SHARE
TWEET

Untitled

a guest Jul 23rd, 2019 97 Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
  1. name: "Openpose"
  2. input: "image"
  3. input_dim: 1
  4. input_dim: 3
  5. input_dim: 1 # This value will be defined at runtime
  6. input_dim: 1 # This value will be defined at runtime
  7. layer {
  8.   name: "conv1_1"
  9.   type: "Convolution"
  10.   bottom: "image"
  11.   top: "conv1_1"
  12.   param {
  13.     lr_mult: 1.0
  14.     decay_mult: 1
  15.   }
  16.   param {
  17.     lr_mult: 2.0
  18.     decay_mult: 0
  19.   }
  20.   convolution_param {
  21.     num_output: 64
  22.     pad: 1
  23.     kernel_size: 3
  24.     weight_filler {
  25.       type: "gaussian"
  26.       std: 0.01
  27.     }
  28.     bias_filler {
  29.       type: "constant"
  30.     }
  31.   }
  32. }
  33. layer {
  34.   name: "relu1_1"
  35.   type: "ReLU"
  36.   bottom: "conv1_1"
  37.   top: "conv1_1"
  38. }
  39. layer {
  40.   name: "conv1_2"
  41.   type: "Convolution"
  42.   bottom: "conv1_1"
  43.   top: "conv1_2"
  44.   param {
  45.     lr_mult: 1.0
  46.     decay_mult: 1
  47.   }
  48.   param {
  49.     lr_mult: 2.0
  50.     decay_mult: 0
  51.   }
  52.   convolution_param {
  53.     num_output: 64
  54.     pad: 1
  55.     kernel_size: 3
  56.     weight_filler {
  57.       type: "gaussian"
  58.       std: 0.01
  59.     }
  60.     bias_filler {
  61.       type: "constant"
  62.     }
  63.   }
  64. }
  65. layer {
  66.   name: "relu1_2"
  67.   type: "ReLU"
  68.   bottom: "conv1_2"
  69.   top: "conv1_2"
  70. }
  71. layer {
  72.   name: "pool1_stage1"
  73.   type: "Pooling"
  74.   bottom: "conv1_2"
  75.   top: "pool1_stage1"
  76.   pooling_param {
  77.     pool: MAX
  78.     kernel_size: 2
  79.     stride: 2
  80.   }
  81. }
  82. layer {
  83.   name: "conv2_1"
  84.   type: "Convolution"
  85.   bottom: "pool1_stage1"
  86.   top: "conv2_1"
  87.   param {
  88.     lr_mult: 1.0
  89.     decay_mult: 1
  90.   }
  91.   param {
  92.     lr_mult: 2.0
  93.     decay_mult: 0
  94.   }
  95.   convolution_param {
  96.     num_output: 128
  97.     pad: 1
  98.     kernel_size: 3
  99.     weight_filler {
  100.       type: "gaussian"
  101.       std: 0.01
  102.     }
  103.     bias_filler {
  104.       type: "constant"
  105.     }
  106.   }
  107. }
  108. layer {
  109.   name: "relu2_1"
  110.   type: "ReLU"
  111.   bottom: "conv2_1"
  112.   top: "conv2_1"
  113. }
  114. layer {
  115.   name: "conv2_2"
  116.   type: "Convolution"
  117.   bottom: "conv2_1"
  118.   top: "conv2_2"
  119.   param {
  120.     lr_mult: 1.0
  121.     decay_mult: 1
  122.   }
  123.   param {
  124.     lr_mult: 2.0
  125.     decay_mult: 0
  126.   }
  127.   convolution_param {
  128.     num_output: 128
  129.     pad: 1
  130.     kernel_size: 3
  131.     weight_filler {
  132.       type: "gaussian"
  133.       std: 0.01
  134.     }
  135.     bias_filler {
  136.       type: "constant"
  137.     }
  138.   }
  139. }
  140. layer {
  141.   name: "relu2_2"
  142.   type: "ReLU"
  143.   bottom: "conv2_2"
  144.   top: "conv2_2"
  145. }
  146. layer {
  147.   name: "pool2_stage1"
  148.   type: "Pooling"
  149.   bottom: "conv2_2"
  150.   top: "pool2_stage1"
  151.   pooling_param {
  152.     pool: MAX
  153.     kernel_size: 2
  154.     stride: 2
  155.   }
  156. }
  157. layer {
  158.   name: "conv3_1"
  159.   type: "Convolution"
  160.   bottom: "pool2_stage1"
  161.   top: "conv3_1"
  162.   param {
  163.     lr_mult: 1.0
  164.     decay_mult: 1
  165.   }
  166.   param {
  167.     lr_mult: 2.0
  168.     decay_mult: 0
  169.   }
  170.   convolution_param {
  171.     num_output: 256
  172.     pad: 1
  173.     kernel_size: 3
  174.     weight_filler {
  175.       type: "gaussian"
  176.       std: 0.01
  177.     }
  178.     bias_filler {
  179.       type: "constant"
  180.     }
  181.   }
  182. }
  183. layer {
  184.   name: "relu3_1"
  185.   type: "ReLU"
  186.   bottom: "conv3_1"
  187.   top: "conv3_1"
  188. }
  189. layer {
  190.   name: "conv3_2"
  191.   type: "Convolution"
  192.   bottom: "conv3_1"
  193.   top: "conv3_2"
  194.   param {
  195.     lr_mult: 1.0
  196.     decay_mult: 1
  197.   }
  198.   param {
  199.     lr_mult: 2.0
  200.     decay_mult: 0
  201.   }
  202.   convolution_param {
  203.     num_output: 256
  204.     pad: 1
  205.     kernel_size: 3
  206.     weight_filler {
  207.       type: "gaussian"
  208.       std: 0.01
  209.     }
  210.     bias_filler {
  211.       type: "constant"
  212.     }
  213.   }
  214. }
  215. layer {
  216.   name: "relu3_2"
  217.   type: "ReLU"
  218.   bottom: "conv3_2"
  219.   top: "conv3_2"
  220. }
  221. layer {
  222.   name: "conv3_3"
  223.   type: "Convolution"
  224.   bottom: "conv3_2"
  225.   top: "conv3_3"
  226.   param {
  227.     lr_mult: 1.0
  228.     decay_mult: 1
  229.   }
  230.   param {
  231.     lr_mult: 2.0
  232.     decay_mult: 0
  233.   }
  234.   convolution_param {
  235.     num_output: 256
  236.     pad: 1
  237.     kernel_size: 3
  238.     weight_filler {
  239.       type: "gaussian"
  240.       std: 0.01
  241.     }
  242.     bias_filler {
  243.       type: "constant"
  244.     }
  245.   }
  246. }
  247. layer {
  248.   name: "relu3_3"
  249.   type: "ReLU"
  250.   bottom: "conv3_3"
  251.   top: "conv3_3"
  252. }
  253. layer {
  254.   name: "conv3_4"
  255.   type: "Convolution"
  256.   bottom: "conv3_3"
  257.   top: "conv3_4"
  258.   param {
  259.     lr_mult: 1.0
  260.     decay_mult: 1
  261.   }
  262.   param {
  263.     lr_mult: 2.0
  264.     decay_mult: 0
  265.   }
  266.   convolution_param {
  267.     num_output: 256
  268.     pad: 1
  269.     kernel_size: 3
  270.     weight_filler {
  271.       type: "gaussian"
  272.       std: 0.01
  273.     }
  274.     bias_filler {
  275.       type: "constant"
  276.     }
  277.   }
  278. }
  279. layer {
  280.   name: "relu3_4"
  281.   type: "ReLU"
  282.   bottom: "conv3_4"
  283.   top: "conv3_4"
  284. }
  285. layer {
  286.   name: "pool3_stage1"
  287.   type: "Pooling"
  288.   bottom: "conv3_4"
  289.   top: "pool3_stage1"
  290.   pooling_param {
  291.     pool: MAX
  292.     kernel_size: 2
  293.     stride: 2
  294.   }
  295. }
  296. layer {
  297.   name: "conv4_1"
  298.   type: "Convolution"
  299.   bottom: "pool3_stage1"
  300.   top: "conv4_1"
  301.   param {
  302.     lr_mult: 1.0
  303.     decay_mult: 1
  304.   }
  305.   param {
  306.     lr_mult: 2.0
  307.     decay_mult: 0
  308.   }
  309.   convolution_param {
  310.     num_output: 512
  311.     pad: 1
  312.     kernel_size: 3
  313.     weight_filler {
  314.       type: "gaussian"
  315.       std: 0.01
  316.     }
  317.     bias_filler {
  318.       type: "constant"
  319.     }
  320.   }
  321. }
  322. layer {
  323.   name: "relu4_1"
  324.   type: "ReLU"
  325.   bottom: "conv4_1"
  326.   top: "conv4_1"
  327. }
  328. layer {
  329.   name: "conv4_2"
  330.   type: "Convolution"
  331.   bottom: "conv4_1"
  332.   top: "conv4_2"
  333.   param {
  334.     lr_mult: 1.0
  335.     decay_mult: 1
  336.   }
  337.   param {
  338.     lr_mult: 2.0
  339.     decay_mult: 0
  340.   }
  341.   convolution_param {
  342.     num_output: 512
  343.     pad: 1
  344.     kernel_size: 3
  345.     weight_filler {
  346.       type: "gaussian"
  347.       std: 0.01
  348.     }
  349.     bias_filler {
  350.       type: "constant"
  351.     }
  352.   }
  353. }
  354. layer {
  355.   name: "relu4_2"
  356.   type: "ReLU"
  357.   bottom: "conv4_2"
  358.   top: "conv4_2"
  359. }
  360. layer {
  361.   name: "conv4_3_CPM"
  362.   type: "Convolution"
  363.   bottom: "conv4_2"
  364.   top: "conv4_3_CPM"
  365.   param {
  366.     lr_mult: 1.0
  367.     decay_mult: 1
  368.   }
  369.   param {
  370.     lr_mult: 2.0
  371.     decay_mult: 0
  372.   }
  373.   convolution_param {
  374.     num_output: 256
  375.     pad: 1
  376.     kernel_size: 3
  377.     weight_filler {
  378.       type: "gaussian"
  379.       std: 0.01
  380.     }
  381.     bias_filler {
  382.       type: "constant"
  383.     }
  384.   }
  385. }
  386. layer {
  387.   name: "relu4_3_CPM"
  388.   type: "ReLU"
  389.   bottom: "conv4_3_CPM"
  390.   top: "conv4_3_CPM"
  391. }
