {
4087         bias_term: true
4088     }
4089 }
4090 
4091 layer {
4092     bottom: "res4b15_branch2b"
4093     top: "res4b15_branch2b"
4094     name: "res4b15_branch2b_relu"
4095     type: "ReLU"
4096 }
4097 
4098 layer {
4099     bottom: "res4b15_branch2b"
4100     top: "res4b15_branch2c"
4101     name: "res4b15_branch2c"
4102     type: "Convolution"
4103     convolution_param {
4104         num_output: 1024
4105         kernel_size: 1
4106         pad: 0
4107         stride: 1
4108         weight_filler {
4109             type: "msra"
4110         }
4111         bias_term: false
4112 
4113     }
4114 }
4115 
4116 layer {
4117     bottom: "res4b15_branch2c"
4118     top: "res4b15_branch2c"
4119     name: "bn4b15_branch2c"
4120     type: "BatchNorm"
4121     batch_norm_param {
4122         use_global_stats: false
4123     }
4124 }
4125 
4126 layer {
4127     bottom: "res4b15_branch2c"
4128     top: "res4b15_branch2c"
4129     name: "scale4b15_branch2c"
4130     type: "Scale"
4131     scale_param {
4132         bias_term: true
4133     }
4134 }
4135 
4136 layer {
4137     bottom: "res4b14"
4138     bottom: "res4b15_branch2c"
4139     top: "res4b15"
4140     name: "res4b15"
4141     type: "Eltwise"
4142     eltwise_param {
4143         operation: SUM
4144     }
4145 }
4146 
4147 layer {
4148     bottom: "res4b15"
4149     top: "res4b15"
4150     name: "res4b15_relu"
4151     type: "ReLU"
4152 }
4153 
4154 layer {
4155     bottom: "res4b15"
4156     top: "res4b16_branch2a"
4157     name: "res4b16_branch2a"
4158     type: "Convolution"
4159     convolution_param {
4160         num_output: 256
4161         kernel_size: 1
4162         pad: 0
4163         stride: 1
4164         weight_filler {
4165             type: "msra"
4166         }
4167         bias_term: false
4168 
4169     }
4170 }
4171 
4172 layer {
4173     bottom: "res4b16_branch2a"
4174     top: "res4b16_branch2a"
4175     name: "bn4b16_branch2a"
4176     type: "BatchNorm"
4177     batch_norm_param {
4178         use_global_stats: false
4179     }
4180 }
4181 
4182 layer {
4183     bottom: "res4b16_branch2a"
4184     top: "res4b16_branch2a"
4185     name: "scale4b16_branch2a"
4186     type: "Scale"
4187     scale_param {
4188         bias_term: true
4189     }
4190 }
4191 
4192 layer {
4193     bottom: "res4b16_branch2a"
4194     top: "res4b16_branch2a"
4195     name: "res4b16_branch2a_relu"
4196     type: "ReLU"
4197 }
4198 
4199 layer {
4200     bottom: "res4b16_branch2a"
4201     top: "res4b16_branch2b"
4202     name: "res4b16_branch2b"
4203     type: "Convolution"
4204     convolution_param {
4205         num_output: 256
4206         kernel_size: 3
4207         pad: 1
4208         stride: 1
4209         weight_filler {
4210             type: "msra"
4211         }
4212         bias_term: false
4213 
4214     }
4215 }
4216 
4217 layer {
4218     bottom: "res4b16_branch2b"
4219     top: "res4b16_branch2b"
4220     name: "bn4b16_branch2b"
4221     type: "BatchNorm"
4222     batch_norm_param {
4223         use_global_stats: false
4224     }
4225 }
4226 
4227 layer {
4228     bottom: "res4b16_branch2b"
4229     top: "res4b16_branch2b"
4230     name: "scale4b16_branch2b"
4231     type: "Scale"
4232     scale_param {
4233         bias_term: true
4234     }
4235 }
4236 
4237 layer {
4238     bottom: "res4b16_branch2b"
4239     top: "res4b16_branch2b"
4240     name: "res4b16_branch2b_relu"
4241     type: "ReLU"
4242 }
4243 
4244 layer {
4245     bottom: "res4b16_branch2b"
4246     top: "res4b16_branch2c"
4247     name: "res4b16_branch2c"
4248     type: "Convolution"
4249     convolution_param {
4250         num_output: 1024
4251         kernel_size: 1
4252         pad: 0
4253         stride: 1
4254         weight_filler {
4255             type: "msra"
4256         }
4257         bias_term: false
4258 
4259     }
4260 }
4261 
4262 layer {
4263     bottom: "res4b16_branch2c"
4264     top: "res4b16_branch2c"
4265     name: "bn4b16_branch2c"
4266     type: "BatchNorm"
4267     batch_norm_param {
4268         use_global_stats: false
4269     }
4270 }
4271 
4272 layer {
4273     bottom: "res4b16_branch2c"
4274     top: "res4b16_branch2c"
4275     name: "scale4b16_branch2c"
4276     type: "Scale"
4277     scale_param {
4278         bias_term: true
4279     }
4280 }
4281 
4282 layer {
4283     bottom: "res4b15"
4284     bottom: "res4b16_branch2c"
4285     top: "res4b16"
4286     name: "res4b16"
4287     type: "Eltwise"
4288     eltwise_param {
4289         operation: SUM
4290     }
4291 }
4292 
4293 layer {
4294     bottom: "res4b16"
4295     top: "res4b16"
4296     name: "res4b16_relu"
4297     type: "ReLU"
4298 }
4299 
4300 layer {
4301     bottom: "res4b16"
4302     top: "res4b17_branch2a"
4303     name: "res4b17_branch2a"
4304     type: "Convolution"
4305     convolution_param {
4306         num_output: 256
4307         kernel_size: 1
4308         pad: 0
4309         stride: 1
4310         weight_filler {
4311             type: "msra"
4312         }
4313         bias_term: false
4314 
4315     }
4316 }
4317 
4318 layer {
4319     bottom: "res4b17_branch2a"
4320     top: "res4b17_branch2a"
4321     name: "bn4b17_branch2a"
4322     type: "BatchNorm"
4323     batch_norm_param {
4324         use_global_stats: false
4325     }
4326 }
4327 
4328 layer {
4329     bottom: "res4b17_branch2a"
4330     top: "res4b17_branch2a"
4331     name: "scale4b17_branch2a"
4332     type: "Scale"
4333     scale_param {
4334         bias_term: true
4335     }
4336 }
4337 
4338 layer {
4339     bottom: "res4b17_branch2a"
4340     top: "res4b17_branch2a"
4341     name: "res4b17_branch2a_relu"
4342     type: "ReLU"
4343 }
4344 
4345 layer {
4346     bottom: "res4b17_branch2a"
4347     top: "res4b17_branch2b"
4348     name: "res4b17_branch2b"
4349     type: "Convolution"
4350     convolution_param {
4351         num_output: 256
4352         kernel_size: 3
4353         pad: 1
4354         stride: 1
4355         weight_filler {
4356             type: "msra"
4357         }
4358         bias_term: false
4359 
4360     }
4361 }
4362 
4363 layer {
4364     bottom: "res4b17_branch2b"
4365     top: "res4b17_branch2b"
4366     name: "bn4b17_branch2b"
4367     type: "BatchNorm"
4368     batch_norm_param {
4369         use_global_stats: false
4370     }
4371 }
4372 
4373 layer {
4374     bottom: "res4b17_branch2b"
4375     top: "res4b17_branch2b"
4376     name: "scale4b17_branch2b"
4377     type: "Scale"
4378     scale_param {
4379         bias_term: true
4380     }
4381 }
4382 
4383 layer {
4384     bottom: "res4b17_branch2b"
4385     top: "res4b17_branch2b"
4386     name: "res4b17_branch2b_relu"
4387     type: "ReLU"
4388 }
4389 
4390 layer {
4391     bottom: "res4b17_branch2b"
4392     top: "res4b17_branch2c"
4393     name: "res4b17_branch2c"
4394     type: "Convolution"
4395     convolution_param {
4396         num_output: 1024
4397         kernel_size: 1
4398         pad: 0
4399         stride: 1
4400         weight_filler {
4401             type: "msra"
4402         }
4403         bias_term: false
4404 
4405     }
4406 }
4407 
4408 layer {
4409     bottom: "res4b17_branch2c"
4410     top: "res4b17_branch2c"
4411     name: "bn4b17_branch2c"
4412     type: "BatchNorm"
4413     batch_norm_param {
4414         use_global_stats: false
4415     }
4416 }
4417 
4418 layer {
4419     bottom: "res4b17_branch2c"
4420     top: "res4b17_branch2c"
4421     name: "scale4b17_branch2c"
4422     type: "Scale"
4423     scale_param {
4424         bias_term: true
4425     }
4426 }
4427 
4428 layer {
4429     bottom: "res4b16"
4430     bottom: "res4b17_branch2c"
4431     top: "res4b17"
4432     name: "res4b17"
4433     type: "Eltwise"
4434     eltwise_param {
4435         operation: SUM
4436     }
4437 }
4438 
4439 layer {
4440     bottom: "res4b17"
4441     top: "res4b17"
4442     name: "res4b17_relu"
4443     type: "ReLU"
4444 }
4445 
4446 layer {
4447     bottom: "res4b17"
4448     top: "res4b18_branch2a"
4449     name: "res4b18_branch2a"
4450     type: "Convolution"
4451     convolution_param {
4452         num_output: 256
4453         kernel_size: 1
4454         pad: 0
4455         stride: 1
4456         weight_filler {
4457             type: "msra"
4458         }
4459         bias_term: false
4460 
4461     }
4462 }
4463 
4464 layer {
4465     bottom: "res4b18_branch2a"
4466     top: "res4b18_branch2a"
4467     name: "bn4b18_branch2a"
4468     type: "BatchNorm"
4469     batch_norm_param {
4470         use_global_stats: false
4471     }
4472 }
4473 
4474 layer {
4475     bottom: "res4b18_branch2a"
4476     top: "res4b18_branch2a"
4477     name: "scale4b18_branch2a"
4478     type: "Scale"
4479     scale_param {
4480         bias_term: true
4481     }
4482 }
4483 
4484 layer {
4485     bottom: "res4b18_branch2a"
4486     top: "res4b18_branch2a"
4487     name: "res4b18_branch2a_relu"
4488     type: "ReLU"
4489 }
4490 
4491 layer {
4492     bottom: "res4b18_branch2a"
4493     top: "res4b18_branch2b"
4494     name: "res4b18_branch2b"
4495     type: "Convolution"
4496     convolution_param {
4497         num_output: 256
4498         kernel_size: 3
4499         pad: 1
4500         stride: 1
4501         weight_filler {
4502             type: "msra"
4503         }
4504         bias_term: false
4505 
4506     }
4507 }
4508 
4509 layer {
4510     bottom: "res4b18_branch2b"
4511     top: "res4b18_branch2b"
4512     name: "bn4b18_branch2b"
4513     type: "BatchNorm"
4514     batch_norm_param {
4515         use_global_stats: false
4516     }
4517 }
4518 
4519 layer {
4520     bottom: "res4b18_branch2b"
4521     top: "res4b18_branch2b"
4522     name: "scale4b18_branch2b"
4523     type: "Scale"
4524     scale_param {
4525         bias_term: true
4526     }
4527 }
4528 
4529 layer {
4530     bottom: "res4b18_branch2b"
4531     top: "res4b18_branch2b"
4532     name: "res4b18_branch2b_relu"
4533     type: "ReLU"
4534 }
4535 
4536 layer {
4537     bottom: "res4b18_branch2b"
4538     top: "res4b18_branch2c"
4539     name: "res4b18_branch2c"
4540     type: "Convolution"
4541     convolution_param {
4542         num_output: 1024
4543         kernel_size: 1
4544         pad: 0
4545         stride: 1
4546         weight_filler {
4547             type: "msra"
4548         }
4549         bias_term: false
4550 
4551     }
4552 }
4553 
4554 layer {
4555     bottom: "res4b18_branch2c"
4556     top: "res4b18_branch2c"
4557     name: "bn4b18_branch2c"
4558     type: "BatchNorm"
4559     batch_norm_param {
4560         use_global_stats: false
4561     }
4562 }
4563 
4564 layer {
4565     bottom: "res4b18_branch2c"
4566     top: "res4b18_branch2c"
4567     name: "scale4b18_branch2c"
4568     type: "Scale"
4569     scale_param {
4570         bias_term: true
4571     }
4572 }
4573 
4574 layer {
4575     bottom: "res4b17"
4576     bottom: "res4b18_branch2c"
4577     top: "res4b18"
4578     name: "res4b18"
4579     type: "Eltwise"
4580     eltwise_param {
4581         operation: SUM
4582     }
4583 }
4584 
4585 layer {
4586     bottom: "res4b18"
4587     top: "res4b18"
4588     name: "res4b18_relu"
4589     type: "ReLU"
4590 }
4591 
4592 layer {
4593     bottom: "res4b18"
4594     top: "res4b19_branch2a"
4595     name: "res4b19_branch2a"
4596     type: "Convolution"
4597     convolution_param {
4598         num_output: 256
4599         kernel_size: 1
4600         pad: 0
4601         stride: 1
4602         weight_filler {
4603             type: "msra"
4604         }
4605         bias_term: false
4606 
4607     }
4608 }
4609 
4610 layer {
4611     bottom: "res4b19_branch2a"
4612     top: "res4b19_branch2a"
4613     name: "bn4b19_branch2a"
4614     type: "BatchNorm"
4615     batch_norm_param {
4616         use_global_stats: false
4617     }
4618 }
4619 
4620 layer {
4621     bottom: "res4b19_branch2a"
4622     top: "res4b19_branch2a"
4623     name: "scale4b19_branch2a"
4624     type: "Scale"
4625     scale_param {
4626         bias_term: true
4627     }
4628 }
4629 
4630 layer {
4631     bottom: "res4b19_branch2a"
4632     top: "res4b19_branch2a"
4633     name: "res4b19_branch2a_relu"
4634     type: "ReLU"
4635 }
4636 
4637 layer {
4638     bottom: "res4b19_branch2a"
4639     top: "res4b19_branch2b"
4640     name: "res4b19_branch2b"
4641     type: "Convolution"