  392. layer {
  393.   name: "conv4_4_CPM"
  394.   type: "Convolution"
  395.   bottom: "conv4_3_CPM"
  396.   top: "conv4_4_CPM"
  397.   param {
  398.     lr_mult: 1.0
  399.     decay_mult: 1
  400.   }
  401.   param {
  402.     lr_mult: 2.0
  403.     decay_mult: 0
  404.   }
  405.   convolution_param {
  406.     num_output: 128
  407.     pad: 1
  408.     kernel_size: 3
  409.     weight_filler {
  410.       type: "gaussian"
  411.       std: 0.01
  412.     }
  413.     bias_filler {
  414.       type: "constant"
  415.     }
  416.   }
  417. }
  418. layer {
  419.   name: "relu4_4_CPM"
  420.   type: "ReLU"
  421.   bottom: "conv4_4_CPM"
  422.   top: "conv4_4_CPM"
  423. }
  424. layer {
  425.   name: "conv5_1_CPM_L1"
  426.   type: "Convolution"
  427.   bottom: "conv4_4_CPM"
  428.   top: "conv5_1_CPM_L1"
  429.   param {
  430.     lr_mult: 1.0
  431.     decay_mult: 1
  432.   }
  433.   param {
  434.     lr_mult: 2.0
  435.     decay_mult: 0
  436.   }
  437.   convolution_param {
  438.     num_output: 128
  439.     pad: 1
  440.     kernel_size: 3
  441.     weight_filler {
  442.       type: "gaussian"
  443.       std: 0.01
  444.     }
  445.     bias_filler {
  446.       type: "constant"
  447.     }
  448.   }
  449. }
  450. layer {
  451.   name: "relu5_1_CPM_L1"
  452.   type: "ReLU"
  453.   bottom: "conv5_1_CPM_L1"
  454.   top: "conv5_1_CPM_L1"
  455. }
  456. layer {
  457.   name: "conv5_1_CPM_L2"
  458.   type: "Convolution"
  459.   bottom: "conv4_4_CPM"
  460.   top: "conv5_1_CPM_L2"
  461.   param {
  462.     lr_mult: 1.0
  463.     decay_mult: 1
  464.   }
  465.   param {
  466.     lr_mult: 2.0
  467.     decay_mult: 0
  468.   }
  469.   convolution_param {
  470.     num_output: 128
  471.     pad: 1
  472.     kernel_size: 3
  473.     weight_filler {
  474.       type: "gaussian"
  475.       std: 0.01
  476.     }
  477.     bias_filler {
  478.       type: "constant"
  479.     }
  480.   }
  481. }
  482. layer {
  483.   name: "relu5_1_CPM_L2"
  484.   type: "ReLU"
  485.   bottom: "conv5_1_CPM_L2"
  486.   top: "conv5_1_CPM_L2"
  487. }
  488. layer {
  489.   name: "conv5_2_CPM_L1"
  490.   type: "Convolution"
  491.   bottom: "conv5_1_CPM_L1"
  492.   top: "conv5_2_CPM_L1"
  493.   param {
  494.     lr_mult: 1.0
  495.     decay_mult: 1
  496.   }
  497.   param {
  498.     lr_mult: 2.0
  499.     decay_mult: 0
  500.   }
  501.   convolution_param {
  502.     num_output: 128
  503.     pad: 1
  504.     kernel_size: 3
  505.     weight_filler {
  506.       type: "gaussian"
  507.       std: 0.01
  508.     }
  509.     bias_filler {
  510.       type: "constant"
  511.     }
  512.   }
  513. }
  514. layer {
  515.   name: "relu5_2_CPM_L1"
  516.   type: "ReLU"
  517.   bottom: "conv5_2_CPM_L1"
  518.   top: "conv5_2_CPM_L1"
  519. }
  520. layer {
  521.   name: "conv5_2_CPM_L2"
  522.   type: "Convolution"
  523.   bottom: "conv5_1_CPM_L2"
  524.   top: "conv5_2_CPM_L2"
  525.   param {
  526.     lr_mult: 1.0
  527.     decay_mult: 1
  528.   }
  529.   param {
  530.     lr_mult: 2.0
  531.     decay_mult: 0
  532.   }
  533.   convolution_param {
  534.     num_output: 128
  535.     pad: 1
  536.     kernel_size: 3
  537.     weight_filler {
  538.       type: "gaussian"
  539.       std: 0.01
  540.     }
  541.     bias_filler {
  542.       type: "constant"
  543.     }
  544.   }
  545. }
  546. layer {
  547.   name: "relu5_2_CPM_L2"
  548.   type: "ReLU"
  549.   bottom: "conv5_2_CPM_L2"
  550.   top: "conv5_2_CPM_L2"
  551. }
  552. layer {
  553.   name: "conv5_3_CPM_L1"
  554.   type: "Convolution"
  555.   bottom: "conv5_2_CPM_L1"
  556.   top: "conv5_3_CPM_L1"
  557.   param {
  558.     lr_mult: 1.0
  559.     decay_mult: 1
  560.   }
  561.   param {
  562.     lr_mult: 2.0
  563.     decay_mult: 0
  564.   }
  565.   convolution_param {
  566.     num_output: 128
  567.     pad: 1
  568.     kernel_size: 3
  569.     weight_filler {
  570.       type: "gaussian"
  571.       std: 0.01
  572.     }
  573.     bias_filler {
  574.       type: "constant"
  575.     }
  576.   }
  577. }
  578. layer {
  579.   name: "relu5_3_CPM_L1"
  580.   type: "ReLU"
  581.   bottom: "conv5_3_CPM_L1"
  582.   top: "conv5_3_CPM_L1"
  583. }
  584. layer {
  585.   name: "conv5_3_CPM_L2"
  586.   type: "Convolution"
  587.   bottom: "conv5_2_CPM_L2"
  588.   top: "conv5_3_CPM_L2"
  589.   param {
  590.     lr_mult: 1.0
  591.     decay_mult: 1
  592.   }
  593.   param {
  594.     lr_mult: 2.0
  595.     decay_mult: 0
  596.   }
  597.   convolution_param {
  598.     num_output: 128
  599.     pad: 1
  600.     kernel_size: 3
  601.     weight_filler {
  602.       type: "gaussian"
  603.       std: 0.01
  604.     }
  605.     bias_filler {
  606.       type: "constant"
  607.     }
  608.   }
  609. }
  610. layer {
  611.   name: "relu5_3_CPM_L2"
  612.   type: "ReLU"
  613.   bottom: "conv5_3_CPM_L2"
  614.   top: "conv5_3_CPM_L2"
  615. }
  616. layer {
  617.   name: "conv5_4_CPM_L1"
  618.   type: "Convolution"
  619.   bottom: "conv5_3_CPM_L1"
  620.   top: "conv5_4_CPM_L1"
  621.   param {
  622.     lr_mult: 1.0
  623.     decay_mult: 1
  624.   }
  625.   param {
  626.     lr_mult: 2.0
  627.     decay_mult: 0
  628.   }
  629.   convolution_param {
  630.     num_output: 512
  631.     pad: 0
  632.     kernel_size: 1
  633.     weight_filler {
  634.       type: "gaussian"
  635.       std: 0.01
  636.     }
  637.     bias_filler {
  638.       type: "constant"
  639.     }
  640.   }
  641. }
  642. layer {
  643.   name: "relu5_4_CPM_L1"
  644.   type: "ReLU"
  645.   bottom: "conv5_4_CPM_L1"
  646.   top: "conv5_4_CPM_L1"
  647. }
  648. layer {
  649.   name: "conv5_4_CPM_L2"
  650.   type: "Convolution"
  651.   bottom: "conv5_3_CPM_L2"
  652.   top: "conv5_4_CPM_L2"
  653.   param {
  654.     lr_mult: 1.0
  655.     decay_mult: 1
  656.   }
  657.   param {
  658.     lr_mult: 2.0
  659.     decay_mult: 0
  660.   }
  661.   convolution_param {
  662.     num_output: 512
  663.     pad: 0
  664.     kernel_size: 1
  665.     weight_filler {
  666.       type: "gaussian"
  667.       std: 0.01
  668.     }
  669.     bias_filler {
  670.       type: "constant"
  671.     }
  672.   }
  673. }
  674. layer {
  675.   name: "relu5_4_CPM_L2"
  676.   type: "ReLU"
  677.   bottom: "conv5_4_CPM_L2"
  678.   top: "conv5_4_CPM_L2"
  679. }
  680. layer {
  681.   name: "conv5_5_CPM_L1"
  682.   type: "Convolution"
  683.   bottom: "conv5_4_CPM_L1"
  684.   top: "conv5_5_CPM_L1"
  685.   param {
  686.     lr_mult: 1.0
  687.     decay_mult: 1
  688.   }
  689.   param {
  690.     lr_mult: 2.0
  691.     decay_mult: 0
  692.   }
  693.   convolution_param {
  694.     num_output: 38
  695.     pad: 0
  696.     kernel_size: 1
  697.     weight_filler {
  698.       type: "gaussian"
  699.       std: 0.01
  700.     }
  701.     bias_filler {
  702.       type: "constant"
  703.     }
  704.   }
  705. }
  706. layer {
  707.   name: "conv5_5_CPM_L2"
  708.   type: "Convolution"
  709.   bottom: "conv5_4_CPM_L2"
  710.   top: "conv5_5_CPM_L2"
  711.   param {
  712.     lr_mult: 1.0
  713.     decay_mult: 1
  714.   }
  715.   param {
  716.     lr_mult: 2.0
  717.     decay_mult: 0
  718.   }
  719.   convolution_param {
  720.     num_output: 19
  721.     pad: 0
  722.     kernel_size: 1
  723.     weight_filler {
  724.       type: "gaussian"
  725.       std: 0.01
  726.     }
  727.     bias_filler {
  728.       type: "constant"
  729.     }
  730.   }
  731. }
  732. layer {
  733.   name: "concat_stage2"
  734.   type: "Concat"
  735.   bottom: "conv5_5_CPM_L1"
  736.   bottom: "conv5_5_CPM_L2"
  737.   bottom: "conv4_4_CPM"
  738.   top: "concat_stage2"
  739.   concat_param {
  740.     axis: 1
  741.   }
  742. }
  743. layer {
  744.   name: "Mconv1_stage2_L1"
  745.   type: "Convolution"
  746.   bottom: "concat_stage2"
  747.   top: "Mconv1_stage2_L1"
  748.   param {
  749.     lr_mult: 4.0
  750.     decay_mult: 1
  751.   }
  752.   param {
  753.     lr_mult: 8.0
  754.     decay_mult: 0
  755.   }
  756.   convolution_param {
  757.     num_output: 128
  758.     pad: 3
  759.     kernel_size: 7
  760.     weight_filler {
  761.       type: "gaussian"
  762.       std: 0.01
  763.     }
  764.     bias_filler {
  765.       type: "constant"