4642     convolution_param {
4643         num_output: 256
4644         kernel_size: 3
4645         pad: 1
4646         stride: 1
4647         weight_filler {
4648             type: "msra"
4649         }
4650         bias_term: false
4651 
4652     }
4653 }
4654 
4655 layer {
4656     bottom: "res4b19_branch2b"
4657     top: "res4b19_branch2b"
4658     name: "bn4b19_branch2b"
4659     type: "BatchNorm"
4660     batch_norm_param {
4661         use_global_stats: false
4662     }
4663 }
4664 
4665 layer {
4666     bottom: "res4b19_branch2b"
4667     top: "res4b19_branch2b"
4668     name: "scale4b19_branch2b"
4669     type: "Scale"
4670     scale_param {
4671         bias_term: true
4672     }
4673 }
4674 
4675 layer {
4676     bottom: "res4b19_branch2b"
4677     top: "res4b19_branch2b"
4678     name: "res4b19_branch2b_relu"
4679     type: "ReLU"
4680 }
4681 
4682 layer {
4683     bottom: "res4b19_branch2b"
4684     top: "res4b19_branch2c"
4685     name: "res4b19_branch2c"
4686     type: "Convolution"
4687     convolution_param {
4688         num_output: 1024
4689         kernel_size: 1
4690         pad: 0
4691         stride: 1
4692         weight_filler {
4693             type: "msra"
4694         }
4695         bias_term: false
4696 
4697     }
4698 }
4699 
4700 layer {
4701     bottom: "res4b19_branch2c"
4702     top: "res4b19_branch2c"
4703     name: "bn4b19_branch2c"
4704     type: "BatchNorm"
4705     batch_norm_param {
4706         use_global_stats: false
4707     }
4708 }
4709 
4710 layer {
4711     bottom: "res4b19_branch2c"
4712     top: "res4b19_branch2c"
4713     name: "scale4b19_branch2c"
4714     type: "Scale"
4715     scale_param {
4716         bias_term: true
4717     }
4718 }
4719 
4720 layer {
4721     bottom: "res4b18"
4722     bottom: "res4b19_branch2c"
4723     top: "res4b19"
4724     name: "res4b19"
4725     type: "Eltwise"
4726     eltwise_param {
4727         operation: SUM
4728     }
4729 }
4730 
4731 layer {
4732     bottom: "res4b19"
4733     top: "res4b19"
4734     name: "res4b19_relu"
4735     type: "ReLU"
4736 }
4737 
4738 layer {
4739     bottom: "res4b19"
4740     top: "res4b20_branch2a"
4741     name: "res4b20_branch2a"
4742     type: "Convolution"
4743     convolution_param {
4744         num_output: 256
4745         kernel_size: 1
4746         pad: 0
4747         stride: 1
4748         weight_filler {
4749             type: "msra"
4750         }
4751         bias_term: false
4752 
4753     }
4754 }
4755 
4756 layer {
4757     bottom: "res4b20_branch2a"
4758     top: "res4b20_branch2a"
4759     name: "bn4b20_branch2a"
4760     type: "BatchNorm"
4761     batch_norm_param {
4762         use_global_stats: false
4763     }
4764 }
4765 
4766 layer {
4767     bottom: "res4b20_branch2a"
4768     top: "res4b20_branch2a"
4769     name: "scale4b20_branch2a"
4770     type: "Scale"
4771     scale_param {
4772         bias_term: true
4773     }
4774 }
4775 
4776 layer {
4777     bottom: "res4b20_branch2a"
4778     top: "res4b20_branch2a"
4779     name: "res4b20_branch2a_relu"
4780     type: "ReLU"
4781 }
4782 
4783 layer {
4784     bottom: "res4b20_branch2a"
4785     top: "res4b20_branch2b"
4786     name: "res4b20_branch2b"
4787     type: "Convolution"
4788     convolution_param {
4789         num_output: 256
4790         kernel_size: 3
4791         pad: 1
4792         stride: 1
4793         weight_filler {
4794             type: "msra"
4795         }
4796         bias_term: false
4797 
4798     }
4799 }
4800 
4801 layer {
4802     bottom: "res4b20_branch2b"
4803     top: "res4b20_branch2b"
4804     name: "bn4b20_branch2b"
4805     type: "BatchNorm"
4806     batch_norm_param {
4807         use_global_stats: false
4808     }
4809 }
4810 
4811 layer {
4812     bottom: "res4b20_branch2b"
4813     top: "res4b20_branch2b"
4814     name: "scale4b20_branch2b"
4815     type: "Scale"
4816     scale_param {
4817         bias_term: true
4818     }
4819 }
4820 
4821 layer {
4822     bottom: "res4b20_branch2b"
4823     top: "res4b20_branch2b"
4824     name: "res4b20_branch2b_relu"
4825     type: "ReLU"
4826 }
4827 
4828 layer {
4829     bottom: "res4b20_branch2b"
4830     top: "res4b20_branch2c"
4831     name: "res4b20_branch2c"
4832     type: "Convolution"
4833     convolution_param {
4834         num_output: 1024
4835         kernel_size: 1
4836         pad: 0
4837         stride: 1
4838         weight_filler {
4839             type: "msra"
4840         }
4841         bias_term: false
4842 
4843     }
4844 }
4845 
4846 layer {
4847     bottom: "res4b20_branch2c"
4848     top: "res4b20_branch2c"
4849     name: "bn4b20_branch2c"
4850     type: "BatchNorm"
4851     batch_norm_param {
4852         use_global_stats: false
4853     }
4854 }
4855 
4856 layer {
4857     bottom: "res4b20_branch2c"
4858     top: "res4b20_branch2c"
4859     name: "scale4b20_branch2c"
4860     type: "Scale"
4861     scale_param {
4862         bias_term: true
4863     }
4864 }
4865 
4866 layer {
4867     bottom: "res4b19"
4868     bottom: "res4b20_branch2c"
4869     top: "res4b20"
4870     name: "res4b20"
4871     type: "Eltwise"
4872     eltwise_param {
4873         operation: SUM
4874     }
4875 }
4876 
4877 layer {
4878     bottom: "res4b20"
4879     top: "res4b20"
4880     name: "res4b20_relu"
4881     type: "ReLU"
4882 }
4883 
4884 layer {
4885     bottom: "res4b20"
4886     top: "res4b21_branch2a"
4887     name: "res4b21_branch2a"
4888     type: "Convolution"
4889     convolution_param {
4890         num_output: 256
4891         kernel_size: 1
4892         pad: 0
4893         stride: 1
4894         weight_filler {
4895             type: "msra"
4896         }
4897         bias_term: false
4898 
4899     }
4900 }
4901 
4902 layer {
4903     bottom: "res4b21_branch2a"
4904     top: "res4b21_branch2a"
4905     name: "bn4b21_branch2a"
4906     type: "BatchNorm"
4907     batch_norm_param {
4908         use_global_stats: false
4909     }
4910 }
4911 
4912 layer {
4913     bottom: "res4b21_branch2a"
4914     top: "res4b21_branch2a"
4915     name: "scale4b21_branch2a"
4916     type: "Scale"
4917     scale_param {
4918         bias_term: true
4919     }
4920 }
4921 
4922 layer {
4923     bottom: "res4b21_branch2a"
4924     top: "res4b21_branch2a"
4925     name: "res4b21_branch2a_relu"
4926     type: "ReLU"
4927 }
4928 
4929 layer {
4930     bottom: "res4b21_branch2a"
4931     top: "res4b21_branch2b"
4932     name: "res4b21_branch2b"
4933     type: "Convolution"
4934     convolution_param {
4935         num_output: 256
4936         kernel_size: 3
4937         pad: 1
4938         stride: 1
4939         weight_filler {
4940             type: "msra"
4941         }
4942         bias_term: false
4943 
4944     }
4945 }
4946 
4947 layer {
4948     bottom: "res4b21_branch2b"
4949     top: "res4b21_branch2b"
4950     name: "bn4b21_branch2b"
4951     type: "BatchNorm"
4952     batch_norm_param {
4953         use_global_stats: false
4954     }
4955 }
4956 
4957 layer {
4958     bottom: "res4b21_branch2b"
4959     top: "res4b21_branch2b"
4960     name: "scale4b21_branch2b"
4961     type: "Scale"
4962     scale_param {
4963         bias_term: true
4964     }
4965 }
4966 
4967 layer {
4968     bottom: "res4b21_branch2b"
4969     top: "res4b21_branch2b"
4970     name: "res4b21_branch2b_relu"
4971     type: "ReLU"
4972 }
4973 
4974 layer {
4975     bottom: "res4b21_branch2b"
4976     top: "res4b21_branch2c"
4977     name: "res4b21_branch2c"
4978     type: "Convolution"
4979     convolution_param {
4980         num_output: 1024
4981         kernel_size: 1
4982         pad: 0
4983         stride: 1
4984         weight_filler {
4985             type: "msra"
4986         }
4987         bias_term: false
4988 
4989     }
4990 }
4991 
4992 layer {
4993     bottom: "res4b21_branch2c"
4994     top: "res4b21_branch2c"
4995     name: "bn4b21_branch2c"
4996     type: "BatchNorm"
4997     batch_norm_param {
4998         use_global_stats: false
4999     }
5000 }
5001 
5002 layer {
5003     bottom: "res4b21_branch2c"
5004     top: "res4b21_branch2c"
5005     name: "scale4b21_branch2c"
5006     type: "Scale"
5007     scale_param {
5008         bias_term: true
5009     }
5010 }
5011 
5012 layer {
5013     bottom: "res4b20"
5014     bottom: "res4b21_branch2c"
5015     top: "res4b21"
5016     name: "res4b21"
5017     type: "Eltwise"
5018     eltwise_param {
5019         operation: SUM
5020     }
5021 }
5022 
5023 layer {
5024     bottom: "res4b21"
5025     top: "res4b21"
5026     name: "res4b21_relu"
5027     type: "ReLU"
5028 }
5029 
5030 layer {
5031     bottom: "res4b21"
5032     top: "res4b22_branch2a"
5033     name: "res4b22_branch2a"
5034     type: "Convolution"
5035     convolution_param {
5036         num_output: 256
5037         kernel_size: 1
5038         pad: 0
5039         stride: 1
5040         weight_filler {
5041             type: "msra"
5042         }
5043         bias_term: false
5044 
5045     }
5046 }
5047 
5048 layer {
5049     bottom: "res4b22_branch2a"
5050     top: "res4b22_branch2a"
5051     name: "bn4b22_branch2a"
5052     type: "BatchNorm"
5053     batch_norm_param {
5054         use_global_stats: false
5055     }
5056 }
5057 
5058 layer {
5059     bottom: "res4b22_branch2a"
5060     top: "res4b22_branch2a"
5061     name: "scale4b22_branch2a"
5062     type: "Scale"
5063     scale_param {
5064         bias_term: true
5065     }
5066 }
5067 
5068 layer {
5069     bottom: "res4b22_branch2a"
5070     top: "res4b22_branch2a"
5071     name: "res4b22_branch2a_relu"
5072     type: "ReLU"
5073 }
5074 
5075 layer {
5076     bottom: "res4b22_branch2a"
5077     top: "res4b22_branch2b"
5078     name: "res4b22_branch2b"
5079     type: "Convolution"
5080     convolution_param {
5081         num_output: 256
5082         kernel_size: 3
5083         pad: 1
5084         stride: 1
5085         weight_filler {
5086             type: "msra"
5087         }
5088         bias_term: false
5089 
5090     }
5091 }
5092 
5093 layer {
5094     bottom: "res4b22_branch2b"
5095     top: "res4b22_branch2b"
5096     name: "bn4b22_branch2b"
5097     type: "BatchNorm"
5098     batch_norm_param {
5099         use_global_stats: false
5100     }
5101 }
5102 
5103 layer {
5104     bottom: "res4b22_branch2b"
5105     top: "res4b22_branch2b"
5106     name: "scale4b22_branch2b"
5107     type: "Scale"
5108     scale_param {
5109         bias_term: true
5110     }
5111 }
5112 
5113 layer {
5114     bottom: "res4b22_branch2b"
5115     top: "res4b22_branch2b"
5116     name: "res4b22_branch2b_relu"
5117     type: "ReLU"
5118 }
5119 
5120 layer {
5121     bottom: "res4b22_branch2b"
5122     top: "res4b22_branch2c"
5123     name: "res4b22_branch2c"
5124     type: "Convolution"
5125     convolution_param {
5126         num_output: 1024
5127         kernel_size: 1
5128         pad: 0
5129         stride: 1
5130         weight_filler {
5131             type: "msra"
5132         }
5133         bias_term: false
5134 
5135     }
5136 }
5137 
5138 layer {
5139     bottom: "res4b22_branch2c"
5140     top: "res4b22_branch2c"
5141     name: "bn4b22_branch2c"
5142     type: "BatchNorm"
5143     batch_norm_param {
5144         use_global_stats: false
5145     }
5146 }
5147 
5148 layer {
5149     bottom: "res4b22_branch2c"
5150     top: "res4b22_branch2c"
5151     name: "scale4b22_branch2c"
5152     type: "Scale"
5153     scale_param {
5154         bias_term: true
5155     }
5156 }
5157 
5158 layer {
5159     bottom: "res4b21"
5160     bottom: "res4b22_branch2c"
5161     top: "res4b22"
5162     name: "res4b22"
5163     type: "Eltwise"
5164     eltwise_param {
5165         operation: SUM
5166     }
5167 }
5168 
5169 layer {
5170     bottom: "res4b22"
5171     top: "res4b22"
5172     name: "res4b22_relu"
5173     type: "ReLU"
5174 }
5175 
5176 layer {
5177     bottom: "res4b22"
5178     top: "res4b23_branch2a"
5179     name: "res4b23_branch2a"
5180     type: "Convolution"
5181     convolution_param {
5182         num_output: 256
5183         kernel_size: 1
5184         pad: 0
5185         stride: 1
5186         weight_filler {
5187             type: "msra"
5188         }
5189         bias_term: false
5190 
5191     }
5192 }
5193 
5194 layer {
5195     bottom: "res4b23_branch2a"
5196     top: "res4b23_branch2a"
5197     name: "bn4b23_branch2a"
5198     type: "BatchNorm"
5199     batch_norm_param {
5200         use_global_stats: false
5201     }
5202 }
5203 
5204 layer {
5205     bottom: "res4b23_branch2a"
5206     top: "res4b23_branch2a"
5207     name: "scale4b23_branch2a"
5208     type: "Scale"
5209     scale_param {
5210         bias_term: true
5211     }
5212 }
5213 
5214 layer {
5215     bottom: "res4b23_branch2a"
5216     top: "res4b23_branch2a"
5217     name: "res4b23_branch2a_relu"
5218     type: "ReLU"
5219 }
5220 
5221 layer {
5222     bottom: "res4b23_branch2a"
5223     top: "res4b23_branch2b"
5224     name: "res4b23_branch2b"
5225     type: "Convolution"
5226     convolution_param {
5227         num_output: 256
5228         kernel_size: 3
5229         pad: 1
5230         stride: 1
5231         weight_filler {
5232             type: "msra"
5233         }
5234         bias_term: false
5235 
5236     }
5237 }
5238 
5239 layer {
5240     bottom: "res4b23_branch2b"
5241     top: "res4b23_branch2b"
5242     name: "bn4b23_branch2b"
5243     type: "BatchNorm"
5244     batch_norm_param {
5245         use_global_stats: false
5246     }
5247 }
5248 
5249 layer {
5250     bottom: "res4b23_branch2b"
5251     top: "res4b23_branch2b"
5252     name: "scale4b23_branch2b"
5253     type: "Scale"
5254     scale_param {
5255         bias_term: true
5256     }
5257 }
5258 
5259 layer {
5260     bottom: "res4b23_branch2b"
5261     top: "res4b23_branch2b"
5262     name: "res4b23_branch2b_relu"
5263     type: "ReLU"
5264 }
5265 
5266 layer {
5267     bottom: "res4b23_branch2b"
5268     top: "res4b23_branch2c"
5269     name: "res4b23_branch2c"
5270     type: "Convolution"
5271     convolution_param {
5272         num_output: 1024
5273         kernel_size: 1
5274         pad: 0
5275         stride: 1
5276         weight_filler {
5277             type: "msra"
5278         }
5279         bias_term: false
5280 
5281     }
5282 }
5283 
5284 layer {
5285     bottom: "res4b23_branch2c"
5286     top: "res4b23_branch2c"
5287     name: "bn4b23_branch2c"
5288     type: "BatchNorm"
5289     batch_norm_param {
5290         use_global_stats: false
5291     }
5292 }
5293 
5294 layer {
5295     bottom: "res4b23_branch2c"
5296     top: "res4b23_branch2c"
5297     name: "scale4b23_branch2c"
5298     type: "Scale"
5299     scale_param {
5300         bias_term: true
5301     }
5302 }
5303 
5304 layer {
5305     bottom: "res4b22"
5306     bottom: "res4b23_branch2c"
5307     top: "res4b23"
5308     name: "res4b23"
5309     type: "Eltwise"
5310     eltwise_param {
5311         operation: SUM
5312     }
5313 }
5314 
5315 layer {
5316     bottom: "res4b23"
5317     top: "res4b23"
5318     name: "res4b23_relu"
5319     type: "ReLU"
5320 }
5321 
5322 layer {
5323     bottom: "res4b23"
5324     top: "res4b24_branch2a"
5325     name: "res4b24_branch2a"
5326     type: "Convolution"
5327     convolution_param {
5328         num_output: 256
5329         kernel_size: 1
5330         pad: 0
5331         stride: 1
5332         weight_filler {
5333             type: "msra"
5334         }
5335         bias_term: false
5336 
5337     }
5338 }
5339 
5340 layer {
5341     bottom: "res4b24_branch2a"
5342     top: "res4b24_branch2a"
5343     name: "bn4b24_branch2a"
5344     type: "BatchNorm"
5345     batch_norm_param {
5346         use_global_stats: false
5347     }
5348 }
5349 
5350 layer {
5351     bottom: "res4b24_branch2a"
5352     top: "res4b24_branch2a"
5353     name: "scale4b24_branch2a"
5354     type: "Scale"
5355     scale_param {
5356         bias_term: true
5357     }
5358 }
5359 
5360 layer {
5361     bottom: "res4b24_branch2a"
5362     top: "res4b24_branch2a"
5363     name: "res4b24_branch2a_relu"
5364     type: "ReLU"
5365 }
5366 
5367 layer {
5368     bottom: "res4b24_branch2a"
5369     top: "res4b24_branch2b"
5370     name: "res4b24_branch2b"
5371     type: "Convolution"
5372     convolution_param {
5373         num_output: 256
5374         kernel_size: 3
5375         pad: 1
5376         stride: 1
5377         weight_filler {
5378             type: "msra"
5379         }
5380         bias_term: false
5381 
5382     }
5383 }
5384 
5385 layer {
5386     bottom: "res4b24_branch2b"
5387     top: "res4b24_branch2b"
5388     name: "bn4b24_branch2b"
5389     type: "BatchNorm"
5390     batch_norm_param {
5391         use_global_stats: false
5392     }
5393 }
5394 
5395 layer {
5396     bottom: "res4b24_branch2b"
5397     top: "res4b24_branch2b"
5398     name: "scale4b24_branch2b"
5399     type: "Scale"
5400     scale_param {
5401         bias_term: true
5402     }
5403 }
5404 
5405 layer {
5406     bottom: "res4b24_branch2b"
5407     top: "res4b24_branch2b"
5408     name: "res4b24_branch2b_relu"
5409     type: "ReLU"
5410 }
5411 
5412 layer {
5413     bottom: "res4b24_branch2b"
5414     top: "res4b24_branch2c"
5415     name: "res4b24_branch2c"
5416     type: "Convolution"
5417     convolution_param {
5418         num_output: 1024
5419         kernel_size: 1
5420         pad: 0
5421         stride: 1
5422         weight_filler {
5423             type: "msra"
5424         }
5425         bias_term: false
5426 
5427     }
5428 }
5429 
5430 layer {
5431     bottom: "res4b24_branch2c"
5432     top: "res4b24_branch2c"
5433     name: "bn4b24_branch2c"
5434     type: "BatchNorm"
5435     batch_norm_param {
5436         use_global_stats: false
5437     }
5438 }
5439 
5440 layer {
5441     bottom: "res4b24_branch2c"
5442     top: "res4b24_branch2c"
5443     name: "scale4b24_branch2c"
5444     type: "Scale"
5445     scale_param {
5446         bias_term: true
5447     }
5448 }
5449 
5450 layer {
5451     bottom: "res4b23"
5452     bottom: "res4b24_branch2c"
5453     top: "res4b24"
5454     name: "res4b24"
5455     type: "Eltwise"
5456     eltwise_param {
5457         operation: SUM
5458     }
5459 }
5460 
5461 layer {
5462     bottom: "res4b24"
5463     top: "res4b24"
5464     name: "res4b24_relu"
5465     type: "ReLU"
5466 }
5467 
5468 layer {
5469     bottom: "res4b24"
5470     top: "res4b25_branch2a"
5471     name: "res4b25_branch2a"
5472     type: "Convolution"
5473     convolution_param {
5474         num_output: 256
5475         kernel_size: 1
5476         pad: 0
5477         stride: 1
5478         weight_filler {
5479             type: "msra"
5480         }
5481         bias_term: false
5482 
5483     }
5484 }
5485 
5486 layer {
5487     bottom: "res4b25_branch2a"
5488     top: "res4b25_branch2a"
5489     name: "bn4b25_branch2a"
5490     type: "BatchNorm"
5491     batch_norm_param {
5492         use_global_stats: false
5493     }
5494 }
5495 
5496 layer {
5497     bottom: "res4b25_branch2a"
5498     top: "res4b25_branch2a"
5499     name: "scale4b25_branch2a"
5500     type: "Scale"
5501     scale_param {
5502         bias_term: true
5503     }
5504 }
5505 
5506 layer {
5507     bottom: "res4b25_branch2a"
5508     top: "res4b25_branch2a"
5509     name: "res4b25_branch2a_relu"
5510     type: "ReLU"
5511 }
5512 
5513 layer {
5514     bottom: "res4b25_branch2a"
5515     top: "res4b25_branch2b"
5516     name: "res4b25_branch2b"
5517     type: "Convolution"
5518     convolution_param {
5519         num_output: 256
5520         kernel_size: 3
5521         pad: 1
5522         stride: 1
5523         weight_filler {
5524             type: "msra"
5525         }
5526         bias_term: false
5527 
5528     }
5529 }
5530 
5531 layer {
5532     bottom: "res4b25_branch2b"
5533     top: "res4b25_branch2b"
5534     name: "bn4b25_branch2b"
5535     type: "BatchNorm"
5536     batch_norm_param {
5537         use_global_stats: false
5538     }
5539 }
5540 
5541 layer {
5542     bottom: "res4b25_branch2b"
5543     top: "res4b25_branch2b"
5544     name: "scale4b25_branch2b"
5545     type: "Scale"
5546     scale_param {
5547         bias_term: true
5548     }
5549 }
5550 
5551 layer {
5552     bottom: "res4b25_branch2b"
5553     top: "res4b25_branch2b"
5554     name: "res4b25_branch2b_relu"
5555     type: "ReLU"
5556 }
5557 
5558 layer {
5559     bottom: "res4b25_branch2b"
5560     top: "res4b25_branch2c"
5561     name: "res4b25_branch2c"
5562     type: "Convolution"
5563     convolution_param {
5564         num_output: 1024
5565         kernel_size: 1
5566         pad: 0
5567         stride: 1
5568         weight_filler {
5569             type: "msra"
5570         }
5571         bias_term: false
5572 
5573     }
5574 }
5575 
5576 layer {
5577     bottom: "res4b25_branch2c"
5578     top: "res4b25_branch2c"
5579     name: "bn4b25_branch2c"
5580     type: "BatchNorm"
5581     batch_norm_param {
5582         use_global_stats: false
5583     }
5584 }
5585 
5586 layer {
5587     bottom: "res4b25_branch2c"
5588     top: "res4b25_branch2c"
5589     name: "scale4b25_branch2c"
5590     type: "Scale"
5591     scale_param {
5592         bias_term: true
5593     }
5594 }
5595 
5596 layer {
5597     bottom: "res4b24"
5598     bottom: "res4b25_branch2c"
5599     top: "res4b25"
5600     name: "res4b25"
5601     type: "Eltwise"
5602     eltwise_param {
5603         operation: SUM
5604     }
5605 }
5606 
5607 layer {
5608     bottom: "res4b25"
5609     top: "res4b25"
5610     name: "res4b25_relu"
5611     type: "ReLU"
5612 }
5613 
5614 layer {
5615     bottom: "res4b25"
5616     top: "res4b26_branch2a"
5617     name: "res4b26_branch2a"
5618     type: "Convolution"
5619     convolution_param {
5620         num_output: 256
5621         kernel_size: 1
5622         pad: 0
5623         stride: 1
5624         weight_filler {
5625             type: "msra"
5626         }
5627         bias_term: false
5628 
5629     }
5630 }
5631 
5632 layer {
5633     bottom: "res4b26_branch2a"
5634     top: "res4b26_branch2a"
5635     name: "bn4b26_branch2a"
5636     type: "BatchNorm"
5637     batch_norm_param {
5638         use_global_stats: false
5639     }
5640 }
5641 
5642 layer {
5643     bottom: "res4b26_branch2a"
5644     top: "res4b26_branch2a"
5645     name: "scale4b26_branch2a"
5646     type: "Scale"
5647     scale_param {
5648         bias_term: true
5649     }
5650 }
5651 
5652 layer {
5653     bottom: "res4b26_branch2a"
5654     top: "res4b26_branch2a"
5655     name: "res4b26_branch2a_relu"
5656     type: "ReLU"
5657 }
5658 
5659 layer {
5660     bottom: "res4b26_branch2a"
5661     top: "res4b26_branch2b"
5662     name: "res4b26_branch2b"
5663     type: "Convolution"
5664     convolution_param {
5665         num_output: 256
5666         kernel_size: 3
5667         pad: 1
5668         stride: 1
5669         weight_filler {
5670             type: "msra"
5671         }
5672         bias_term: false
5673 
5674     }
5675 }
5676 
5677 layer {
5678     bottom: "res4b26_branch2b"
5679     top: "res4b26_branch2b"
5680     name: "bn4b26_branch2b"
5681     type: "BatchNorm"
5682     batch_norm_param {
5683         use_global_stats: false
5684     }
5685 }
5686 
5687 layer {
5688     bottom: "res4b26_branch2b"
5689     top: "res4b26_branch2b"
5690     name: "scale4b26_branch2b"
5691     type: "Scale"
5692     scale_param {
5693         bias_term: true
5694     }
5695 }
5696 
5697 layer {
5698     bottom: "res4b26_branch2b"
5699     top: "res4b26_branch2b"
5700     name: "res4b26_branch2b_relu"
5701     type: "ReLU"
5702 }
5703 
5704 layer {
5705     bottom: "res4b26_branch2b"
5706     top: "res4b26_branch2c"
5707     name: "res4b26_branch2c"
5708     type: "Convolution"
5709     convolution_param {
5710         num_output: 1024
5711         kernel_size: 1
5712         pad: 0
5713         stride: 1
5714         weight_filler {
5715             type: "msra"
5716         }
5717         bias_term: false
5718 
5719     }
5720 }
5721 
5722 layer {
5723     bottom: "res4b26_branch2c"
5724     top: "res4b26_branch2c"
5725     name: "bn4b26_branch2c"
5726     type: "BatchNorm"
5727     batch_norm_param {
5728         use_global_stats: false
5729     }
5730 }
5731 
5732 layer {
5733     bottom: "res4b26_branch2c"
5734     top: "res4b26_branch2c"
5735     name: "scale4b26_branch2c"
5736     type: "Scale"
5737     scale_param {
5738         bias_term: true
5739     }
5740 }
5741 
5742 layer {
5743     bottom: "res4b25"
5744     bottom: "res4b26_branch2c"
5745     top: "res4b26"
5746     name: "res4b26"
5747     type: "Eltwise"
5748     eltwise_param {
5749         operation: SUM