  766.     }
  767.   }
  768. }
  769. layer {
  770.   name: "Mrelu1_stage2_L1"
  771.   type: "ReLU"
  772.   bottom: "Mconv1_stage2_L1"
  773.   top: "Mconv1_stage2_L1"
  774. }
  775. layer {
  776.   name: "Mconv1_stage2_L2"
  777.   type: "Convolution"
  778.   bottom: "concat_stage2"
  779.   top: "Mconv1_stage2_L2"
  780.   param {
  781.     lr_mult: 4.0
  782.     decay_mult: 1
  783.   }
  784.   param {
  785.     lr_mult: 8.0
  786.     decay_mult: 0
  787.   }
  788.   convolution_param {
  789.     num_output: 128
  790.     pad: 3
  791.     kernel_size: 7
  792.     weight_filler {
  793.       type: "gaussian"
  794.       std: 0.01
  795.     }
  796.     bias_filler {
  797.       type: "constant"
  798.     }
  799.   }
  800. }
  801. layer {
  802.   name: "Mrelu1_stage2_L2"
  803.   type: "ReLU"
  804.   bottom: "Mconv1_stage2_L2"
  805.   top: "Mconv1_stage2_L2"
  806. }
  807. layer {
  808.   name: "Mconv2_stage2_L1"
  809.   type: "Convolution"
  810.   bottom: "Mconv1_stage2_L1"
  811.   top: "Mconv2_stage2_L1"
  812.   param {
  813.     lr_mult: 4.0
  814.     decay_mult: 1
  815.   }
  816.   param {
  817.     lr_mult: 8.0
  818.     decay_mult: 0
  819.   }
  820.   convolution_param {
  821.     num_output: 128
  822.     pad: 3
  823.     kernel_size: 7
  824.     weight_filler {
  825.       type: "gaussian"
  826.       std: 0.01
  827.     }
  828.     bias_filler {
  829.       type: "constant"
  830.     }
  831.   }
  832. }
  833. layer {
  834.   name: "Mrelu2_stage2_L1"
  835.   type: "ReLU"
  836.   bottom: "Mconv2_stage2_L1"
  837.   top: "Mconv2_stage2_L1"
  838. }
  839. layer {
  840.   name: "Mconv2_stage2_L2"
  841.   type: "Convolution"
  842.   bottom: "Mconv1_stage2_L2"
  843.   top: "Mconv2_stage2_L2"
  844.   param {
  845.     lr_mult: 4.0
  846.     decay_mult: 1
  847.   }
  848.   param {
  849.     lr_mult: 8.0
  850.     decay_mult: 0
  851.   }
  852.   convolution_param {
  853.     num_output: 128
  854.     pad: 3
  855.     kernel_size: 7
  856.     weight_filler {
  857.       type: "gaussian"
  858.       std: 0.01
  859.     }
  860.     bias_filler {
  861.       type: "constant"
  862.     }
  863.   }
  864. }
  865. layer {
  866.   name: "Mrelu2_stage2_L2"
  867.   type: "ReLU"
  868.   bottom: "Mconv2_stage2_L2"
  869.   top: "Mconv2_stage2_L2"
  870. }
  871. layer {
  872.   name: "Mconv3_stage2_L1"
  873.   type: "Convolution"
  874.   bottom: "Mconv2_stage2_L1"
  875.   top: "Mconv3_stage2_L1"
  876.   param {
  877.     lr_mult: 4.0
  878.     decay_mult: 1
  879.   }
  880.   param {
  881.     lr_mult: 8.0
  882.     decay_mult: 0
  883.   }
  884.   convolution_param {
  885.     num_output: 128
  886.     pad: 3
  887.     kernel_size: 7
  888.     weight_filler {
  889.       type: "gaussian"
  890.       std: 0.01
  891.     }
  892.     bias_filler {
  893.       type: "constant"
  894.     }
  895.   }
  896. }
  897. layer {
  898.   name: "Mrelu3_stage2_L1"
  899.   type: "ReLU"
  900.   bottom: "Mconv3_stage2_L1"
  901.   top: "Mconv3_stage2_L1"
  902. }
  903. layer {
  904.   name: "Mconv3_stage2_L2"
  905.   type: "Convolution"
  906.   bottom: "Mconv2_stage2_L2"
  907.   top: "Mconv3_stage2_L2"
  908.   param {
  909.     lr_mult: 4.0
  910.     decay_mult: 1
  911.   }
  912.   param {
  913.     lr_mult: 8.0
  914.     decay_mult: 0
  915.   }
  916.   convolution_param {
  917.     num_output: 128
  918.     pad: 3
  919.     kernel_size: 7
  920.     weight_filler {
  921.       type: "gaussian"
  922.       std: 0.01
  923.     }
  924.     bias_filler {
  925.       type: "constant"
  926.     }
  927.   }
  928. }
  929. layer {
  930.   name: "Mrelu3_stage2_L2"
  931.   type: "ReLU"
  932.   bottom: "Mconv3_stage2_L2"
  933.   top: "Mconv3_stage2_L2"
  934. }
  935. layer {
  936.   name: "Mconv4_stage2_L1"
  937.   type: "Convolution"
  938.   bottom: "Mconv3_stage2_L1"
  939.   top: "Mconv4_stage2_L1"
  940.   param {
  941.     lr_mult: 4.0
  942.     decay_mult: 1
  943.   }
  944.   param {
  945.     lr_mult: 8.0
  946.     decay_mult: 0
  947.   }
  948.   convolution_param {
  949.     num_output: 128
  950.     pad: 3
  951.     kernel_size: 7
  952.     weight_filler {
  953.       type: "gaussian"
  954.       std: 0.01
  955.     }
  956.     bias_filler {
  957.       type: "constant"
  958.     }
  959.   }
  960. }
  961. layer {
  962.   name: "Mrelu4_stage2_L1"
  963.   type: "ReLU"
  964.   bottom: "Mconv4_stage2_L1"
  965.   top: "Mconv4_stage2_L1"
  966. }
  967. layer {
  968.   name: "Mconv4_stage2_L2"
  969.   type: "Convolution"
  970.   bottom: "Mconv3_stage2_L2"
  971.   top: "Mconv4_stage2_L2"
  972.   param {
  973.     lr_mult: 4.0
  974.     decay_mult: 1
  975.   }
  976.   param {
  977.     lr_mult: 8.0
  978.     decay_mult: 0
  979.   }
  980.   convolution_param {
  981.     num_output: 128
  982.     pad: 3
  983.     kernel_size: 7
  984.     weight_filler {
  985.       type: "gaussian"
  986.       std: 0.01
  987.     }
  988.     bias_filler {
  989.       type: "constant"
  990.     }
  991.   }
  992. }
  993. layer {
  994.   name: "Mrelu4_stage2_L2"
  995.   type: "ReLU"
  996.   bottom: "Mconv4_stage2_L2"
  997.   top: "Mconv4_stage2_L2"
  998. }
  999. layer {
  1000.   name: "Mconv5_stage2_L1"
  1001.   type: "Convolution"
  1002.   bottom: "Mconv4_stage2_L1"
  1003.   top: "Mconv5_stage2_L1"
  1004.   param {
  1005.     lr_mult: 4.0
  1006.     decay_mult: 1
  1007.   }
  1008.   param {
  1009.     lr_mult: 8.0
  1010.     decay_mult: 0
  1011.   }
  1012.   convolution_param {
  1013.     num_output: 128
  1014.     pad: 3
  1015.     kernel_size: 7
  1016.     weight_filler {
  1017.       type: "gaussian"
  1018.       std: 0.01
  1019.     }
  1020.     bias_filler {
  1021.       type: "constant"
  1022.     }
  1023.   }
  1024. }
  1025. layer {
  1026.   name: "Mrelu5_stage2_L1"
  1027.   type: "ReLU"
  1028.   bottom: "Mconv5_stage2_L1"
  1029.   top: "Mconv5_stage2_L1"
  1030. }
  1031. layer {
  1032.   name: "Mconv5_stage2_L2"
  1033.   type: "Convolution"
  1034.   bottom: "Mconv4_stage2_L2"
  1035.   top: "Mconv5_stage2_L2"
  1036.   param {
  1037.     lr_mult: 4.0
  1038.     decay_mult: 1
  1039.   }
  1040.   param {
  1041.     lr_mult: 8.0
  1042.     decay_mult: 0
  1043.   }
  1044.   convolution_param {
  1045.     num_output: 128
  1046.     pad: 3
  1047.     kernel_size: 7
  1048.     weight_filler {
  1049.       type: "gaussian"
  1050.       std: 0.01
  1051.     }
  1052.     bias_filler {
  1053.       type: "constant"
  1054.     }
  1055.   }
  1056. }
  1057. layer {
  1058.   name: "Mrelu5_stage2_L2"
  1059.   type: "ReLU"
  1060.   bottom: "Mconv5_stage2_L2"
  1061.   top: "Mconv5_stage2_L2"
  1062. }
  1063. layer {
  1064.   name: "Mconv6_stage2_L1"
  1065.   type: "Convolution"
  1066.   bottom: "Mconv5_stage2_L1"
  1067.   top: "Mconv6_stage2_L1"
  1068.   param {
  1069.     lr_mult: 4.0
  1070.     decay_mult: 1
  1071.   }
  1072.   param {
  1073.     lr_mult: 8.0
  1074.     decay_mult: 0
  1075.   }
  1076.   convolution_param {
  1077.     num_output: 128
  1078.     pad: 0
  1079.     kernel_size: 1
  1080.     weight_filler {
  1081.       type: "gaussian"
  1082.       std: 0.01
  1083.     }
  1084.     bias_filler {
  1085.       type: "constant"
  1086.     }
  1087.   }
  1088. }
  1089. layer {
  1090.   name: "Mrelu6_stage2_L1"
  1091.   type: "ReLU"
  1092.   bottom: "Mconv6_stage2_L1"
  1093.   top: "Mconv6_stage2_L1"
  1094. }
  1095. layer {
  1096.   name: "Mconv6_stage2_L2"
  1097.   type: "Convolution"
  1098.   bottom: "Mconv5_stage2_L2"
  1099.   top: "Mconv6_stage2_L2"
  1100.   param {
  1101.     lr_mult: 4.0
  1102.     decay_mult: 1
  1103.   }
  1104.   param {
  1105.     lr_mult: 8.0
  1106.     decay_mult: 0
  1107.   }
  1108.   convolution_param {
  1109.     num_output: 128
  1110.     pad: 0
  1111.     kernel_size: 1
  1112.     weight_filler {
  1113.       type: "gaussian"
  1114.       std: 0.01
  1115.     }
  1116.     bias_filler {
  1117.       type: "constant"
  1118.     }
  1119.   }
  1120. }
  1121. layer {
  1122.   name: "Mrelu6_stage2_L2"
  1123.   type: "ReLU"
  1124.   bottom: "Mconv6_stage2_L2"
  1125.   top: "Mconv6_stage2_L2"
  1126. }
  1127. layer {
  1128.   name: "Mconv7_stage2_L1"
  1129.   type: "Convolution"
  1130.   bottom: "Mconv6_stage2_L1"
  1131.   top: "Mconv7_stage2_L1"
  1132.   param {
  1133.     lr_mult: 4.0
  1134.     decay_mult: 1
  1135.   }
  1136.   param {
  1137.     lr_mult: 8.0
  1138.     decay_mult: 0
  1139.   }
  1140.   convolution_param {
  1141.     num_output: 38
  1142.     pad: 0
  1143.     kernel_size: 1
  1144.     weight_filler {
  1145.       type: "gaussian"
  1146.       std: 0.01
  1147.     }
  1148.     bias_filler {
  1149.       type: "constant"
  1150.     }
  1151.   }
  1152. }
  1153. layer {
  1154.   name: "Mconv7_stage2_L2"
  1155.   type: "Convolution"
  1156.   bottom: "Mconv6_stage2_L2"
  1157.   top: "Mconv7_stage2_L2"
  1158.   param {
  1159.     lr_mult: 4.0
  1160.     decay_mult: 1
  1161.   }
  1162.   param {
  1163.     lr_mult: 8.0
  1164.     decay_mult: 0
  1165.   }
  1166.   convolution_param {
  1167.     num_output: 19
  1168.     pad: 0
  1169.     kernel_size: 1
  1170.     weight_filler {
  1171.       type: "gaussian"
  1172.       std: 0.01
  1173.     }
  1174.     bias_filler {
  1175.       type: "constant"
  1176.     }
  1177.   }
  1178. }
  1179. layer {
  1180.   name: "concat_stage3"
  1181.   type: "Concat"
  1182.   bottom: "Mconv7_stage2_L1"
  1183.   bottom: "Mconv7_stage2_L2"
  1184.   bottom: "conv4_4_CPM"
  1185.   top: "concat_stage3"
  1186.   concat_param {
  1187.     axis: 1
  1188.   }
  1189. }
  1190. layer {
  1191.   name: "Mconv1_stage3_L1"
  1192.   type: "Convolution"
  1193.   bottom: "concat_stage3"
  1194.   top: "Mconv1_stage3_L1"
  1195.   param {
  1196.     lr_mult: 4.0
  1197.     decay_mult: 1
  1198.   }
  1199.   param {
  1200.     lr_mult: 8.0
  1201.     decay_mult: 0
  1202.   }
  1203.   convolution_param {
  1204.     num_output: 128
  1205.     pad: 3
  1206.     kernel_size: 7
  1207.     weight_filler {
  1208.       type: "gaussian"
  1209.       std: 0.01
  1210.     }
  1211.     bias_filler {
  1212.       type: "constant"
  1213.     }
  1214.   }
  1215. }
  1216. layer {
  1217.   name: "Mrelu1_stage3_L1"
  1218.   type: "ReLU"
  1219.   bottom: "Mconv1_stage3_L1"
  1220.   top: "Mconv1_stage3_L1"
  1221. }
  1222. layer {
  1223.   name: "Mconv1_stage3_L2"
  1224.   type: "Convolution"
  1225.   bottom: "concat_stage3"
  1226.   top: "Mconv1_stage3_L2"
  1227.   param {
  1228.     lr_mult: 4.0
  1229.     decay_mult: 1
  1230.   }
  1231.   param {
  1232.     lr_mult: 8.0
  1233.     decay_mult: 0
  1234.   }
  1235.   convolution_param {
  1236.     num_output: 128
  1237.     pad: 3
  1238.     kernel_size: 7
  1239.     weight_filler {
  1240.       type: "gaussian"
  1241.       std: 0.01
  1242.     }
  1243.     bias_filler {
  1244.       type: "constant"
  1245.     }
  1246.   }
  1247. }
  1248. layer {
  1249.   name: "Mrelu1_stage3_L2"
  1250.   type: "ReLU"
  1251.   bottom: "Mconv1_stage3_L2"
  1252.   top: "Mconv1_stage3_L2"
  1253. }
  1254. layer {
  1255.   name: "Mconv2_stage3_L1"
  1256.   type: "Convolution"
  1257.   bottom: "Mconv1_stage3_L1"
  1258.   top: "Mconv2_stage3_L1"
  1259.   param {
  1260.     lr_mult: 4.0
  1261.     decay_mult: 1
  1262.   }
  1263.   param {
  1264.     lr_mult: 8.0
  1265.     decay_mult: 0
  1266.   }
  1267.   convolution_param {
  1268.     num_output: 128
  1269.     pad: 3
  1270.     kernel_size: 7
  1271.     weight_filler {
  1272.       type: "gaussian"
  1273.       std: 0.01
  1274.     }
  1275.     bias_filler {
  1276.       type: "constant"
  1277.     }
  1278.   }
  1279. }
  1280. layer {
  1281.   name: "Mrelu2_stage3_L1"
  1282.   type: "ReLU"
  1283.   bottom: "Mconv2_stage3_L1"
  1284.   top: "Mconv2_stage3_L1"
  1285. }
  1286. layer {
  1287.   name: "Mconv2_stage3_L2"
  1288.   type: "Convolution"
  1289.   bottom: "Mconv1_stage3_L2"
  1290.   top: "Mconv2_stage3_L2"
  1291.   param {
  1292.     lr_mult: 4.0
  1293.     decay_mult: 1
  1294.   }
  1295.   param {
  1296.     lr_mult: 8.0
  1297.     decay_mult: 0
  1298.   }
  1299.   convolution_param {
  1300.     num_output: 128
  1301.     pad: 3
  1302.     kernel_size: 7
  1303.     weight_filler {
  1304.       type: "gaussian"
  1305.       std: 0.01
  1306.     }
  1307.     bias_filler {
  1308.       type: "constant"
  1309.     }
  1310.   }
  1311. }
  1312. layer {
  1313.   name: "Mrelu2_stage3_L2"
  1314.   type: "ReLU"
  1315.   bottom: "Mconv2_stage3_L2"
  1316.   top: "Mconv2_stage3_L2"
  1317. }
  1318. layer {
  1319.   name: "Mconv3_stage3_L1"
  1320.   type: "Convolution"
  1321.   bottom: "Mconv2_stage3_L1"
  1322.   top: "Mconv3_stage3_L1"
  1323.   param {
  1324.     lr_mult: 4.0
  1325.     decay_mult: 1
  1326.   }
  1327.   param {
  1328.     lr_mult: 8.0
  1329.     decay_mult: 0
  1330.   }
  1331.   convolution_param {
  1332.     num_output: 128
  1333.     pad: 3
  1334.     kernel_size: 7
  1335.     weight_filler {
  1336.       type: "gaussian"
  1337.       std: 0.01
  1338.     }
  1339.     bias_filler {
  1340.       type: "constant"
  1341.     }
  1342.   }
  1343. }
  1344. layer {
  1345.   name: "Mrelu3_stage3_L1"
  1346.   type: "ReLU"
  1347.   bottom: "Mconv3_stage3_L1"
  1348.   top: "Mconv3_stage3_L1"
  1349. }
  1350. layer {
  1351.   name: "Mconv3_stage3_L2"
  1352.   type: "Convolution"
  1353.   bottom: "Mconv2_stage3_L2"
  1354.   top: "Mconv3_stage3_L2"
  1355.   param {
  1356.     lr_mult: 4.0
  1357.     decay_mult: 1
  1358.   }
  1359.   param {
  1360.     lr_mult: 8.0
  1361.     decay_mult: 0
  1362.   }
  1363.   convolution_param {
  1364.     num_output: 128
  1365.     pad: 3
  1366.     kernel_size: 7
  1367.     weight_filler {
  1368.       type: "gaussian"
  1369.       std: 0.01
  1370.     }
  1371.     bias_filler {
  1372.       type: "constant"
  1373.     }
  1374.   }
  1375. }
  1376. layer {
  1377.   name: "Mrelu3_stage3_L2"
  1378.   type: "ReLU"
  1379.   bottom: "Mconv3_stage3_L2"
  1380.   top: "Mconv3_stage3_L2"
  1381. }
  1382. layer {
  1383.   name: "Mconv4_stage3_L1"
  1384.   type: "Convolution"
  1385.   bottom: "Mconv3_stage3_L1"
  1386.   top: "Mconv4_stage3_L1"
  1387.   param {
  1388.     lr_mult: 4.0
  1389.     decay_mult: 1
  1390.   }
  1391.   param {
  1392.     lr_mult: 8.0
  1393.     decay_mult: 0
  1394.   }
  1395.   convolution_param {
  1396.     num_output: 128
  1397.     pad: 3
  1398.     kernel_size: 7
  1399.     weight_filler {
  1400.       type: "gaussian"
  1401.       std: 0.01
  1402.     }
  1403.     bias_filler {
  1404.       type: "constant"
  1405.     }
  1406.   }
  1407. }
  1408. layer {
  1409.   name: "Mrelu4_stage3_L1"
  1410.   type: "ReLU"
  1411.   bottom: "Mconv4_stage3_L1"
  1412.   top: "Mconv4_stage3_L1"
  1413. }
  1414. layer {
  1415.   name: "Mconv4_stage3_L2"
  1416.   type: "Convolution"
  1417.   bottom: "Mconv3_stage3_L2"
  1418.   top: "Mconv4_stage3_L2"
  1419.   param {
  1420.     lr_mult: 4.0
  1421.     decay_mult: 1
  1422.   }
  1423.   param {
  1424.     lr_mult: 8.0
  1425.     decay_mult: 0
  1426.   }
  1427.   convolution_param {
  1428.     num_output: 128
  1429.     pad: 3
  1430.     kernel_size: 7
  1431.     weight_filler {
  1432.       type: "gaussian"
  1433.       std: 0.01
  1434.     }
  1435.     bias_filler {
  1436.       type: "constant"
  1437.     }
  1438.   }
  1439. }
  1440. layer {
  1441.   name: "Mrelu4_stage3_L2"
  1442.   type: "ReLU"
  1443.   bottom: "Mconv4_stage3_L2"
  1444.   top: "Mconv4_stage3_L2"
  1445. }
  1446. layer {
  1447.   name: "Mconv5_stage3_L1"
  1448.   type: "Convolution"
  1449.   bottom: "Mconv4_stage3_L1"
  1450.   top: "Mconv5_stage3_L1"
  1451.   param {
  1452.     lr_mult: 4.0
  1453.     decay_mult: 1
  1454.   }
  1455.   param {
  1456.     lr_mult: 8.0
  1457.     decay_mult: 0
  1458.   }
  1459.   convolution_param {
  1460.     num_output: 128
  1461.     pad: 3
  1462.     kernel_size: 7
  1463.     weight_filler {
  1464.       type: "gaussian"
  1465.       std: 0.01
  1466.     }
  1467.     bias_filler {
  1468.       type: "constant"
  1469.     }
  1470.   }
  1471. }
  1472. layer {
  1473.   name: "Mrelu5_stage3_L1"
  1474.   type: "ReLU"
  1475.   bottom: "Mconv5_stage3_L1"
  1476.   top: "Mconv5_stage3_L1"
  1477. }
  1478. layer {
  1479.   name: "Mconv5_stage3_L2"
  1480.   type: "Convolution"