5750     }
5751 }
5752 
5753 layer {
5754     bottom: "res4b26"
5755     top: "res4b26"
5756     name: "res4b26_relu"
5757     type: "ReLU"
5758 }
5759 
5760 layer {
5761     bottom: "res4b26"
5762     top: "res4b27_branch2a"
5763     name: "res4b27_branch2a"
5764     type: "Convolution"
5765     convolution_param {
5766         num_output: 256
5767         kernel_size: 1
5768         pad: 0
5769         stride: 1
5770         weight_filler {
5771             type: "msra"
5772         }
5773         bias_term: false
5774 
5775     }
5776 }
5777 
5778 layer {
5779     bottom: "res4b27_branch2a"
5780     top: "res4b27_branch2a"
5781     name: "bn4b27_branch2a"
5782     type: "BatchNorm"
5783     batch_norm_param {
5784         use_global_stats: false
5785     }
5786 }
5787 
5788 layer {
5789     bottom: "res4b27_branch2a"
5790     top: "res4b27_branch2a"
5791     name: "scale4b27_branch2a"
5792     type: "Scale"
5793     scale_param {
5794         bias_term: true
5795     }
5796 }
5797 
5798 layer {
5799     bottom: "res4b27_branch2a"
5800     top: "res4b27_branch2a"
5801     name: "res4b27_branch2a_relu"
5802     type: "ReLU"
5803 }
5804 
5805 layer {
5806     bottom: "res4b27_branch2a"
5807     top: "res4b27_branch2b"
5808     name: "res4b27_branch2b"
5809     type: "Convolution"
5810     convolution_param {
5811         num_output: 256
5812         kernel_size: 3
5813         pad: 1
5814         stride: 1
5815         weight_filler {
5816             type: "msra"
5817         }
5818         bias_term: false
5819 
5820     }
5821 }
5822 
5823 layer {
5824     bottom: "res4b27_branch2b"
5825     top: "res4b27_branch2b"
5826     name: "bn4b27_branch2b"
5827     type: "BatchNorm"
5828     batch_norm_param {
5829         use_global_stats: false
5830     }
5831 }
5832 
5833 layer {
5834     bottom: "res4b27_branch2b"
5835     top: "res4b27_branch2b"
5836     name: "scale4b27_branch2b"
5837     type: "Scale"
5838     scale_param {
5839         bias_term: true
5840     }
5841 }
5842 
5843 layer {
5844     bottom: "res4b27_branch2b"
5845     top: "res4b27_branch2b"
5846     name: "res4b27_branch2b_relu"
5847     type: "ReLU"
5848 }
5849 
5850 layer {
5851     bottom: "res4b27_branch2b"
5852     top: "res4b27_branch2c"
5853     name: "res4b27_branch2c"
5854     type: "Convolution"
5855     convolution_param {
5856         num_output: 1024
5857         kernel_size: 1
5858         pad: 0
5859         stride: 1
5860         weight_filler {
5861             type: "msra"
5862         }
5863         bias_term: false
5864 
5865     }
5866 }
5867 
5868 layer {
5869     bottom: "res4b27_branch2c"
5870     top: "res4b27_branch2c"
5871     name: "bn4b27_branch2c"
5872     type: "BatchNorm"
5873     batch_norm_param {
5874         use_global_stats: false
5875     }
5876 }
5877 
5878 layer {
5879     bottom: "res4b27_branch2c"
5880     top: "res4b27_branch2c"
5881     name: "scale4b27_branch2c"
5882     type: "Scale"
5883     scale_param {
5884         bias_term: true
5885     }
5886 }
5887 
5888 layer {
5889     bottom: "res4b26"
5890     bottom: "res4b27_branch2c"
5891     top: "res4b27"
5892     name: "res4b27"
5893     type: "Eltwise"
5894     eltwise_param {
5895         operation: SUM
5896     }
5897 }
5898 
5899 layer {
5900     bottom: "res4b27"
5901     top: "res4b27"
5902     name: "res4b27_relu"
5903     type: "ReLU"
5904 }
5905 
5906 layer {
5907     bottom: "res4b27"
5908     top: "res4b28_branch2a"
5909     name: "res4b28_branch2a"
5910     type: "Convolution"
5911     convolution_param {
5912         num_output: 256
5913         kernel_size: 1
5914         pad: 0
5915         stride: 1
5916         weight_filler {
5917             type: "msra"
5918         }
5919         bias_term: false
5920 
5921     }
5922 }
5923 
5924 layer {
5925     bottom: "res4b28_branch2a"
5926     top: "res4b28_branch2a"
5927     name: "bn4b28_branch2a"
5928     type: "BatchNorm"
5929     batch_norm_param {
5930         use_global_stats: false
5931     }
5932 }
5933 
5934 layer {
5935     bottom: "res4b28_branch2a"
5936     top: "res4b28_branch2a"
5937     name: "scale4b28_branch2a"
5938     type: "Scale"
5939     scale_param {
5940         bias_term: true
5941     }
5942 }
5943 
5944 layer {
5945     bottom: "res4b28_branch2a"
5946     top: "res4b28_branch2a"
5947     name: "res4b28_branch2a_relu"
5948     type: "ReLU"
5949 }
5950 
5951 layer {
5952     bottom: "res4b28_branch2a"
5953     top: "res4b28_branch2b"
5954     name: "res4b28_branch2b"
5955     type: "Convolution"
5956     convolution_param {
5957         num_output: 256
5958         kernel_size: 3
5959         pad: 1
5960         stride: 1
5961         weight_filler {
5962             type: "msra"
5963         }
5964         bias_term: false
5965 
5966     }
5967 }
5968 
5969 layer {
5970     bottom: "res4b28_branch2b"
5971     top: "res4b28_branch2b"
5972     name: "bn4b28_branch2b"
5973     type: "BatchNorm"
5974     batch_norm_param {
5975         use_global_stats: false
5976     }
5977 }
5978 
5979 layer {
5980     bottom: "res4b28_branch2b"
5981     top: "res4b28_branch2b"
5982     name: "scale4b28_branch2b"
5983     type: "Scale"
5984     scale_param {
5985         bias_term: true
5986     }
5987 }
5988 
5989 layer {
5990     bottom: "res4b28_branch2b"
5991     top: "res4b28_branch2b"
5992     name: "res4b28_branch2b_relu"
5993     type: "ReLU"
5994 }
5995 
5996 layer {
5997     bottom: "res4b28_branch2b"
5998     top: "res4b28_branch2c"
5999     name: "res4b28_branch2c"
6000     type: "Convolution"
6001     convolution_param {
6002         num_output: 1024
6003         kernel_size: 1
6004         pad: 0
6005         stride: 1
6006         weight_filler {
6007             type: "msra"
6008         }
6009         bias_term: false
6010 
6011     }
6012 }
6013 
6014 layer {
6015     bottom: "res4b28_branch2c"
6016     top: "res4b28_branch2c"
6017     name: "bn4b28_branch2c"
6018     type: "BatchNorm"
6019     batch_norm_param {
6020         use_global_stats: false
6021     }
6022 }
6023 
6024 layer {
6025     bottom: "res4b28_branch2c"
6026     top: "res4b28_branch2c"
6027     name: "scale4b28_branch2c"
6028     type: "Scale"
6029     scale_param {
6030         bias_term: true
6031     }
6032 }
6033 
6034 layer {
6035     bottom: "res4b27"
6036     bottom: "res4b28_branch2c"
6037     top: "res4b28"
6038     name: "res4b28"
6039     type: "Eltwise"
6040     eltwise_param {
6041         operation: SUM
6042     }
6043 }
6044 
6045 layer {
6046     bottom: "res4b28"
6047     top: "res4b28"
6048     name: "res4b28_relu"
6049     type: "ReLU"
6050 }
6051 
6052 layer {
6053     bottom: "res4b28"
6054     top: "res4b29_branch2a"
6055     name: "res4b29_branch2a"
6056     type: "Convolution"
6057     convolution_param {
6058         num_output: 256
6059         kernel_size: 1
6060         pad: 0
6061         stride: 1
6062         weight_filler {
6063             type: "msra"
6064         }
6065         bias_term: false
6066 
6067     }
6068 }
6069 
6070 layer {
6071     bottom: "res4b29_branch2a"
6072     top: "res4b29_branch2a"
6073     name: "bn4b29_branch2a"
6074     type: "BatchNorm"
6075     batch_norm_param {
6076         use_global_stats: false
6077     }
6078 }
6079 
6080 layer {
6081     bottom: "res4b29_branch2a"
6082     top: "res4b29_branch2a"
6083     name: "scale4b29_branch2a"
6084     type: "Scale"
6085     scale_param {
6086         bias_term: true
6087     }
6088 }
6089 
6090 layer {
6091     bottom: "res4b29_branch2a"
6092     top: "res4b29_branch2a"
6093     name: "res4b29_branch2a_relu"
6094     type: "ReLU"
6095 }
6096 
6097 layer {
6098     bottom: "res4b29_branch2a"
6099     top: "res4b29_branch2b"
6100     name: "res4b29_branch2b"
6101     type: "Convolution"
6102     convolution_param {
6103         num_output: 256
6104         kernel_size: 3
6105         pad: 1
6106         stride: 1
6107         weight_filler {
6108             type: "msra"
6109         }
6110         bias_term: false
6111 
6112     }
6113 }
6114 
6115 layer {
6116     bottom: "res4b29_branch2b"
6117     top: "res4b29_branch2b"
6118     name: "bn4b29_branch2b"
6119     type: "BatchNorm"
6120     batch_norm_param {
6121         use_global_stats: false
6122     }
6123 }
6124 
6125 layer {
6126     bottom: "res4b29_branch2b"
6127     top: "res4b29_branch2b"
6128     name: "scale4b29_branch2b"
6129     type: "Scale"
6130     scale_param {
6131         bias_term: true
6132     }
6133 }
6134 
6135 layer {
6136     bottom: "res4b29_branch2b"
6137     top: "res4b29_branch2b"
6138     name: "res4b29_branch2b_relu"
6139     type: "ReLU"
6140 }
6141 
6142 layer {
6143     bottom: "res4b29_branch2b"
6144     top: "res4b29_branch2c"
6145     name: "res4b29_branch2c"
6146     type: "Convolution"
6147     convolution_param {
6148         num_output: 1024
6149         kernel_size: 1
6150         pad: 0
6151         stride: 1
6152         weight_filler {
6153             type: "msra"
6154         }
6155         bias_term: false
6156 
6157     }
6158 }
6159 
6160 layer {
6161     bottom: "res4b29_branch2c"
6162     top: "res4b29_branch2c"
6163     name: "bn4b29_branch2c"
6164     type: "BatchNorm"
6165     batch_norm_param {
6166         use_global_stats: false
6167     }
6168 }
6169 
6170 layer {
6171     bottom: "res4b29_branch2c"
6172     top: "res4b29_branch2c"
6173     name: "scale4b29_branch2c"
6174     type: "Scale"
6175     scale_param {
6176         bias_term: true
6177     }
6178 }
6179 
6180 layer {
6181     bottom: "res4b28"
6182     bottom: "res4b29_branch2c"
6183     top: "res4b29"
6184     name: "res4b29"
6185     type: "Eltwise"
6186     eltwise_param {
6187         operation: SUM
6188     }
6189 }
6190 
6191 layer {
6192     bottom: "res4b29"
6193     top: "res4b29"
6194     name: "res4b29_relu"
6195     type: "ReLU"
6196 }
6197 
6198 layer {
6199     bottom: "res4b29"
6200     top: "res4b30_branch2a"
6201     name: "res4b30_branch2a"
6202     type: "Convolution"
6203     convolution_param {
6204         num_output: 256
6205         kernel_size: 1
6206         pad: 0
6207         stride: 1
6208         weight_filler {
6209             type: "msra"
6210         }
6211         bias_term: false
6212 
6213     }
6214 }
6215 
6216 layer {
6217     bottom: "res4b30_branch2a"
6218     top: "res4b30_branch2a"
6219     name: "bn4b30_branch2a"
6220     type: "BatchNorm"
6221     batch_norm_param {
6222         use_global_stats: false
6223     }
6224 }
6225 
6226 layer {
6227     bottom: "res4b30_branch2a"
6228     top: "res4b30_branch2a"
6229     name: "scale4b30_branch2a"
6230     type: "Scale"
6231     scale_param {
6232         bias_term: true
6233     }
6234 }
6235 
6236 layer {
6237     bottom: "res4b30_branch2a"
6238     top: "res4b30_branch2a"
6239     name: "res4b30_branch2a_relu"
6240     type: "ReLU"
6241 }
6242 
6243 layer {
6244     bottom: "res4b30_branch2a"
6245     top: "res4b30_branch2b"
6246     name: "res4b30_branch2b"
6247     type: "Convolution"
6248     convolution_param {
6249         num_output: 256
6250         kernel_size: 3
6251         pad: 1
6252         stride: 1
6253         weight_filler {
6254             type: "msra"
6255         }
6256         bias_term: false
6257 
6258     }
6259 }
6260 
6261 layer {
6262     bottom: "res4b30_branch2b"
6263     top: "res4b30_branch2b"
6264     name: "bn4b30_branch2b"
6265     type: "BatchNorm"
6266     batch_norm_param {
6267         use_global_stats: false
6268     }
6269 }
6270 
6271 layer {
6272     bottom: "res4b30_branch2b"
6273     top: "res4b30_branch2b"
6274     name: "scale4b30_branch2b"
6275     type: "Scale"
6276     scale_param {
6277         bias_term: true
6278     }
6279 }
6280 
6281 layer {
6282     bottom: "res4b30_branch2b"
6283     top: "res4b30_branch2b"
6284     name: "res4b30_branch2b_relu"
6285     type: "ReLU"
6286 }
6287 
6288 layer {
6289     bottom: "res4b30_branch2b"
6290     top: "res4b30_branch2c"
6291     name: "res4b30_branch2c"
6292     type: "Convolution"
6293     convolution_param {
6294         num_output: 1024
6295         kernel_size: 1
6296         pad: 0
6297         stride: 1
6298         weight_filler {
6299             type: "msra"
6300         }
6301         bias_term: false
6302 
6303     }
6304 }
6305 
6306 layer {
6307     bottom: "res4b30_branch2c"
6308     top: "res4b30_branch2c"
6309     name: "bn4b30_branch2c"
6310     type: "BatchNorm"
6311     batch_norm_param {
6312         use_global_stats: false
6313     }
6314 }
6315 
6316 layer {
6317     bottom: "res4b30_branch2c"
6318     top: "res4b30_branch2c"
6319     name: "scale4b30_branch2c"
6320     type: "Scale"
6321     scale_param {
6322         bias_term: true
6323     }
6324 }
6325 
6326 layer {
6327     bottom: "res4b29"
6328     bottom: "res4b30_branch2c"
6329     top: "res4b30"
6330     name: "res4b30"
6331     type: "Eltwise"
6332     eltwise_param {
6333         operation: SUM
6334     }
6335 }
6336 
6337 layer {
6338     bottom: "res4b30"
6339     top: "res4b30"
6340     name: "res4b30_relu"
6341     type: "ReLU"
6342 }
6343 
6344 layer {
6345     bottom: "res4b30"
6346     top: "res4b31_branch2a"
6347     name: "res4b31_branch2a"
6348     type: "Convolution"
6349     convolution_param {
6350         num_output: 256
6351         kernel_size: 1
6352         pad: 0
6353         stride: 1
6354         weight_filler {
6355             type: "msra"
6356         }
6357         bias_term: false
6358 
6359     }
6360 }
6361 
6362 layer {
6363     bottom: "res4b31_branch2a"
6364     top: "res4b31_branch2a"
6365     name: "bn4b31_branch2a"
6366     type: "BatchNorm"
6367     batch_norm_param {
6368         use_global_stats: false
6369     }
6370 }
6371 
6372 layer {
6373     bottom: "res4b31_branch2a"
6374     top: "res4b31_branch2a"
6375     name: "scale4b31_branch2a"
6376     type: "Scale"
6377     scale_param {
6378         bias_term: true
6379     }
6380 }
6381 
6382 layer {
6383     bottom: "res4b31_branch2a"
6384     top: "res4b31_branch2a"
6385     name: "res4b31_branch2a_relu"
6386     type: "ReLU"
6387 }
6388 
6389 layer {
6390     bottom: "res4b31_branch2a"
6391     top: "res4b31_branch2b"
6392     name: "res4b31_branch2b"
6393     type: "Convolution"
6394     convolution_param {
6395         num_output: 256
6396         kernel_size: 3
6397         pad: 1
6398         stride: 1
6399         weight_filler {
6400             type: "msra"
6401         }
6402         bias_term: false
6403 
6404     }
6405 }
6406 
6407 layer {
6408     bottom: "res4b31_branch2b"
6409     top: "res4b31_branch2b"
6410     name: "bn4b31_branch2b"
6411     type: "BatchNorm"
6412     batch_norm_param {
6413         use_global_stats: false
6414     }
6415 }
6416 
6417 layer {
6418     bottom: "res4b31_branch2b"
6419     top: "res4b31_branch2b"
6420     name: "scale4b31_branch2b"
6421     type: "Scale"
6422     scale_param {
6423         bias_term: true
6424     }
6425 }
6426 
6427 layer {
6428     bottom: "res4b31_branch2b"
6429     top: "res4b31_branch2b"
6430     name: "res4b31_branch2b_relu"
6431     type: "ReLU"
6432 }
6433 
6434 layer {
6435     bottom: "res4b31_branch2b"
6436     top: "res4b31_branch2c"
6437     name: "res4b31_branch2c"
6438     type: "Convolution"
6439     convolution_param {
6440         num_output: 1024
6441         kernel_size: 1
6442         pad: 0
6443         stride: 1
6444         weight_filler {
6445             type: "msra"
6446         }
6447         bias_term: false
6448 
6449     }
6450 }
6451 
6452 layer {
6453     bottom: "res4b31_branch2c"
6454     top: "res4b31_branch2c"
6455     name: "bn4b31_branch2c"
6456     type: "BatchNorm"
6457     batch_norm_param {
6458         use_global_stats: false
6459     }
6460 }
6461 
6462 layer {
6463     bottom: "res4b31_branch2c"
6464     top: "res4b31_branch2c"
6465     name: "scale4b31_branch2c"
6466     type: "Scale"
6467     scale_param {
6468         bias_term: true
6469     }
6470 }
6471 
6472 layer {
6473     bottom: "res4b30"
6474     bottom: "res4b31_branch2c"
6475     top: "res4b31"
6476     name: "res4b31"
6477     type: "Eltwise"
6478     eltwise_param {
6479         operation: SUM
6480     }
6481 }
6482 
6483 layer {
6484     bottom: "res4b31"
6485     top: "res4b31"
6486     name: "res4b31_relu"
6487     type: "ReLU"
6488 }
6489 
6490 layer {
6491     bottom: "res4b31"
6492     top: "res4b32_branch2a"
6493     name: "res4b32_branch2a"
6494     type: "Convolution"
6495     convolution_param {
6496         num_output: 256
6497         kernel_size: 1
6498         pad: 0
6499         stride: 1
6500         weight_filler {
6501             type: "msra"
6502         }
6503         bias_term: false
6504 
6505     }
6506 }
6507 
6508 layer {
6509     bottom: "res4b32_branch2a"
6510     top: "res4b32_branch2a"
6511     name: "bn4b32_branch2a"
6512     type: "BatchNorm"
6513     batch_norm_param {
6514         use_global_stats: false
6515     }
6516 }
6517 
6518 layer {
6519     bottom: "res4b32_branch2a"
6520     top: "res4b32_branch2a"
6521     name: "scale4b32_branch2a"
6522     type: "Scale"
6523     scale_param {
6524         bias_term: true
6525     }
6526 }
6527 
6528 layer {
6529     bottom: "res4b32_branch2a"
6530     top: "res4b32_branch2a"
6531     name: "res4b32_branch2a_relu"
6532     type: "ReLU"
6533 }
6534 
6535 layer {
6536     bottom: "res4b32_branch2a"
6537     top: "res4b32_branch2b"
6538     name: "res4b32_branch2b"
6539     type: "Convolution"
6540     convolution_param {
6541         num_output: 256
6542         kernel_size: 3
6543         pad: 1
6544         stride: 1
6545         weight_filler {
6546             type: "msra"
6547         }
6548         bias_term: false
6549 
6550     }
6551 }
6552 
6553 layer {
6554     bottom: "res4b32_branch2b"
6555     top: "res4b32_branch2b"
6556     name: "bn4b32_branch2b"
6557     type: "BatchNorm"
6558     batch_norm_param {
6559         use_global_stats: false
6560     }
6561 }
6562 
6563 layer {
6564     bottom: "res4b32_branch2b"
6565     top: "res4b32_branch2b"
6566     name: "scale4b32_branch2b"
6567     type: "Scale"
6568     scale_param {
6569         bias_term: true
6570     }
6571 }
6572 
6573 layer {
6574     bottom: "res4b32_branch2b"
6575     top: "res4b32_branch2b"
6576     name: "res4b32_branch2b_relu"
6577     type: "ReLU"
6578 }
6579 
6580 layer {
6581     bottom: "res4b32_branch2b"
6582     top: "res4b32_branch2c"
6583     name: "res4b32_branch2c"
6584     type: "Convolution"
6585     convolution_param {
6586         num_output: 1024
6587         kernel_size: 1
6588         pad: 0
6589         stride: 1
6590         weight_filler {
6591             type: "msra"
6592         }
6593         bias_term: false
6594 
6595     }
6596 }
6597 
6598 layer {
6599     bottom: "res4b32_branch2c"
6600     top: "res4b32_branch2c"
6601     name: "bn4b32_branch2c"
6602     type: "BatchNorm"
6603     batch_norm_param {
6604         use_global_stats: false
6605     }
6606 }
6607 
6608 layer {
6609     bottom: "res4b32_branch2c"
6610     top: "res4b32_branch2c"
6611     name: "scale4b32_branch2c"
6612     type: "Scale"
6613     scale_param {
6614         bias_term: true
6615     }
6616 }
6617 
6618 layer {
6619     bottom: "res4b31"
6620     bottom: "res4b32_branch2c"
6621     top: "res4b32"
6622     name: "res4b32"
6623     type: "Eltwise"
6624     eltwise_param {
6625         operation: SUM
6626     }
6627 }
6628 
6629 layer {
6630     bottom: "res4b32"
6631     top: "res4b32"
6632     name: "res4b32_relu"
6633     type: "ReLU"
6634 }
6635 
6636 layer {
6637     bottom: "res4b32"
6638     top: "res4b33_branch2a"
6639     name: "res4b33_branch2a"
6640     type: "Convolution"
6641     convolution_param {
6642         num_output: 256
6643         kernel_size: 1
6644         pad: 0
6645         stride: 1
6646         weight_filler {
6647             type: "msra"
6648         }
6649         bias_term: false
6650 
6651     }
6652 }
6653 
6654 layer {
6655     bottom: "res4b33_branch2a"
6656     top: "res4b33_branch2a"
6657     name: "bn4b33_branch2a"
6658     type: "BatchNorm"
6659     batch_norm_param {
6660         use_global_stats: false
6661     }
6662 }
6663 
6664 layer {
6665     bottom: "res4b33_branch2a"
6666     top: "res4b33_branch2a"
6667     name: "scale4b33_branch2a"
6668     type: "Scale"
6669     scale_param {
6670         bias_term: true
6671     }
6672 }
6673 
6674 layer {
6675     bottom: "res4b33_branch2a"
6676     top: "res4b33_branch2a"
6677     name: "res4b33_branch2a_relu"
6678     type: "ReLU"
6679 }
6680 
6681 layer {
6682     bottom: "res4b33_branch2a"
6683     top: "res4b33_branch2b"
6684     name: "res4b33_branch2b"
6685     type: "Convolution"
6686     convolution_param {
6687         num_output: 256
6688         kernel_size: 3
6689         pad: 1
6690         stride: 1
6691         weight_filler {
6692             type: "msra"
6693         }
6694         bias_term: false
6695 
6696     }
6697 }
6698 
6699 layer {
6700     bottom: "res4b33_branch2b"
6701     top: "res4b33_branch2b"
6702     name: "bn4b33_branch2b"
6703     type: "BatchNorm"
6704     batch_norm_param {
6705         use_global_stats: false
6706     }
6707 }
6708 
6709 layer {
6710     bottom: "res4b33_branch2b"
6711     top: "res4b33_branch2b"
6712     name: "scale4b33_branch2b"
6713     type: "Scale"
6714     scale_param {
6715         bias_term: true
6716     }
6717 }
6718 
6719 layer {
6720     bottom: "res4b33_branch2b"
6721     top: "res4b33_branch2b"
6722     name: "res4b33_branch2b_relu"
6723     type: "ReLU"
6724 }
6725 
6726 layer {
6727     bottom: "res4b33_branch2b"
6728     top: "res4b33_branch2c"
6729     name: "res4b33_branch2c"
6730     type: "Convolution"
6731     convolution_param {
6732         num_output: 1024
6733         kernel_size: 1
6734         pad: 0
6735         stride: 1
6736         weight_filler {
6737             type: "msra"
6738         }
6739         bias_term: false
6740 
6741     }
6742 }
6743 
6744 layer {
6745     bottom: "res4b33_branch2c"
6746     top: "res4b33_branch2c"
6747     name: "bn4b33_branch2c"
6748     type: "BatchNorm"
6749     batch_norm_param {
6750         use_global_stats: false
6751     }
6752 }
6753 
6754 layer {
6755     bottom: "res4b33_branch2c"
6756     top: "res4b33_branch2c"
6757     name: "scale4b33_branch2c"
6758     type: "Scale"
6759     scale_param {
6760         bias_term: true
6761     }
6762 }
6763 
6764 layer {
6765     bottom: "res4b32"
6766     bottom: "res4b33_branch2c"
6767     top: "res4b33"
6768     name: "res4b33"
6769     type: "Eltwise"
6770     eltwise_param {
6771         operation: SUM
6772     }
6773 }
6774 
6775 layer {
6776     bottom: "res4b33"
6777     top: "res4b33"
6778     name: "res4b33_relu"
6779     type: "ReLU"
6780 }
6781 
6782 layer {
6783     bottom: "res4b33"
6784     top: "res4b34_branch2a"
6785     name: "res4b34_branch2a"
6786     type: "Convolution"
6787     convolution_param {
6788         num_output: 256
6789         kernel_size: 1
6790         pad: 0
6791         stride: 1
6792         weight_filler {
6793             type: "msra"
6794         }
6795         bias_term: false
6796 
6797     }
6798 }
6799 
6800 layer {
6801     bottom: "res4b34_branch2a"
6802     top: "res4b34_branch2a"
6803     name: "bn4b34_branch2a"
6804     type: "BatchNorm"
6805     batch_norm_param {
6806         use_global_stats: false
6807     }
6808 }
6809 
6810 layer {
6811     bottom: "res4b34_branch2a"
6812     top: "res4b34_branch2a"
6813     name: "scale4b34_branch2a"
6814     type: "Scale"
6815     scale_param {
6816         bias_term: true
6817     }
6818 }
6819 
6820 layer {
6821     bottom: "res4b34_branch2a"
6822     top: "res4b34_branch2a"
6823     name: "res4b34_branch2a_relu"
6824     type: "ReLU"
6825 }
6826 
6827 layer {
6828     bottom: "res4b34_branch2a"
6829     top: "res4b34_branch2b"
6830     name: "res4b34_branch2b"
6831     type: "Convolution"
6832     convolution_param {
6833         num_output: 256
6834         kernel_size: 3
6835         pad: 1
6836         stride: 1
6837         weight_filler {
6838             type: "msra"
6839         }
6840         bias_term: false
6841 
6842     }
6843 }
6844 
6845 layer {
6846     bottom: "res4b34_branch2b"
6847     top: "res4b34_branch2b"
6848     name: "bn4b34_branch2b"
6849     type: "BatchNorm"
6850     batch_norm_param {
6851         use_global_stats: false
6852     }
6853 }
6854 
6855 layer {
6856     bottom: "res4b34_branch2b"
6857     top: "res4b34_branch2b"
6858     name: "scale4b34_branch2b"