  1481.   bottom: "Mconv4_stage3_L2"
  1482.   top: "Mconv5_stage3_L2"
  1483.   param {
  1484.     lr_mult: 4.0
  1485.     decay_mult: 1
  1486.   }
  1487.   param {
  1488.     lr_mult: 8.0
  1489.     decay_mult: 0
  1490.   }
  1491.   convolution_param {
  1492.     num_output: 128
  1493.     pad: 3
  1494.     kernel_size: 7
  1495.     weight_filler {
  1496.       type: "gaussian"
  1497.       std: 0.01
  1498.     }
  1499.     bias_filler {
  1500.       type: "constant"
  1501.     }
  1502.   }
  1503. }
  1504. layer {
  1505.   name: "Mrelu5_stage3_L2"
  1506.   type: "ReLU"
  1507.   bottom: "Mconv5_stage3_L2"
  1508.   top: "Mconv5_stage3_L2"
  1509. }
  1510. layer {
  1511.   name: "Mconv6_stage3_L1"
  1512.   type: "Convolution"
  1513.   bottom: "Mconv5_stage3_L1"
  1514.   top: "Mconv6_stage3_L1"
  1515.   param {
  1516.     lr_mult: 4.0
  1517.     decay_mult: 1
  1518.   }
  1519.   param {
  1520.     lr_mult: 8.0
  1521.     decay_mult: 0
  1522.   }
  1523.   convolution_param {
  1524.     num_output: 128
  1525.     pad: 0
  1526.     kernel_size: 1
  1527.     weight_filler {
  1528.       type: "gaussian"
  1529.       std: 0.01
  1530.     }
  1531.     bias_filler {
  1532.       type: "constant"
  1533.     }
  1534.   }
  1535. }
  1536. layer {
  1537.   name: "Mrelu6_stage3_L1"
  1538.   type: "ReLU"
  1539.   bottom: "Mconv6_stage3_L1"
  1540.   top: "Mconv6_stage3_L1"
  1541. }
  1542. layer {
  1543.   name: "Mconv6_stage3_L2"
  1544.   type: "Convolution"
  1545.   bottom: "Mconv5_stage3_L2"
  1546.   top: "Mconv6_stage3_L2"
  1547.   param {
  1548.     lr_mult: 4.0
  1549.     decay_mult: 1
  1550.   }
  1551.   param {
  1552.     lr_mult: 8.0
  1553.     decay_mult: 0
  1554.   }
  1555.   convolution_param {
  1556.     num_output: 128
  1557.     pad: 0
  1558.     kernel_size: 1
  1559.     weight_filler {
  1560.       type: "gaussian"
  1561.       std: 0.01
  1562.     }
  1563.     bias_filler {
  1564.       type: "constant"
  1565.     }
  1566.   }
  1567. }
  1568. layer {
  1569.   name: "Mrelu6_stage3_L2"
  1570.   type: "ReLU"
  1571.   bottom: "Mconv6_stage3_L2"
  1572.   top: "Mconv6_stage3_L2"
  1573. }
  1574. layer {
  1575.   name: "Mconv7_stage3_L1"
  1576.   type: "Convolution"
  1577.   bottom: "Mconv6_stage3_L1"
  1578.   top: "Mconv7_stage3_L1"
  1579.   param {
  1580.     lr_mult: 4.0
  1581.     decay_mult: 1
  1582.   }
  1583.   param {
  1584.     lr_mult: 8.0
  1585.     decay_mult: 0
  1586.   }
  1587.   convolution_param {
  1588.     num_output: 38
  1589.     pad: 0
  1590.     kernel_size: 1
  1591.     weight_filler {
  1592.       type: "gaussian"
  1593.       std: 0.01
  1594.     }
  1595.     bias_filler {
  1596.       type: "constant"
  1597.     }
  1598.   }
  1599. }
  1600. layer {
  1601.   name: "Mconv7_stage3_L2"
  1602.   type: "Convolution"
  1603.   bottom: "Mconv6_stage3_L2"
  1604.   top: "Mconv7_stage3_L2"
  1605.   param {
  1606.     lr_mult: 4.0
  1607.     decay_mult: 1
  1608.   }
  1609.   param {
  1610.     lr_mult: 8.0
  1611.     decay_mult: 0
  1612.   }
  1613.   convolution_param {
  1614.     num_output: 19
  1615.     pad: 0
  1616.     kernel_size: 1
  1617.     weight_filler {
  1618.       type: "gaussian"
  1619.       std: 0.01
  1620.     }
  1621.     bias_filler {
  1622.       type: "constant"
  1623.     }
  1624.   }
  1625. }
  1626. layer {
  1627.   name: "concat_stage4"
  1628.   type: "Concat"
  1629.   bottom: "Mconv7_stage3_L1"
  1630.   bottom: "Mconv7_stage3_L2"
  1631.   bottom: "conv4_4_CPM"
  1632.   top: "concat_stage4"
  1633.   concat_param {
  1634.     axis: 1
  1635.   }
  1636. }
  1637. layer {
  1638.   name: "Mconv1_stage4_L1"
  1639.   type: "Convolution"
  1640.   bottom: "concat_stage4"
  1641.   top: "Mconv1_stage4_L1"
  1642.   param {
  1643.     lr_mult: 4.0
  1644.     decay_mult: 1
  1645.   }
  1646.   param {
  1647.     lr_mult: 8.0
  1648.     decay_mult: 0
  1649.   }
  1650.   convolution_param {
  1651.     num_output: 128
  1652.     pad: 3
  1653.     kernel_size: 7
  1654.     weight_filler {
  1655.       type: "gaussian"
  1656.       std: 0.01
  1657.     }
  1658.     bias_filler {
  1659.       type: "constant"
  1660.     }
  1661.   }
  1662. }
  1663. layer {
  1664.   name: "Mrelu1_stage4_L1"
  1665.   type: "ReLU"
  1666.   bottom: "Mconv1_stage4_L1"
  1667.   top: "Mconv1_stage4_L1"
  1668. }
  1669. layer {
  1670.   name: "Mconv1_stage4_L2"
  1671.   type: "Convolution"
  1672.   bottom: "concat_stage4"
  1673.   top: "Mconv1_stage4_L2"
  1674.   param {
  1675.     lr_mult: 4.0
  1676.     decay_mult: 1
  1677.   }
  1678.   param {
  1679.     lr_mult: 8.0
  1680.     decay_mult: 0
  1681.   }
  1682.   convolution_param {
  1683.     num_output: 128
  1684.     pad: 3
  1685.     kernel_size: 7
  1686.     weight_filler {
  1687.       type: "gaussian"
  1688.       std: 0.01
  1689.     }
  1690.     bias_filler {
  1691.       type: "constant"
  1692.     }
  1693.   }
  1694. }
  1695. layer {
  1696.   name: "Mrelu1_stage4_L2"
  1697.   type: "ReLU"
  1698.   bottom: "Mconv1_stage4_L2"
  1699.   top: "Mconv1_stage4_L2"
  1700. }
  1701. layer {
  1702.   name: "Mconv2_stage4_L1"
  1703.   type: "Convolution"
  1704.   bottom: "Mconv1_stage4_L1"
  1705.   top: "Mconv2_stage4_L1"
  1706.   param {
  1707.     lr_mult: 4.0
  1708.     decay_mult: 1
  1709.   }
  1710.   param {
  1711.     lr_mult: 8.0
  1712.     decay_mult: 0
  1713.   }
  1714.   convolution_param {
  1715.     num_output: 128
  1716.     pad: 3
  1717.     kernel_size: 7
  1718.     weight_filler {
  1719.       type: "gaussian"
  1720.       std: 0.01
  1721.     }
  1722.     bias_filler {
  1723.       type: "constant"
  1724.     }
  1725.   }
  1726. }
  1727. layer {
  1728.   name: "Mrelu2_stage4_L1"
  1729.   type: "ReLU"
  1730.   bottom: "Mconv2_stage4_L1"
  1731.   top: "Mconv2_stage4_L1"
  1732. }
  1733. layer {
  1734.   name: "Mconv2_stage4_L2"
  1735.   type: "Convolution"
  1736.   bottom: "Mconv1_stage4_L2"
  1737.   top: "Mconv2_stage4_L2"
  1738.   param {
  1739.     lr_mult: 4.0
  1740.     decay_mult: 1
  1741.   }
  1742.   param {
  1743.     lr_mult: 8.0
  1744.     decay_mult: 0
  1745.   }
  1746.   convolution_param {
  1747.     num_output: 128
  1748.     pad: 3
  1749.     kernel_size: 7
  1750.     weight_filler {
  1751.       type: "gaussian"
  1752.       std: 0.01
  1753.     }
  1754.     bias_filler {
  1755.       type: "constant"
  1756.     }
  1757.   }
  1758. }
  1759. layer {
  1760.   name: "Mrelu2_stage4_L2"
  1761.   type: "ReLU"
  1762.   bottom: "Mconv2_stage4_L2"
  1763.   top: "Mconv2_stage4_L2"
  1764. }
  1765. layer {
  1766.   name: "Mconv3_stage4_L1"
  1767.   type: "Convolution"
  1768.   bottom: "Mconv2_stage4_L1"
  1769.   top: "Mconv3_stage4_L1"
  1770.   param {
  1771.     lr_mult: 4.0
  1772.     decay_mult: 1
  1773.   }
  1774.   param {
  1775.     lr_mult: 8.0
  1776.     decay_mult: 0
  1777.   }
  1778.   convolution_param {
  1779.     num_output: 128
  1780.     pad: 3
  1781.     kernel_size: 7
  1782.     weight_filler {
  1783.       type: "gaussian"
  1784.       std: 0.01
  1785.     }
  1786.     bias_filler {
  1787.       type: "constant"
  1788.     }
  1789.   }
  1790. }
  1791. layer {
  1792.   name: "Mrelu3_stage4_L1"
  1793.   type: "ReLU"
  1794.   bottom: "Mconv3_stage4_L1"
  1795.   top: "Mconv3_stage4_L1"
  1796. }
  1797. layer {
  1798.   name: "Mconv3_stage4_L2"
  1799.   type: "Convolution"
  1800.   bottom: "Mconv2_stage4_L2"
  1801.   top: "Mconv3_stage4_L2"
  1802.   param {
  1803.     lr_mult: 4.0
  1804.     decay_mult: 1
  1805.   }
  1806.   param {
  1807.     lr_mult: 8.0
  1808.     decay_mult: 0
  1809.   }
  1810.   convolution_param {
  1811.     num_output: 128
  1812.     pad: 3
  1813.     kernel_size: 7
  1814.     weight_filler {
  1815.       type: "gaussian"
  1816.       std: 0.01
  1817.     }
  1818.     bias_filler {
  1819.       type: "constant"
  1820.     }
  1821.   }
  1822. }
  1823. layer {
  1824.   name: "Mrelu3_stage4_L2"
  1825.   type: "ReLU"
  1826.   bottom: "Mconv3_stage4_L2"
  1827.   top: "Mconv3_stage4_L2"
  1828. }
  1829. layer {
  1830.   name: "Mconv4_stage4_L1"
  1831.   type: "Convolution"
  1832.   bottom: "Mconv3_stage4_L1"