6859     type: "Scale"
6860     scale_param {
6861         bias_term: true
6862     }
6863 }
6864 
6865 layer {
6866     bottom: "res4b34_branch2b"
6867     top: "res4b34_branch2b"
6868     name: "res4b34_branch2b_relu"
6869     type: "ReLU"
6870 }
6871 
6872 layer {
6873     bottom: "res4b34_branch2b"
6874     top: "res4b34_branch2c"
6875     name: "res4b34_branch2c"
6876     type: "Convolution"
6877     convolution_param {
6878         num_output: 1024
6879         kernel_size: 1
6880         pad: 0
6881         stride: 1
6882         weight_filler {
6883             type: "msra"
6884         }
6885         bias_term: false
6886 
6887     }
6888 }
6889 
6890 layer {
6891     bottom: "res4b34_branch2c"
6892     top: "res4b34_branch2c"
6893     name: "bn4b34_branch2c"
6894     type: "BatchNorm"
6895     batch_norm_param {
6896         use_global_stats: false
6897     }
6898 }
6899 
6900 layer {
6901     bottom: "res4b34_branch2c"
6902     top: "res4b34_branch2c"
6903     name: "scale4b34_branch2c"
6904     type: "Scale"
6905     scale_param {
6906         bias_term: true
6907     }
6908 }
6909 
6910 layer {
6911     bottom: "res4b33"
6912     bottom: "res4b34_branch2c"
6913     top: "res4b34"
6914     name: "res4b34"
6915     type: "Eltwise"
6916     eltwise_param {
6917         operation: SUM
6918     }
6919 }
6920 
6921 layer {
6922     bottom: "res4b34"
6923     top: "res4b34"
6924     name: "res4b34_relu"
6925     type: "ReLU"
6926 }
6927 
6928 layer {
6929     bottom: "res4b34"
6930     top: "res4b35_branch2a"
6931     name: "res4b35_branch2a"
6932     type: "Convolution"
6933     convolution_param {
6934         num_output: 256
6935         kernel_size: 1
6936         pad: 0
6937         stride: 1
6938         weight_filler {
6939             type: "msra"
6940         }
6941         bias_term: false
6942 
6943     }
6944 }
6945 
6946 layer {
6947     bottom: "res4b35_branch2a"
6948     top: "res4b35_branch2a"
6949     name: "bn4b35_branch2a"
6950     type: "BatchNorm"
6951     batch_norm_param {
6952         use_global_stats: false
6953     }
6954 }
6955 
6956 layer {
6957     bottom: "res4b35_branch2a"
6958     top: "res4b35_branch2a"
6959     name: "scale4b35_branch2a"
6960     type: "Scale"
6961     scale_param {
6962         bias_term: true
6963     }
6964 }
6965 
6966 layer {
6967     bottom: "res4b35_branch2a"
6968     top: "res4b35_branch2a"
6969     name: "res4b35_branch2a_relu"
6970     type: "ReLU"
6971 }
6972 
6973 layer {
6974     bottom: "res4b35_branch2a"
6975     top: "res4b35_branch2b"
6976     name: "res4b35_branch2b"
6977     type: "Convolution"
6978     convolution_param {
6979         num_output: 256
6980         kernel_size: 3
6981         pad: 1
6982         stride: 1
6983         weight_filler {
6984             type: "msra"
6985         }
6986         bias_term: false
6987 
6988     }
6989 }
6990 
6991 layer {
6992     bottom: "res4b35_branch2b"
6993     top: "res4b35_branch2b"
6994     name: "bn4b35_branch2b"
6995     type: "BatchNorm"
6996     batch_norm_param {
6997         use_global_stats: false
6998     }
6999 }
7000 
7001 layer {
7002     bottom: "res4b35_branch2b"
7003     top: "res4b35_branch2b"
7004     name: "scale4b35_branch2b"
7005     type: "Scale"
7006     scale_param {
7007         bias_term: true
7008     }
7009 }
7010 
7011 layer {
7012     bottom: "res4b35_branch2b"
7013     top: "res4b35_branch2b"
7014     name: "res4b35_branch2b_relu"
7015     type: "ReLU"
7016 }
7017 
7018 layer {
7019     bottom: "res4b35_branch2b"
7020     top: "res4b35_branch2c"
7021     name: "res4b35_branch2c"
7022     type: "Convolution"
7023     convolution_param {
7024         num_output: 1024
7025         kernel_size: 1
7026         pad: 0
7027         stride: 1
7028         weight_filler {
7029             type: "msra"
7030         }
7031         bias_term: false
7032 
7033     }
7034 }
7035 
7036 layer {
7037     bottom: "res4b35_branch2c"
7038     top: "res4b35_branch2c"
7039     name: "bn4b35_branch2c"
7040     type: "BatchNorm"
7041     batch_norm_param {
7042         use_global_stats: false
7043     }
7044 }
7045 
7046 layer {
7047     bottom: "res4b35_branch2c"
7048     top: "res4b35_branch2c"
7049     name: "scale4b35_branch2c"
7050     type: "Scale"
7051     scale_param {
7052         bias_term: true
7053     }
7054 }
7055 
7056 layer {
7057     bottom: "res4b34"
7058     bottom: "res4b35_branch2c"
7059     top: "res4b35"
7060     name: "res4b35"
7061     type: "Eltwise"
7062     eltwise_param {
7063         operation: SUM
7064     }
7065 }
7066 
7067 layer {
7068     bottom: "res4b35"
7069     top: "res4b35"
7070     name: "res4b35_relu"
7071     type: "ReLU"
7072 }
7073 
7074 layer {
7075     bottom: "res4b35"
7076     top: "res5a_branch1"
7077     name: "res5a_branch1"
7078     type: "Convolution"
7079     convolution_param {
7080         num_output: 2048
7081         kernel_size: 1
7082         pad: 0
7083         stride: 2
7084         weight_filler {
7085             type: "msra"
7086         }
7087         bias_term: false
7088 
7089     }
7090 }
7091 
7092 layer {
7093     bottom: "res5a_branch1"
7094     top: "res5a_branch1"
7095     name: "bn5a_branch1"
7096     type: "BatchNorm"
7097     batch_norm_param {
7098         use_global_stats: false
7099     }
7100 }
7101 
7102 layer {
7103     bottom: "res5a_branch1"
7104     top: "res5a_branch1"
7105     name: "scale5a_branch1"
7106     type: "Scale"
7107     scale_param {
7108         bias_term: true
7109     }
7110 }
7111 
7112 layer {
7113     bottom: "res4b35"
7114     top: "res5a_branch2a"
7115     name: "res5a_branch2a"
7116     type: "Convolution"
7117     convolution_param {
7118         num_output: 512
7119         kernel_size: 1
7120         pad: 0
7121         stride: 2
7122         weight_filler {
7123             type: "msra"
7124         }
7125         bias_term: false
7126 
7127     }
7128 }
7129 
7130 layer {
7131     bottom: "res5a_branch2a"
7132     top: "res5a_branch2a"
7133     name: "bn5a_branch2a"
7134     type: "BatchNorm"
7135     batch_norm_param {
7136         use_global_stats: false
7137     }
7138 }
7139 
7140 layer {
7141     bottom: "res5a_branch2a"
7142     top: "res5a_branch2a"
7143     name: "scale5a_branch2a"
7144     type: "Scale"
7145     scale_param {
7146         bias_term: true
7147     }
7148 }
7149 
7150 layer {
7151     bottom: "res5a_branch2a"
7152     top: "res5a_branch2a"
7153     name: "res5a_branch2a_relu"
7154     type: "ReLU"
7155 }
7156 
7157 layer {
7158     bottom: "res5a_branch2a"
7159     top: "res5a_branch2b"
7160     name: "res5a_branch2b"
7161     type: "Convolution"
7162     convolution_param {
7163         num_output: 512
7164         kernel_size: 3
7165         pad: 1
7166         stride: 1
7167         weight_filler {
7168             type: "msra"
7169         }
7170         bias_term: false
7171 
7172     }
7173 }
7174 
7175 layer {
7176     bottom: "res5a_branch2b"
7177     top: "res5a_branch2b"
7178     name: "bn5a_branch2b"
7179     type: "BatchNorm"
7180     batch_norm_param {
7181         use_global_stats: false
7182     }
7183 }
7184 
7185 layer {
7186     bottom: "res5a_branch2b"
7187     top: "res5a_branch2b"
7188     name: "scale5a_branch2b"
7189     type: "Scale"
7190     scale_param {
7191         bias_term: true
7192     }
7193 }
7194 
7195 layer {
7196     bottom: "res5a_branch2b"
7197     top: "res5a_branch2b"
7198     name: "res5a_branch2b_relu"
7199     type: "ReLU"
7200 }
7201 
7202 layer {
7203     bottom: "res5a_branch2b"
7204     top: "res5a_branch2c"
7205     name: "res5a_branch2c"
7206     type: "Convolution"
7207     convolution_param {
7208         num_output: 2048
7209         kernel_size: 1
7210         pad: 0
7211         stride: 1
7212         weight_filler {
7213             type: "msra"
7214         }
7215         bias_term: false
7216 
7217     }
7218 }
7219 
7220 layer {
7221     bottom: "res5a_branch2c"
7222     top: "res5a_branch2c"
7223     name: "bn5a_branch2c"
7224     type: "BatchNorm"
7225     batch_norm_param {
7226         use_global_stats: false
7227     }
7228 }
7229 
7230 layer {
7231     bottom: "res5a_branch2c"
7232     top: "res5a_branch2c"
7233     name: "scale5a_branch2c"
7234     type: "Scale"
7235     scale_param {
7236         bias_term: true
7237     }
7238 }
7239 
7240 layer {
7241     bottom: "res5a_branch1"
7242     bottom: "res5a_branch2c"
7243     top: "res5a"
7244     name: "res5a"
7245     type: "Eltwise"
7246     eltwise_param {
7247         operation: SUM
7248     }
7249 }
7250 
7251 layer {
7252     bottom: "res5a"
7253     top: "res5a"
7254     name: "res5a_relu"
7255     type: "ReLU"
7256 }
7257 
7258 layer {
7259     bottom: "res5a"
7260     top: "res5b_branch2a"
7261     name: "res5b_branch2a"
7262     type: "Convolution"
7263     convolution_param {
7264         num_output: 512
7265         kernel_size: 1
7266         pad: 0
7267         stride: 1
7268         weight_filler {
7269             type: "msra"
7270         }
7271         bias_term: false
7272 
7273     }
7274 }
7275 
7276 layer {
7277     bottom: "res5b_branch2a"
7278     top: "res5b_branch2a"
7279     name: "bn5b_branch2a"
7280     type: "BatchNorm"
7281     batch_norm_param {
7282         use_global_stats: false
7283     }
7284 }
7285 
7286 layer {
7287     bottom: "res5b_branch2a"
7288     top: "res5b_branch2a"
7289     name: "scale5b_branch2a"
7290     type: "Scale"
7291     scale_param {
7292         bias_term: true
7293     }
7294 }
7295 
7296 layer {
7297     bottom: "res5b_branch2a"
7298     top: "res5b_branch2a"
7299     name: "res5b_branch2a_relu"
7300     type: "ReLU"
7301 }
7302 
7303 layer {
7304     bottom: "res5b_branch2a"
7305     top: "res5b_branch2b"
7306     name: "res5b_branch2b"
7307     type: "Convolution"
7308     convolution_param {
7309         num_output: 512
7310         kernel_size: 3
7311         pad: 1
7312         stride: 1
7313         weight_filler {
7314             type: "msra"
7315         }
7316         bias_term: false
7317 
7318     }
7319 }
7320 
7321 layer {
7322     bottom: "res5b_branch2b"
7323     top: "res5b_branch2b"
7324     name: "bn5b_branch2b"
7325     type: "BatchNorm"
7326     batch_norm_param {
7327         use_global_stats: false
7328     }
7329 }
7330 
7331 layer {
7332     bottom: "res5b_branch2b"
7333     top: "res5b_branch2b"
7334     name: "scale5b_branch2b"
7335     type: "Scale"
7336     scale_param {
7337         bias_term: true
7338     }
7339 }
7340 
7341 layer {
7342     bottom: "res5b_branch2b"
7343     top: "res5b_branch2b"
7344     name: "res5b_branch2b_relu"
7345     type: "ReLU"
7346 }
7347 
7348 layer {
7349     bottom: "res5b_branch2b"
7350     top: "res5b_branch2c"
7351     name: "res5b_branch2c"
7352     type: "Convolution"
7353     convolution_param {
7354         num_output: 2048
7355         kernel_size: 1
7356         pad: 0
7357         stride: 1
7358         weight_filler {
7359             type: "msra"
7360         }
7361         bias_term: false
7362 
7363     }
7364 }
7365 
7366 layer {
7367     bottom: "res5b_branch2c"
7368     top: "res5b_branch2c"
7369     name: "bn5b_branch2c"
7370     type: "BatchNorm"
7371     batch_norm_param {
7372         use_global_stats: false
7373     }
7374 }
7375 
7376 layer {
7377     bottom: "res5b_branch2c"
7378     top: "res5b_branch2c"
7379     name: "scale5b_branch2c"
7380     type: "Scale"
7381     scale_param {
7382         bias_term: true
7383     }
7384 }
7385 
7386 layer {
7387     bottom: "res5a"
7388     bottom: "res5b_branch2c"
7389     top: "res5b"
7390     name: "res5b"
7391     type: "Eltwise"
7392     eltwise_param {
7393         operation: SUM
7394     }
7395 }
7396 
7397 layer {
7398     bottom: "res5b"
7399     top: "res5b"
7400     name: "res5b_relu"
7401     type: "ReLU"
7402 }
7403 
7404 layer {
7405     bottom: "res5b"
7406     top: "res5c_branch2a"
7407     name: "res5c_branch2a"
7408     type: "Convolution"
7409     convolution_param {
7410         num_output: 512
7411         kernel_size: 1
7412         pad: 0
7413         stride: 1
7414         weight_filler {
7415             type: "msra"
7416         }
7417         bias_term: false
7418 
7419     }
7420 }
7421 
7422 layer {
7423     bottom: "res5c_branch2a"
7424     top: "res5c_branch2a"
7425     name: "bn5c_branch2a"
7426     type: "BatchNorm"
7427     batch_norm_param {
7428         use_global_stats: false
7429     }
7430 }
7431 
7432 layer {
7433     bottom: "res5c_branch2a"
7434     top: "res5c_branch2a"
7435     name: "scale5c_branch2a"
7436     type: "Scale"
7437     scale_param {
7438         bias_term: true
7439     }
7440 }
7441 
7442 layer {
7443     bottom: "res5c_branch2a"
7444     top: "res5c_branch2a"
7445     name: "res5c_branch2a_relu"
7446     type: "ReLU"
7447 }
7448 
7449 layer {
7450     bottom: "res5c_branch2a"
7451     top: "res5c_branch2b"
7452     name: "res5c_branch2b"
7453     type: "Convolution"
7454     convolution_param {
7455         num_output: 512
7456         kernel_size: 3
7457         pad: 1
7458         stride: 1
7459         weight_filler {
7460             type: "msra"
7461         }
7462         bias_term: false
7463 
7464     }
7465 }
7466 
7467 layer {
7468     bottom: "res5c_branch2b"
7469     top: "res5c_branch2b"
7470     name: "bn5c_branch2b"
7471     type: "BatchNorm"
7472     batch_norm_param {
7473         use_global_stats: false
7474     }
7475 }
7476 
7477 layer {
7478     bottom: "res5c_branch2b"
7479     top: "res5c_branch2b"
7480     name: "scale5c_branch2b"
7481     type: "Scale"
7482     scale_param {
7483         bias_term: true
7484     }
7485 }
7486 
7487 layer {
7488     bottom: "res5c_branch2b"
7489     top: "res5c_branch2b"
7490     name: "res5c_branch2b_relu"
7491     type: "ReLU"
7492 }
7493 
7494 layer {
7495     bottom: "res5c_branch2b"
7496     top: "res5c_branch2c"
7497     name: "res5c_branch2c"
7498     type: "Convolution"
7499     convolution_param {
7500         num_output: 2048
7501         kernel_size: 1
7502         pad: 0
7503         stride: 1
7504         weight_filler {
7505             type: "msra"
7506         }
7507         bias_term: false
7508 
7509     }
7510 }
7511 
7512 layer {
7513     bottom: "res5c_branch2c"
7514     top: "res5c_branch2c"
7515     name: "bn5c_branch2c"
7516     type: "BatchNorm"
7517     batch_norm_param {
7518         use_global_stats: false
7519     }
7520 }
7521 
7522 layer {
7523     bottom: "res5c_branch2c"
7524     top: "res5c_branch2c"
7525     name: "scale5c_branch2c"
7526     type: "Scale"
7527     scale_param {
7528         bias_term: true
7529     }
7530 }
7531 
7532 layer {
7533     bottom: "res5b"
7534     bottom: "res5c_branch2c"
7535     top: "res5c"
7536     name: "res5c"
7537     type: "Eltwise"
7538     eltwise_param {
7539         operation: SUM
7540     }
7541 }
7542 
7543 layer {
7544     bottom: "res5c"
7545     top: "res5c"
7546     name: "res5c_relu"
7547     type: "ReLU"
7548 }
7549 
7550 layer {
7551     bottom: "res5c"
7552     top: "pool5"
7553     name: "pool5"
7554     type: "Pooling"
7555     pooling_param {
7556         kernel_size: 7
7557         stride: 1
7558         pool: AVE
7559     }
7560 }
7561 
7562 layer {
7563     bottom: "pool5"
7564     top: "fc3"
7565     name: "fc3"
7566     type: "InnerProduct"
7567     param {
7568         lr_mult: 1
7569         decay_mult: 1
7570     }
7571     param {
7572         lr_mult: 2
7573         decay_mult: 1
7574     }
7575     inner_product_param {
7576         num_output: 3
7577         weight_filler {
7578             type: "xavier"
7579         }
7580         bias_filler {
7581             type: "constant"
7582             value: 0
7583         }
7584     }
7585 }
7586 
7587 layer {
7588     bottom: "fc3"
7589     bottom: "label"
7590     name: "loss"
7591     type: "SoftmaxWithLoss"
7592     top: "loss"
7593 }
7594 
7595 layer {
7596        name: "probt"
7597        type: "Softmax"
7598        bottom: "fc3"
7599        top: "probt"
7600        include {
7601        phase: TEST
7602     }
7603 }
7604 
7605 layer {
7606   bottom: "fc3"
7607   bottom: "label"
7608   top: "accuracy@1"
7609   name: "accuracy/top1"
7610   type: "Accuracy"
7611   accuracy_param {
7612     top_k: 1
7613   }
7614 }
View Code

 

       deploy.prototxt

   1 name: "ResNet-152"
   2 input: "data"
   3 input_shape {
   4   dim: 1
   5   dim: 3
   6   dim: 224
   7   dim: 224
   8 }
   9 
  10 layer {
  11     bottom: "data"
  12     top: "conv1"
  13     name: "conv1"
  14     type: "Convolution"
  15     convolution_param {
  16         num_output: 64
  17         kernel_size: 7
  18         pad: 3
  19         stride: 2
  20         bias_term: false
  21     }
  22 }
  23 
  24 layer {
  25     bottom: "conv1"
  26     top: "conv1"
  27     name: "bn_conv1"
  28     type: "BatchNorm"
  29     batch_norm_param {
  30         use_global_stats: true
  31     }
  32 }
  33 
  34 layer {
  35     bottom: "conv1"
  36     top: "conv1"
  37     name: "scale_conv1"
  38     type: "Scale"
  39     scale_param {
  40         bias_term: true
  41     }
  42 }
  43 
  44 layer {
  45     top: "conv1"
  46     bottom: "conv1"
  47     name: "conv1_relu"
  48     type: "ReLU"
  49 }
  50 
  51 layer {
  52     bottom: "conv1"
  53     top: "pool1"
  54     name: "pool1"
  55     type: "Pooling"
  56     pooling_param {
  57         kernel_size: 3
  58         stride: 2
  59         pool: MAX
  60     }
  61 }
  62 
  63 layer {
  64     bottom: "pool1"
  65     top: "res2a_branch1"
  66     name: "res2a_branch1"
  67     type: "Convolution"
  68     convolution_param {
  69         num_output: 256
  70         kernel_size: 1
  71         pad: 0
  72         stride: 1
  73         bias_term: false
  74     }
  75 }
  76 
  77 layer {
  78     bottom: "res2a_branch1"
  79     top: "res2a_branch1"
  80     name: "bn2a_branch1"
  81     type: "BatchNorm"
  82     batch_norm_param {
  83         use_global_stats: true
  84     }
  85 }
  86 
  87 layer {
  88     bottom: "res2a_branch1"
  89     top: "res2a_branch1"
  90     name: "scale2a_branch1"
  91     type: "Scale"
  92     scale_param {
  93         bias_term: true
  94     }
  95 }
  96 
  97 layer {
  98     bottom: "pool1"
  99     top: "res2a_branch2a"
 100     name: "res2a_branch2a"
 101     type: "Convolution"
 102     convolution_param {
 103         num_output: 64
 104         kernel_size: 1
 105         pad: 0
 106         stride: 1
 107         bias_term: false
 108     }
 109 }
 110 
 111 layer {
 112     bottom: "res2a_branch2a"
 113     top: "res2a_branch2a"
 114     name: "bn2a_branch2a"
 115     type: "BatchNorm"
 116     batch_norm_param {
 117         use_global_stats: true
 118     }
 119 }
 120 
 121 layer {
 122     bottom: "res2a_branch2a"
 123     top: "res2a_branch2a"
 124     name: "scale2a_branch2a"
 125     type: "Scale"
 126     scale_param {
 127         bias_term: true
 128     }
 129 }
 130 
 131 layer {
 132     top: "res2a_branch2a"
 133     bottom: "res2a_branch2a"
 134     name: "res2a_branch2a_relu"
 135     type: "ReLU"
 136 }
 137 
 138 layer {
 139     bottom: "res2a_branch2a"
 140     top: "res2a_branch2b"
 141     name: "res2a_branch2b"
 142     type: "Convolution"
 143     convolution_param {
 144         num_output: 64
 145         kernel_size: 3
 146         pad: 1
 147         stride: 1
 148         bias_term: false
 149     }
 150 }
 151 
 152 layer {
 153     bottom: "res2a_branch2b"
 154     top: "res2a_branch2b"
 155     name: "bn2a_branch2b"
 156     type: "BatchNorm"
 157     batch_norm_param {
 158         use_global_stats: true
 159     }
 160 }
 161 
 162 layer {
 163     bottom: "res2a_branch2b"
 164     top: "res2a_branch2b"
 165     name: "scale2a_branch2b"
 166     type: "Scale"
 167     scale_param {
 168         bias_term: true
 169     }
 170 }
 171 
 172 layer {
 173     top: "res2a_branch2b"
 174     bottom: "res2a_branch2b"
 175     name: "res2a_branch2b_relu"
 176     type: "ReLU"
 177 }
 178 
 179 layer {
 180     bottom: "res2a_branch2b"
 181     top: "res2a_branch2c"
 182     name: "res2a_branch2c"
 183     type: "Convolution"
 184     convolution_param {
 185         num_output: 256
 186         kernel_size: 1
 187         pad: 0
 188         stride: 1
 189         bias_term: false
 190     }
 191 }
 192 
 193 layer {
 194     bottom: "res2a_branch2c"
 195     top: "res2a_branch2c"
 196     name: "bn2a_branch2c"
 197     type: "BatchNorm"
 198     batch_norm_param {
 199         use_global_stats: true
 200     }
 201 }
 202 
 203 layer {
 204     bottom: "res2a_branch2c"
 205     top: "res2a_branch2c"
 206     name: "scale2a_branch2c"
 207     type: "Scale"
 208     scale_param {
 209         bias_term: true
 210     }
 211 }
 212 
 213 layer {
 214     bottom: "res2a_branch1"
 215     bottom: "res2a_branch2c"
 216     top: "res2a"
 217     name: "res2a"
 218     type: "Eltwise"
 219 }
 220 
 221 layer {
 222     bottom: "res2a"
 223     top: "res2a"
 224     name: "res2a_relu"
 225     type: "ReLU"
 226 }
 227 
 228 layer {
 229     bottom: "res2a"
 230     top: "res2b_branch2a"
 231     name: "res2b_branch2a"
 232     type: "Convolution"
 233     convolution_param {
 234         num_output: 64
 235         kernel_size: 1
 236         pad: 0
 237         stride: 1
 238         bias_term: false
 239     }
 240 }
 241 
 242 layer {
 243     bottom: "res2b_branch2a"
 244     top: "res2b_branch2a"
 245     name: "bn2b_branch2a"
 246     type: "BatchNorm"
 247     batch_norm_param {
 248         use_global_stats: true
 249     }
 250 }
 251 
 252 layer {
 253     bottom: "res2b_branch2a"
 254     top: "res2b_branch2a"
 255     name: "scale2b_branch2a"
 256     type: "Scale"
 257     scale_param {
 258         bias_term: true
 259     }
 260 }
 261 
 262 layer {
 263     top: "res2b_branch2a"
 264     bottom: "res2b_branch2a"
 265     name: "res2b_branch2a_relu"
 266     type: "ReLU"
 267 }
 268 
 269 layer {
 270     bottom: "res2b_branch2a"
 271     top: "res2b_branch2b"
 272     name: "res2b_branch2b"
 273     type: "Convolution"
 274     convolution_param {
 275         num_output: 64
 276         kernel_size: 3
 277         pad: 1
 278         stride: 1
 279         bias_term: false
 280     }
 281 }
 282 
 283 layer {
 284     bottom: "res2b_branch2b"
 285     top: "res2b_branch2b"
 286     name: "bn2b_branch2b"
 287     type: "BatchNorm"
 288     batch_norm_param {
 289         use_global_stats: true
 290     }
 291 }
 292 
 293 layer {
 294     bottom: "res2b_branch2b"
 295     top: "res2b_branch2b"
 296     name: "scale2b_branch2b"
 297     type: "Scale"
 298     scale_param {
 299         bias_term: true
 300     }
 301 }
 302 
 303 layer {
 304     top: "res2b_branch2b"
 305     bottom: "res2b_branch2b"
 306     name: "res2b_branch2b_relu"
 307     type: "ReLU"
 308 }
 309 
 310 layer {
 311     bottom: "res2b_branch2b"
 312     top: "res2b_branch2c"
 313     name: "res2b_branch2c"
 314     type: "Convolution"
 315     convolution_param {
 316         num_output: 256
 317         kernel_size: 1
 318         pad: 0
 319         stride: 1
 320         bias_term: false
 321     }
 322 }
 323 
 324 layer {
 325     bottom: "res2b_branch2c"
 326     top: "res2b_branch2c"
 327     name: "bn2b_branch2c"
 328     type: "BatchNorm"
 329     batch_norm_param {
 330         use_global_stats: true
 331     }
 332 }
 333 
 334 layer {
 335     bottom: "res2b_branch2c"
 336     top: "res2b_branch2c"
 337     name: "scale2b_branch2c"
 338     type: "Scale"
 339     scale_param {
 340         bias_term: true
 341     }
 342 }
 343 
 344 layer {
 345     bottom: "res2a"
 346     bottom: "res2b_branch2c"
 347     top: "res2b"
 348     name: "res2b"
 349     type: "Eltwise"
 350 }
 351 
 352 layer {
 353     bottom: "res2b"
 354     top: "res2b"
 355     name: "res2b_relu"
 356     type: "ReLU"
 357 }
 358 
 359 layer {
 360     bottom: "res2b"
 361     top: "res2c_branch2a"
 362     name: "res2c_branch2a"
 363     type: "Convolution"
 364     convolution_param {
 365         num_output: 64
 366         kernel_size: 1
 367         pad: 0
 368         stride: 1
 369         bias_term: false
 370     }
 371 }
 372 
 373 layer {
 374     bottom: "res2c_branch2a"
 375     top: "res2c_branch2a"
 376     name: "bn2c_branch2a"
 377     type: "BatchNorm"
 378     batch_norm_param {
 379         use_global_stats: true
 380     }
 381 }
 382 
 383 layer {
 384     bottom: "res2c_branch2a"
 385     top: "res2c_branch2a"