  1833.   top: "Mconv4_stage4_L1"
  1834.   param {
  1835.     lr_mult: 4.0
  1836.     decay_mult: 1
  1837.   }
  1838.   param {
  1839.     lr_mult: 8.0
  1840.     decay_mult: 0
  1841.   }
  1842.   convolution_param {
  1843.     num_output: 128
  1844.     pad: 3
  1845.     kernel_size: 7
  1846.     weight_filler {
  1847.       type: "gaussian"
  1848.       std: 0.01
  1849.     }
  1850.     bias_filler {
  1851.       type: "constant"
  1852.     }
  1853.   }
  1854. }
  1855. layer {
  1856.   name: "Mrelu4_stage4_L1"
  1857.   type: "ReLU"
  1858.   bottom: "Mconv4_stage4_L1"
  1859.   top: "Mconv4_stage4_L1"
  1860. }
  1861. layer {
  1862.   name: "Mconv4_stage4_L2"
  1863.   type: "Convolution"
  1864.   bottom: "Mconv3_stage4_L2"
  1865.   top: "Mconv4_stage4_L2"
  1866.   param {
  1867.     lr_mult: 4.0
  1868.     decay_mult: 1
  1869.   }
  1870.   param {
  1871.     lr_mult: 8.0
  1872.     decay_mult: 0
  1873.   }
  1874.   convolution_param {
  1875.     num_output: 128
  1876.     pad: 3
  1877.     kernel_size: 7
  1878.     weight_filler {
  1879.       type: "gaussian"
  1880.       std: 0.01
  1881.     }
  1882.     bias_filler {
  1883.       type: "constant"
  1884.     }
  1885.   }
  1886. }
  1887. layer {
  1888.   name: "Mrelu4_stage4_L2"
  1889.   type: "ReLU"
  1890.   bottom: "Mconv4_stage4_L2"
  1891.   top: "Mconv4_stage4_L2"
  1892. }
  1893. layer {
  1894.   name: "Mconv5_stage4_L1"
  1895.   type: "Convolution"
  1896.   bottom: "Mconv4_stage4_L1"
  1897.   top: "Mconv5_stage4_L1"
  1898.   param {
  1899.     lr_mult: 4.0
  1900.     decay_mult: 1
  1901.   }
  1902.   param {
  1903.     lr_mult: 8.0
  1904.     decay_mult: 0
  1905.   }
  1906.   convolution_param {
  1907.     num_output: 128
  1908.     pad: 3
  1909.     kernel_size: 7
  1910.     weight_filler {
  1911.       type: "gaussian"
  1912.       std: 0.01
  1913.     }
  1914.     bias_filler {
  1915.       type: "constant"
  1916.     }
  1917.   }
  1918. }
  1919. layer {
  1920.   name: "Mrelu5_stage4_L1"
  1921.   type: "ReLU"
  1922.   bottom: "Mconv5_stage4_L1"
  1923.   top: "Mconv5_stage4_L1"
  1924. }
  1925. layer {
  1926.   name: "Mconv5_stage4_L2"
  1927.   type: "Convolution"
  1928.   bottom: "Mconv4_stage4_L2"
  1929.   top: "Mconv5_stage4_L2"
  1930.   param {
  1931.     lr_mult: 4.0
  1932.     decay_mult: 1
  1933.   }
  1934.   param {
  1935.     lr_mult: 8.0
  1936.     decay_mult: 0
  1937.   }
  1938.   convolution_param {
  1939.     num_output: 128
  1940.     pad: 3
  1941.     kernel_size: 7
  1942.     weight_filler {
  1943.       type: "gaussian"
  1944.       std: 0.01
  1945.     }
  1946.     bias_filler {
  1947.       type: "constant"
  1948.     }
  1949.   }
  1950. }
  1951. layer {
  1952.   name: "Mrelu5_stage4_L2"
  1953.   type: "ReLU"
  1954.   bottom: "Mconv5_stage4_L2"
  1955.   top: "Mconv5_stage4_L2"
  1956. }
  1957. layer {
  1958.   name: "Mconv6_stage4_L1"
  1959.   type: "Convolution"
  1960.   bottom: "Mconv5_stage4_L1"
  1961.   top: "Mconv6_stage4_L1"
  1962.   param {
  1963.     lr_mult: 4.0
  1964.     decay_mult: 1
  1965.   }
  1966.   param {
  1967.     lr_mult: 8.0
  1968.     decay_mult: 0
  1969.   }
  1970.   convolution_param {
  1971.     num_output: 128
  1972.     pad: 0
  1973.     kernel_size: 1
  1974.     weight_filler {
  1975.       type: "gaussian"
  1976.       std: 0.01
  1977.     }
  1978.     bias_filler {
  1979.       type: "constant"
  1980.     }
  1981.   }
  1982. }
  1983. layer {
  1984.   name: "Mrelu6_stage4_L1"
  1985.   type: "ReLU"
  1986.   bottom: "Mconv6_stage4_L1"
  1987.   top: "Mconv6_stage4_L1"
  1988. }
  1989. layer {
  1990.   name: "Mconv6_stage4_L2"
  1991.   type: "Convolution"
  1992.   bottom: "Mconv5_stage4_L2"
  1993.   top: "Mconv6_stage4_L2"
  1994.   param {
  1995.     lr_mult: 4.0
  1996.     decay_mult: 1
  1997.   }
  1998.   param {
  1999.     lr_mult: 8.0
  2000.     decay_mult: 0
  2001.   }
  2002.   convolution_param {
  2003.     num_output: 128
  2004.     pad: 0
  2005.     kernel_size: 1
  2006.     weight_filler {
  2007.       type: "gaussian"
  2008.       std: 0.01
  2009.     }
  2010.     bias_filler {
  2011.       type: "constant"
  2012.     }
  2013.   }
  2014. }
  2015. layer {
  2016.   name: "Mrelu6_stage4_L2"
  2017.   type: "ReLU"
  2018.   bottom: "Mconv6_stage4_L2"
  2019.   top: "Mconv6_stage4_L2"
  2020. }
  2021. layer {
  2022.   name: "Mconv7_stage4_L1"
  2023.   type: "Convolution"
  2024.   bottom: "Mconv6_stage4_L1"
  2025.   top: "Mconv7_stage4_L1"
  2026.   param {
  2027.     lr_mult: 4.0
  2028.     decay_mult: 1
  2029.   }
  2030.   param {
  2031.     lr_mult: 8.0
  2032.     decay_mult: 0
  2033.   }
  2034.   convolution_param {
  2035.     num_output: 38
  2036.     pad: 0
  2037.     kernel_size: 1
  2038.     weight_filler {
  2039.       type: "gaussian"
  2040.       std: 0.01
  2041.     }
  2042.     bias_filler {
  2043.       type: "constant"
  2044.     }
  2045.   }
  2046. }
  2047. layer {
  2048.   name: "Mconv7_stage4_L2"
  2049.   type: "Convolution"
  2050.   bottom: "Mconv6_stage4_L2"
  2051.   top: "Mconv7_stage4_L2"
  2052.   param {
  2053.     lr_mult: 4.0
  2054.     decay_mult: 1
  2055.   }
  2056.   param {
  2057.     lr_mult: 8.0
  2058.     decay_mult: 0
  2059.   }
  2060.   convolution_param {
  2061.     num_output: 19
  2062.     pad: 0
  2063.     kernel_size: 1
  2064.     weight_filler {
  2065.       type: "gaussian"
  2066.       std: 0.01
  2067.     }
  2068.     bias_filler {
  2069.       type: "constant"
  2070.     }
  2071.   }
  2072. }
  2073. layer {
  2074.   name: "concat_stage5"
  2075.   type: "Concat"
  2076.   bottom: "Mconv7_stage4_L1"
  2077.   bottom: "Mconv7_stage4_L2"
  2078.   bottom: "conv4_4_CPM"
  2079.   top: "concat_stage5"
  2080.   concat_param {
  2081.     axis: 1
  2082.   }
  2083. }
  2084. layer {
  2085.   name: "Mconv1_stage5_L1"
  2086.   type: "Convolution"
  2087.   bottom: "concat_stage5"
  2088.   top: "Mconv1_stage5_L1"
  2089.   param {
  2090.     lr_mult: 4.0
  2091.     decay_mult: 1
  2092.   }
  2093.   param {
  2094.     lr_mult: 8.0
  2095.     decay_mult: 0
  2096.   }
  2097.   convolution_param {
  2098.     num_output: 128
  2099.     pad: 3
  2100.     kernel_size: 7
  2101.     weight_filler {
  2102.       type: "gaussian"
  2103.       std: 0.01
  2104.     }
  2105.     bias_filler {
  2106.       type: "constant"
  2107.     }
  2108.   }
  2109. }
  2110. layer {
  2111.   name: "Mrelu1_stage5_L1"
  2112.   type: "ReLU"
  2113.   bottom: "Mconv1_stage5_L1"
  2114.   top: "Mconv1_stage5_L1"
  2115. }
  2116. layer {
  2117.   name: "Mconv1_stage5_L2"
  2118.   type: "Convolution"
  2119.   bottom: "concat_stage5"
  2120.   top: "Mconv1_stage5_L2"
  2121.   param {
  2122.     lr_mult: 4.0
  2123.     decay_mult: 1
  2124.   }
  2125.   param {
  2126.     lr_mult: 8.0
  2127.     decay_mult: 0
  2128.   }
  2129.   convolution_param {
  2130.     num_output: 128
  2131.     pad: 3
  2132.     kernel_size: 7
  2133.     weight_filler {
  2134.       type: "gaussian"
  2135.       std: 0.01
  2136.     }
  2137.     bias_filler {
  2138.       type: "constant"
  2139.     }
  2140.   }
  2141. }
  2142. layer {
  2143.   name: "Mrelu1_stage5_L2"
  2144.   type: "ReLU"
  2145.   bottom: "Mconv1_stage5_L2"
  2146.   top: "Mconv1_stage5_L2"
  2147. }
  2148. layer {
  2149.   name: "Mconv2_stage5_L1"
  2150.   type: "Convolution"
  2151.   bottom: "Mconv1_stage5_L1"
  2152.   top: "Mconv2_stage5_L1"
  2153.   param {
  2154.     lr_mult: 4.0
  2155.     decay_mult: 1
  2156.   }
  2157.   param {
  2158.     lr_mult: 8.0
  2159.     decay_mult: 0
  2160.   }
  2161.   convolution_param {
  2162.     num_output: 128
  2163.     pad: 3
  2164.     kernel_size: 7
  2165.     weight_filler {
  2166.       type: "gaussian"
  2167.       std: 0.01
  2168.     }
  2169.     bias_filler {
  2170.       type: "constant"
  2171.     }
  2172.   }
  2173. }
  2174. layer {
  2175.   name: "Mrelu2_stage5_L1"
  2176.   type: "ReLU"
  2177.   bottom: "Mconv2_stage5_L1"
  2178.   top: "Mconv2_stage5_L1"
  2179. }
  2180. layer {
  2181.   name: "Mconv2_stage5_L2"
  2182.   type: "Convolution"
  2183.   bottom: "Mconv1_stage5_L2"
  2184.   top: "Mconv2_stage5_L2"
  2185.   param {
  2186.     lr_mult: 4.0
  2187.     decay_mult: 1
  2188.   }
  2189.   param {
  2190.     lr_mult: 8.0
  2191.     decay_mult: 0
  2192.   }
  2193.   convolution_param {
  2194.     num_output: 128
  2195.     pad: 3
  2196.     kernel_size: 7
  2197.     weight_filler {
  2198.       type: "gaussian"
  2199.       std: 0.01
  2200.     }
  2201.     bias_filler {
  2202.       type: "constant"
  2203.     }
  2204.   }
  2205. }
  2206. layer {
  2207.   name: "Mrelu2_stage5_L2"
  2208.   type: "ReLU"
  2209.   bottom: "Mconv2_stage5_L2"
  2210.   top: "Mconv2_stage5_L2"
  2211. }
  2212. layer {
  2213.   name: "Mconv3_stage5_L1"
  2214.   type: "Convolution"
  2215.   bottom: "Mconv2_stage5_L1"
  2216.   top: "Mconv3_stage5_L1"
  2217.   param {
  2218.     lr_mult: 4.0
  2219.     decay_mult: 1
  2220.   }
  2221.   param {
  2222.     lr_mult: 8.0
  2223.     decay_mult: 0
  2224.   }
  2225.   convolution_param {
  2226.     num_output: 128
  2227.     pad: 3
  2228.     kernel_size: 7
  2229.     weight_filler {
  2230.       type: "gaussian"
  2231.       std: 0.01
  2232.     }
  2233.     bias_filler {
  2234.       type: "constant"
  2235.     }
  2236.   }
  2237. }
  2238. layer {
  2239.   name: "Mrelu3_stage5_L1"
  2240.   type: "ReLU"
  2241.   bottom: "Mconv3_stage5_L1"
  2242.   top: "Mconv3_stage5_L1"
  2243. }
  2244. layer {
  2245.   name: "Mconv3_stage5_L2"
  2246.   type: "Convolution"
  2247.   bottom: "Mconv2_stage5_L2"
  2248.   top: "Mconv3_stage5_L2"
  2249.   param {
  2250.     lr_mult: 4.0
  2251.     decay_mult: 1
  2252.   }
  2253.   param {
  2254.     lr_mult: 8.0
  2255.     decay_mult: 0
  2256.   }
  2257.   convolution_param {
  2258.     num_output: 128
  2259.     pad: 3
  2260.     kernel_size: 7
  2261.     weight_filler {
  2262.       type: "gaussian"
  2263.       std: 0.01
  2264.     }
  2265.     bias_filler {
  2266.       type: "constant"
  2267.     }
  2268.   }
  2269. }
  2270. layer {
  2271.   name: "Mrelu3_stage5_L2"
  2272.   type: "ReLU"
  2273.   bottom: "Mconv3_stage5_L2"
  2274.   top: "Mconv3_stage5_L2"
  2275. }
  2276. layer {
  2277.   name: "Mconv4_stage5_L1"
  2278.   type: "Convolution"
  2279.   bottom: "Mconv3_stage5_L1"
  2280.   top: "Mconv4_stage5_L1"
  2281.   param {
  2282.     lr_mult: 4.0
  2283.     decay_mult: 1
  2284.   }
  2285.   param {
  2286.     lr_mult: 8.0
  2287.     decay_mult: 0
  2288.   }
  2289.   convolution_param {
  2290.     num_output: 128
  2291.     pad: 3
  2292.     kernel_size: 7
  2293.     weight_filler {
  2294.       type: "gaussian"
  2295.       std: 0.01
  2296.     }
  2297.     bias_filler {
  2298.       type: "constant"
  2299.     }
  2300.   }
  2301. }
  2302. layer {
  2303.   name: "Mrelu4_stage5_L1"
  2304.   type: "ReLU"
  2305.   bottom: "Mconv4_stage5_L1"
  2306.   top: "Mconv4_stage5_L1"
  2307. }
  2308. layer {
  2309.   name: "Mconv4_stage5_L2"
  2310.   type: "Convolution"
  2311.   bottom: "Mconv3_stage5_L2"
  2312.   top: "Mconv4_stage5_L2"
  2313.   param {
  2314.     lr_mult: 4.0
  2315.     decay_mult: 1
  2316.   }
  2317.   param {
  2318.     lr_mult: 8.0
  2319.     decay_mult: 0
  2320.   }
  2321.   convolution_param {
  2322.     num_output: 128
  2323.     pad: 3
  2324.     kernel_size: 7
  2325.     weight_filler {
  2326.       type: "gaussian"
  2327.       std: 0.01
  2328.     }
  2329.     bias_filler {
  2330.       type: "constant"
  2331.     }
  2332.   }
  2333. }
  2334. layer {
  2335.   name: "Mrelu4_stage5_L2"
  2336.   type: "ReLU"
  2337.   bottom: "Mconv4_stage5_L2"
  2338.   top: "Mconv4_stage5_L2"
  2339. }
  2340. layer {
  2341.   name: "Mconv5_stage5_L1"
  2342.   type: "Convolution"
  2343.   bottom: "Mconv4_stage5_L1"
  2344.   top: "Mconv5_stage5_L1"
  2345.   param {
  2346.     lr_mult: 4.0
  2347.     decay_mult: 1
  2348.   }
  2349.   param {
  2350.     lr_mult: 8.0
  2351.     decay_mult: 0
  2352.   }
  2353.   convolution_param {
  2354.     num_output: 128
  2355.     pad: 3
  2356.     kernel_size: 7
  2357.     weight_filler {
  2358.       type: "gaussian"
  2359.       std: 0.01
  2360.     }
  2361.     bias_filler {
  2362.       type: "constant"
  2363.     }
  2364.   }
  2365. }
  2366. layer {
  2367.   name: "Mrelu5_stage5_L1"
  2368.   type: "ReLU"
  2369.   bottom: "Mconv5_stage5_L1"
  2370.   top: "Mconv5_stage5_L1"
  2371. }
  2372. layer {
  2373.   name: "Mconv5_stage5_L2"
  2374.   type: "Convolution"
  2375.   bottom: "Mconv4_stage5_L2"
  2376.   top: "Mconv5_stage5_L2"
  2377.   param {
  2378.     lr_mult: 4.0
  2379.     decay_mult: 1
  2380.   }
  2381.   param {
  2382.     lr_mult: 8.0
  2383.     decay_mult: 0
  2384.   }
  2385.   convolution_param {
  2386.     num_output: 128
  2387.     pad: 3
  2388.     kernel_size: 7
  2389.     weight_filler {
  2390.       type: "gaussian"
  2391.       std: 0.01
  2392.     }
  2393.     bias_filler {
  2394.       type: "constant"
  2395.     }
  2396.   }
  2397. }
  2398. layer {
  2399.   name: "Mrelu5_stage5_L2"
  2400.   type: "ReLU"
  2401.   bottom: "Mconv5_stage5_L2"
  2402.   top: "Mconv5_stage5_L2"
  2403. }
  2404. layer {
  2405.   name: "Mconv6_stage5_L1"
  2406.   type: "Convolution"
  2407.   bottom: "Mconv5_stage5_L1"
  2408.   top: "Mconv6_stage5_L1"
  2409.   param {
  2410.     lr_mult: 4.0
  2411.     decay_mult: 1
  2412.   }
  2413.   param {
  2414.     lr_mult: 8.0
  2415.     decay_mult: 0
  2416.   }
  2417.   convolution_param {
  2418.     num_output: 128
  2419.     pad: 0
  2420.     kernel_size: 1
  2421.     weight_filler {
  2422.       type: "gaussian"
  2423.       std: 0.01
  2424.     }
  2425.     bias_filler {
  2426.       type: "constant"
  2427.     }
  2428.   }
  2429. }
  2430. layer {
  2431.   name: "Mrelu6_stage5_L1"
  2432.   type: "ReLU"
  2433.   bottom: "Mconv6_stage5_L1"
  2434.   top: "Mconv6_stage5_L1"
  2435. }
  2436. layer {
  2437.   name: "Mconv6_stage5_L2"
  2438.   type: "Convolution"
  2439.   bottom: "Mconv5_stage5_L2"
  2440.   top: "Mconv6_stage5_L2"
  2441.   param {
  2442.     lr_mult: 4.0
  2443.     decay_mult: 1
  2444.   }
  2445.   param {
  2446.     lr_mult: 8.0
  2447.     decay_mult: 0
  2448.   }
  2449.   convolution_param {
  2450.     num_output: 128
  2451.     pad: 0
  2452.     kernel_size: 1
  2453.     weight_filler {
  2454.       type: "gaussian"
  2455.       std: 0.01
  2456.     }
  2457.     bias_filler {
  2458.       type: "constant"
  2459.     }
  2460.   }
  2461. }
  2462. layer {
  2463.   name: "Mrelu6_stage5_L2"
  2464.   type: "ReLU"
  2465.   bottom: "Mconv6_stage5_L2"
  2466.   top: "Mconv6_stage5_L2"
  2467. }
  2468. layer {
  2469.   name: "Mconv7_stage5_L1"
  2470.   type: "Convolution"
  2471.   bottom: "Mconv6_stage5_L1"
  2472.   top: "Mconv7_stage5_L1"
  2473.   param {
  2474.     lr_mult: 4.0
  2475.     decay_mult: 1
  2476.   }
  2477.   param {
  2478.     lr_mult: 8.0
  2479.     decay_mult: 0
  2480.   }
  2481.   convolution_param {
  2482.     num_output: 38
  2483.     pad: 0
  2484.     kernel_size: 1
  2485.     weight_filler {
  2486.       type: "gaussian"
  2487.       std: 0.01
  2488.     }
  2489.     bias_filler {
  2490.       type: "constant"
  2491.     }
  2492.   }
  2493. }
  2494. layer {
  2495.   name: "Mconv7_stage5_L2"
  2496.   type: "Convolution"
  2497.   bottom: "Mconv6_stage5_L2"
  2498.   top: "Mconv7_stage5_L2"
  2499.   param {
  2500.     lr_mult: 4.0
  2501.     decay_mult: 1
  2502.   }
  2503.   param {
  2504.     lr_mult: 8.0
  2505.     decay_mult: 0
  2506.   }
  2507.   convolution_param {
  2508.     num_output: 19
  2509.     pad: 0
  2510.     kernel_size: 1
  2511.     weight_filler {
  2512.       type: "gaussian"
  2513.       std: 0.01
  2514.     }
  2515.     bias_filler {
  2516.       type: "constant"
  2517.     }
  2518.   }
  2519. }
  2520. layer {
  2521.   name: "concat_stage6"
  2522.   type: "Concat"
  2523.   bottom: "Mconv7_stage5_L1"
  2524.   bottom: "Mconv7_stage5_L2"
  2525.   bottom: "conv4_4_CPM"
  2526.   top: "concat_stage6"
  2527.   concat_param {
  2528.     axis: 1
  2529.   }
  2530. }
  2531. layer {
  2532.   name: "Mconv1_stage6_L1"
  2533.   type: "Convolution"
  2534.   bottom: "concat_stage6"
  2535.   top: "Mconv1_stage6_L1"
  2536.   param {
  2537.     lr_mult: 4.0