 386     name: "scale2c_branch2a"
 387     type: "Scale"
 388     scale_param {
 389         bias_term: true
 390     }
 391 }
 392 
 393 layer {
 394     top: "res2c_branch2a"
 395     bottom: "res2c_branch2a"
 396     name: "res2c_branch2a_relu"
 397     type: "ReLU"
 398 }
 399 
 400 layer {
 401     bottom: "res2c_branch2a"
 402     top: "res2c_branch2b"
 403     name: "res2c_branch2b"
 404     type: "Convolution"
 405     convolution_param {
 406         num_output: 64
 407         kernel_size: 3
 408         pad: 1
 409         stride: 1
 410         bias_term: false
 411     }
 412 }
 413 
 414 layer {
 415     bottom: "res2c_branch2b"
 416     top: "res2c_branch2b"
 417     name: "bn2c_branch2b"
 418     type: "BatchNorm"
 419     batch_norm_param {
 420         use_global_stats: true
 421     }
 422 }
 423 
 424 layer {
 425     bottom: "res2c_branch2b"
 426     top: "res2c_branch2b"
 427     name: "scale2c_branch2b"
 428     type: "Scale"
 429     scale_param {
 430         bias_term: true
 431     }
 432 }
 433 
 434 layer {
 435     top: "res2c_branch2b"
 436     bottom: "res2c_branch2b"
 437     name: "res2c_branch2b_relu"
 438     type: "ReLU"
 439 }
 440 
 441 layer {
 442     bottom: "res2c_branch2b"
 443     top: "res2c_branch2c"
 444     name: "res2c_branch2c"
 445     type: "Convolution"
 446     convolution_param {
 447         num_output: 256
 448         kernel_size: 1
 449         pad: 0
 450         stride: 1
 451         bias_term: false
 452     }
 453 }
 454 
 455 layer {
 456     bottom: "res2c_branch2c"
 457     top: "res2c_branch2c"
 458     name: "bn2c_branch2c"
 459     type: "BatchNorm"
 460     batch_norm_param {
 461         use_global_stats: true
 462     }
 463 }
 464 
 465 layer {
 466     bottom: "res2c_branch2c"
 467     top: "res2c_branch2c"
 468     name: "scale2c_branch2c"
 469     type: "Scale"
 470     scale_param {
 471         bias_term: true
 472     }
 473 }
 474 
 475 layer {
 476     bottom: "res2b"
 477     bottom: "res2c_branch2c"
 478     top: "res2c"
 479     name: "res2c"
 480     type: "Eltwise"
 481 }
 482 
 483 layer {
 484     bottom: "res2c"
 485     top: "res2c"
 486     name: "res2c_relu"
 487     type: "ReLU"
 488 }
 489 
 490 layer {
 491     bottom: "res2c"
 492     top: "res3a_branch1"
 493     name: "res3a_branch1"
 494     type: "Convolution"
 495     convolution_param {
 496         num_output: 512
 497         kernel_size: 1
 498         pad: 0
 499         stride: 2
 500         bias_term: false
 501     }
 502 }
 503 
 504 layer {
 505     bottom: "res3a_branch1"
 506     top: "res3a_branch1"
 507     name: "bn3a_branch1"
 508     type: "BatchNorm"
 509     batch_norm_param {
 510         use_global_stats: true
 511     }
 512 }
 513 
 514 layer {
 515     bottom: "res3a_branch1"
 516     top: "res3a_branch1"
 517     name: "scale3a_branch1"
 518     type: "Scale"
 519     scale_param {
 520         bias_term: true
 521     }
 522 }
 523 
 524 layer {
 525     bottom: "res2c"
 526     top: "res3a_branch2a"
 527     name: "res3a_branch2a"
 528     type: "Convolution"
 529     convolution_param {
 530         num_output: 128
 531         kernel_size: 1
 532         pad: 0
 533         stride: 2
 534         bias_term: false
 535     }
 536 }
 537 
 538 layer {
 539     bottom: "res3a_branch2a"
 540     top: "res3a_branch2a"
 541     name: "bn3a_branch2a"
 542     type: "BatchNorm"
 543     batch_norm_param {
 544         use_global_stats: true
 545     }
 546 }
 547 
 548 layer {
 549     bottom: "res3a_branch2a"
 550     top: "res3a_branch2a"
 551     name: "scale3a_branch2a"
 552     type: "Scale"
 553     scale_param {
 554         bias_term: true
 555     }
 556 }
 557 
 558 layer {
 559     top: "res3a_branch2a"
 560     bottom: "res3a_branch2a"
 561     name: "res3a_branch2a_relu"
 562     type: "ReLU"
 563 }
 564 
 565 layer {
 566     bottom: "res3a_branch2a"
 567     top: "res3a_branch2b"
 568     name: "res3a_branch2b"
 569     type: "Convolution"
 570     convolution_param {
 571         num_output: 128
 572         kernel_size: 3
 573         pad: 1
 574         stride: 1
 575         bias_term: false
 576     }
 577 }
 578 
 579 layer {
 580     bottom: "res3a_branch2b"
 581     top: "res3a_branch2b"
 582     name: "bn3a_branch2b"
 583     type: "BatchNorm"
 584     batch_norm_param {
 585         use_global_stats: true
 586     }
 587 }
 588 
 589 layer {
 590     bottom: "res3a_branch2b"
 591     top: "res3a_branch2b"
 592     name: "scale3a_branch2b"
 593     type: "Scale"
 594     scale_param {
 595         bias_term: true
 596     }
 597 }
 598 
 599 layer {
 600     top: "res3a_branch2b"
 601     bottom: "res3a_branch2b"
 602     name: "res3a_branch2b_relu"
 603     type: "ReLU"
 604 }
 605 
 606 layer {
 607     bottom: "res3a_branch2b"
 608     top: "res3a_branch2c"
 609     name: "res3a_branch2c"
 610     type: "Convolution"
 611     convolution_param {
 612         num_output: 512
 613         kernel_size: 1
 614         pad: 0
 615         stride: 1
 616         bias_term: false
 617     }
 618 }
 619 
 620 layer {
 621     bottom: "res3a_branch2c"
 622     top: "res3a_branch2c"
 623     name: "bn3a_branch2c"
 624     type: "BatchNorm"
 625     batch_norm_param {
 626         use_global_stats: true
 627     }
 628 }
 629 
 630 layer {
 631     bottom: "res3a_branch2c"
 632     top: "res3a_branch2c"
 633     name: "scale3a_branch2c"
 634     type: "Scale"
 635     scale_param {
 636         bias_term: true
 637     }
 638 }
 639 
 640 layer {
 641     bottom: "res3a_branch1"
 642     bottom: "res3a_branch2c"
 643     top: "res3a"
 644     name: "res3a"
 645     type: "Eltwise"
 646 }
 647 
 648 layer {
 649     bottom: "res3a"
 650     top: "res3a"
 651     name: "res3a_relu"
 652     type: "ReLU"
 653 }
 654 
 655 layer {
 656     bottom: "res3a"
 657     top: "res3b1_branch2a"
 658     name: "res3b1_branch2a"
 659     type: "Convolution"
 660     convolution_param {
 661         num_output: 128
 662         kernel_size: 1
 663         pad: 0
 664         stride: 1
 665         bias_term: false
 666     }
 667 }
 668 
 669 layer {
 670     bottom: "res3b1_branch2a"
 671     top: "res3b1_branch2a"
 672     name: "bn3b1_branch2a"
 673     type: "BatchNorm"
 674     batch_norm_param {
 675         use_global_stats: true
 676     }
 677 }
 678 
 679 layer {
 680     bottom: "res3b1_branch2a"
 681     top: "res3b1_branch2a"
 682     name: "scale3b1_branch2a"
 683     type: "Scale"
 684     scale_param {
 685         bias_term: true
 686     }
 687 }
 688 
 689 layer {
 690     top: "res3b1_branch2a"
 691     bottom: "res3b1_branch2a"
 692     name: "res3b1_branch2a_relu"
 693     type: "ReLU"
 694 }
 695 
 696 layer {
 697     bottom: "res3b1_branch2a"
 698     top: "res3b1_branch2b"
 699     name: "res3b1_branch2b"
 700     type: "Convolution"
 701     convolution_param {
 702         num_output: 128
 703         kernel_size: 3
 704         pad: 1
 705         stride: 1
 706         bias_term: false
 707     }
 708 }
 709 
 710 layer {
 711     bottom: "res3b1_branch2b"
 712     top: "res3b1_branch2b"
 713     name: "bn3b1_branch2b"
 714     type: "BatchNorm"
 715     batch_norm_param {
 716         use_global_stats: true
 717     }
 718 }
 719 
 720 layer {
 721     bottom: "res3b1_branch2b"
 722     top: "res3b1_branch2b"
 723     name: "scale3b1_branch2b"
 724     type: "Scale"
 725     scale_param {
 726         bias_term: true
 727     }
 728 }
 729 
 730 layer {
 731     top: "res3b1_branch2b"
 732     bottom: "res3b1_branch2b"
 733     name: "res3b1_branch2b_relu"
 734     type: "ReLU"
 735 }
 736 
 737 layer {
 738     bottom: "res3b1_branch2b"
 739     top: "res3b1_branch2c"
 740     name: "res3b1_branch2c"
 741     type: "Convolution"
 742     convolution_param {
 743         num_output: 512
 744         kernel_size: 1
 745         pad: 0
 746         stride: 1
 747         bias_term: false
 748     }
 749 }
 750 
 751 layer {
 752     bottom: "res3b1_branch2c"
 753     top: "res3b1_branch2c"
 754     name: "bn3b1_branch2c"
 755     type: "BatchNorm"
 756     batch_norm_param {
 757         use_global_stats: true
 758     }
 759 }
 760 
 761 layer {
 762     bottom: "res3b1_branch2c"
 763     top: "res3b1_branch2c"
 764     name: "scale3b1_branch2c"
 765     type: "Scale"
 766     scale_param {
 767         bias_term: true
 768     }
 769 }
 770 
 771 layer {
 772     bottom: "res3a"
 773     bottom: "res3b1_branch2c"
 774     top: "res3b1"
 775     name: "res3b1"
 776     type: "Eltwise"
 777 }
 778 
 779 layer {
 780     bottom: "res3b1"
 781     top: "res3b1"
 782     name: "res3b1_relu"
 783     type: "ReLU"
 784 }
 785 
 786 layer {
 787     bottom: "res3b1"
 788     top: "res3b2_branch2a"
 789     name: "res3b2_branch2a"
 790     type: "Convolution"
 791     convolution_param {
 792         num_output: 128
 793         kernel_size: 1
 794         pad: 0
 795         stride: 1
 796         bias_term: false
 797     }
 798 }
 799 
 800 layer {
 801     bottom: "res3b2_branch2a"
 802     top: "res3b2_branch2a"
 803     name: "bn3b2_branch2a"
 804     type: "BatchNorm"
 805     batch_norm_param {
 806         use_global_stats: true
 807     }
 808 }
 809 
 810 layer {
 811     bottom: "res3b2_branch2a"
 812     top: "res3b2_branch2a"
 813     name: "scale3b2_branch2a"
 814     type: "Scale"
 815     scale_param {
 816         bias_term: true
 817     }
 818 }
 819 
 820 layer {
 821     top: "res3b2_branch2a"
 822     bottom: "res3b2_branch2a"
 823     name: "res3b2_branch2a_relu"
 824     type: "ReLU"
 825 }
 826 
 827 layer {
 828     bottom: "res3b2_branch2a"
 829     top: "res3b2_branch2b"
 830     name: "res3b2_branch2b"
 831     type: "Convolution"
 832     convolution_param {
 833         num_output: 128
 834         kernel_size: 3
 835         pad: 1
 836         stride: 1
 837         bias_term: false
 838     }
 839 }
 840 
 841 layer {
 842     bottom: "res3b2_branch2b"
 843     top: "res3b2_branch2b"
 844     name: "bn3b2_branch2b"
 845     type: "BatchNorm"
 846     batch_norm_param {
 847         use_global_stats: true
 848     }
 849 }
 850 
 851 layer {
 852     bottom: "res3b2_branch2b"
 853     top: "res3b2_branch2b"
 854     name: "scale3b2_branch2b"
 855     type: "Scale"
 856     scale_param {
 857         bias_term: true
 858     }
 859 }
 860 
 861 layer {
 862     top: "res3b2_branch2b"
 863     bottom: "res3b2_branch2b"
 864     name: "res3b2_branch2b_relu"
 865     type: "ReLU"
 866 }
 867 
 868 layer {
 869     bottom: "res3b2_branch2b"
 870     top: "res3b2_branch2c"
 871     name: "res3b2_branch2c"
 872     type: "Convolution"
 873     convolution_param {
 874         num_output: 512
 875         kernel_size: 1
 876         pad: 0
 877         stride: 1
 878         bias_term: false
 879     }
 880 }
 881 
 882 layer {
 883     bottom: "res3b2_branch2c"
 884     top: "res3b2_branch2c"
 885     name: "bn3b2_branch2c"
 886     type: "BatchNorm"
 887     batch_norm_param {
 888         use_global_stats: true
 889     }
 890 }
 891 
 892 layer {
 893     bottom: "res3b2_branch2c"
 894     top: "res3b2_branch2c"
 895     name: "scale3b2_branch2c"
 896     type: "Scale"
 897     scale_param {
 898         bias_term: true
 899     }
 900 }
 901 
 902 layer {
 903     bottom: "res3b1"
 904     bottom: "res3b2_branch2c"
 905     top: "res3b2"
 906     name: "res3b2"
 907     type: "Eltwise"
 908 }
 909 
 910 layer {
 911     bottom: "res3b2"
 912     top: "res3b2"
 913     name: "res3b2_relu"
 914     type: "ReLU"
 915 }
 916 
 917 layer {
 918     bottom: "res3b2"
 919     top: "res3b3_branch2a"
 920     name: "res3b3_branch2a"
 921     type: "Convolution"
 922     convolution_param {
 923         num_output: 128
 924         kernel_size: 1
 925         pad: 0
 926         stride: 1
 927         bias_term: false
 928     }
 929 }
 930 
 931 layer {
 932     bottom: "res3b3_branch2a"
 933     top: "res3b3_branch2a"
 934     name: "bn3b3_branch2a"
 935     type: "BatchNorm"
 936     batch_norm_param {
 937         use_global_stats: true
 938     }
 939 }
 940 
 941 layer {
 942     bottom: "res3b3_branch2a"
 943     top: "res3b3_branch2a"
 944     name: "scale3b3_branch2a"
 945     type: "Scale"
 946