  2538.     decay_mult: 1
  2539.   }
  2540.   param {
  2541.     lr_mult: 8.0
  2542.     decay_mult: 0
  2543.   }
  2544.   convolution_param {
  2545.     num_output: 128
  2546.     pad: 3
  2547.     kernel_size: 7
  2548.     weight_filler {
  2549.       type: "gaussian"
  2550.       std: 0.01
  2551.     }
  2552.     bias_filler {
  2553.       type: "constant"
  2554.     }
  2555.   }
  2556. }
  2557. layer {
  2558.   name: "Mrelu1_stage6_L1"
  2559.   type: "ReLU"
  2560.   bottom: "Mconv1_stage6_L1"
  2561.   top: "Mconv1_stage6_L1"
  2562. }
  2563. layer {
  2564.   name: "Mconv1_stage6_L2"
  2565.   type: "Convolution"
  2566.   bottom: "concat_stage6"
  2567.   top: "Mconv1_stage6_L2"
  2568.   param {
  2569.     lr_mult: 4.0
  2570.     decay_mult: 1
  2571.   }
  2572.   param {
  2573.     lr_mult: 8.0
  2574.     decay_mult: 0
  2575.   }
  2576.   convolution_param {
  2577.     num_output: 128
  2578.     pad: 3
  2579.     kernel_size: 7
  2580.     weight_filler {
  2581.       type: "gaussian"
  2582.       std: 0.01
  2583.     }
  2584.     bias_filler {
  2585.       type: "constant"
  2586.     }
  2587.   }
  2588. }
  2589. layer {
  2590.   name: "Mrelu1_stage6_L2"
  2591.   type: "ReLU"
  2592.   bottom: "Mconv1_stage6_L2"
  2593.   top: "Mconv1_stage6_L2"
  2594. }
  2595. layer {
  2596.   name: "Mconv2_stage6_L1"
  2597.   type: "Convolution"
  2598.   bottom: "Mconv1_stage6_L1"
  2599.   top: "Mconv2_stage6_L1"
  2600.   param {
  2601.     lr_mult: 4.0
  2602.     decay_mult: 1
  2603.   }
  2604.   param {
  2605.     lr_mult: 8.0
  2606.     decay_mult: 0
  2607.   }
  2608.   convolution_param {
  2609.     num_output: 128
  2610.     pad: 3
  2611.     kernel_size: 7
  2612.     weight_filler {
  2613.       type: "gaussian"
  2614.       std: 0.01
  2615.     }
  2616.     bias_filler {
  2617.       type: "constant"
  2618.     }
  2619.   }
  2620. }
  2621. layer {
  2622.   name: "Mrelu2_stage6_L1"
  2623.   type: "ReLU"
  2624.   bottom: "Mconv2_stage6_L1"
  2625.   top: "Mconv2_stage6_L1"
  2626. }
  2627. layer {
  2628.   name: "Mconv2_stage6_L2"
  2629.   type: "Convolution"
  2630.   bottom: "Mconv1_stage6_L2"
  2631.   top: "Mconv2_stage6_L2"
  2632.   param {
  2633.     lr_mult: 4.0
  2634.     decay_mult: 1
  2635.   }
  2636.   param {
  2637.     lr_mult: 8.0
  2638.     decay_mult: 0
  2639.   }
  2640.   convolution_param {
  2641.     num_output: 128
  2642.     pad: 3
  2643.     kernel_size: 7
  2644.     weight_filler {
  2645.       type: "gaussian"
  2646.       std: 0.01
  2647.     }
  2648.     bias_filler {
  2649.       type: "constant"
  2650.     }
  2651.   }
  2652. }
  2653. layer {
  2654.   name: "Mrelu2_stage6_L2"
  2655.   type: "ReLU"
  2656.   bottom: "Mconv2_stage6_L2"
  2657.   top: "Mconv2_stage6_L2"
  2658. }
  2659. layer {
  2660.   name: "Mconv3_stage6_L1"
  2661.   type: "Convolution"
  2662.   bottom: "Mconv2_stage6_L1"
  2663.   top: "Mconv3_stage6_L1"
  2664.   param {
  2665.     lr_mult: 4.0
  2666.     decay_mult: 1
  2667.   }
  2668.   param {
  2669.     lr_mult: 8.0
  2670.     decay_mult: 0
  2671.   }
  2672.   convolution_param {
  2673.     num_output: 128
  2674.     pad: 3
  2675.     kernel_size: 7
  2676.     weight_filler {
  2677.       type: "gaussian"
  2678.       std: 0.01
  2679.     }
  2680.     bias_filler {
  2681.       type: "constant"
  2682.     }
  2683.   }
  2684. }
  2685. layer {
  2686.   name: "Mrelu3_stage6_L1"
  2687.   type: "ReLU"
  2688.   bottom: "Mconv3_stage6_L1"
  2689.   top: "Mconv3_stage6_L1"
  2690. }
  2691. layer {
  2692.   name: "Mconv3_stage6_L2"
  2693.   type: "Convolution"
  2694.   bottom: "Mconv2_stage6_L2"
  2695.   top: "Mconv3_stage6_L2"
  2696.   param {
  2697.     lr_mult: 4.0
  2698.     decay_mult: 1
  2699.   }
  2700.   param {
  2701.     lr_mult: 8.0
  2702.     decay_mult: 0
  2703.   }
  2704.   convolution_param {
  2705.     num_output: 128
  2706.     pad: 3
  2707.     kernel_size: 7
  2708.     weight_filler {
  2709.       type: "gaussian"
  2710.       std: 0.01
  2711.     }
  2712.     bias_filler {
  2713.       type: "constant"
  2714.     }
  2715.   }
  2716. }
  2717. layer {
  2718.   name: "Mrelu3_stage6_L2"
  2719.   type: "ReLU"
  2720.   bottom: "Mconv3_stage6_L2"
  2721.   top: "Mconv3_stage6_L2"
  2722. }
  2723. layer {
  2724.   name: "Mconv4_stage6_L1"
  2725.   type: "Convolution"
  2726.   bottom: "Mconv3_stage6_L1"
  2727.   top: "Mconv4_stage6_L1"
  2728.   param {
  2729.     lr_mult: 4.0
  2730.     decay_mult: 1
  2731.   }
  2732.   param {
  2733.     lr_mult: 8.0
  2734.     decay_mult: 0
  2735.   }
  2736.   convolution_param {
  2737.     num_output: 128
  2738.     pad: 3
  2739.     kernel_size: 7
  2740.     weight_filler {
  2741.       type: "gaussian"
  2742.       std: 0.01
  2743.     }
  2744.     bias_filler {
  2745.       type: "constant"
  2746.     }
  2747.   }
  2748. }
  2749. layer {
  2750.   name: "Mrelu4_stage6_L1"
  2751.   type: "ReLU"
  2752.   bottom: "Mconv4_stage6_L1"
  2753.   top: "Mconv4_stage6_L1"
  2754. }
  2755. layer {
  2756.   name: "Mconv4_stage6_L2"
  2757.   type: "Convolution"
  2758.   bottom: "Mconv3_stage6_L2"
  2759.   top: "Mconv4_stage6_L2"
  2760.   param {
  2761.     lr_mult: 4.0
  2762.     decay_mult: 1
  2763.   }
  2764.   param {
  2765.     lr_mult: 8.0
  2766.     decay_mult: 0
  2767.   }
  2768.   convolution_param {
  2769.     num_output: 128
  2770.     pad: 3
  2771.     kernel_size: 7
  2772.     weight_filler {
  2773.       type: "gaussian"
  2774.       std: 0.01
  2775.     }
  2776.     bias_filler {
  2777.       type: "constant"
  2778.     }
  2779.   }
  2780. }
  2781. layer {
  2782.   name: "Mrelu4_stage6_L2"
  2783.   type: "ReLU"
  2784.   bottom: "Mconv4_stage6_L2"
  2785.   top: "Mconv4_stage6_L2"
  2786. }
  2787. layer {
  2788.   name: "Mconv5_stage6_L1"
  2789.   type: "Convolution"
  2790.   bottom: "Mconv4_stage6_L1"
  2791.   top: "Mconv5_stage6_L1"
  2792.   param {
  2793.     lr_mult: 4.0
  2794.     decay_mult: 1
  2795.   }
  2796.   param {
  2797.     lr_mult: 8.0
  2798.     decay_mult: 0
  2799.   }
  2800.   convolution_param {
  2801.     num_output: 128
  2802.     pad: 3
  2803.     kernel_size: 7
  2804.     weight_filler {
  2805.       type: "gaussian"
  2806.       std: 0.01
  2807.     }
  2808.     bias_filler {
  2809.       type: "constant"
  2810.     }
  2811.   }
  2812. }
  2813. layer {
  2814.   name: "Mrelu5_stage6_L1"
  2815.   type: "ReLU"
  2816.   bottom: "Mconv5_stage6_L1"
  2817.   top: "Mconv5_stage6_L1"
  2818. }
  2819. layer {
  2820.   name: "Mconv5_stage6_L2"
  2821.   type: "Convolution"
  2822.   bottom: "Mconv4_stage6_L2"
  2823.   top: "Mconv5_stage6_L2"
  2824.   param {
  2825.     lr_mult: 4.0
  2826.     decay_mult: 1
  2827.   }
  2828.   param {
  2829.     lr_mult: 8.0
  2830.     decay_mult: 0
  2831.   }
  2832.   convolution_param {
  2833.     num_output: 128
  2834.     pad: 3
  2835.     kernel_size: 7
  2836.     weight_filler {
  2837.       type: "gaussian"
  2838.       std: 0.01
  2839.     }
  2840.     bias_filler {
  2841.       type: "constant"
  2842.     }
  2843.   }
  2844. }
  2845. layer {
  2846.   name: "Mrelu5_stage6_L2"
  2847.   type: "ReLU"
  2848.   bottom: "Mconv5_stage6_L2"
  2849.   top: "Mconv5_stage6_L2"
  2850. }
  2851. layer {
  2852.   name: "Mconv6_stage6_L1"
  2853.   type: "Convolution"
  2854.   bottom: "Mconv5_stage6_L1"
  2855.   top: "Mconv6_stage6_L1"
  2856.   param {
  2857.     lr_mult: 4.0
  2858.     decay_mult: 1
  2859.   }
  2860.   param {
  2861.     lr_mult: 8.0
  2862.     decay_mult: 0
  2863.   }
  2864.   convolution_param {
  2865.     num_output: 128
  2866.     pad: 0
  2867.     kernel_size: 1
  2868.     weight_filler {
  2869.       type: "gaussian"
  2870.       std: 0.01
  2871.     }
  2872.     bias_filler {
  2873.       type: "constant"
  2874.     }
  2875.   }
  2876. }
  2877. layer {
  2878.   name: "Mrelu6_stage6_L1"
  2879.   type: "ReLU"
  2880.   bottom: "Mconv6_stage6_L1"
  2881.   top: "Mconv6_stage6_L1"
  2882. }
  2883. layer {
  2884.   name: "Mconv6_stage6_L2"
  2885.   type: "Convolution"
  2886.   bottom: "Mconv5_stage6_L2"
  2887.   top: "Mconv6_stage6_L2"
  2888.   param {
  2889.     lr_mult: 4.0
  2890.     decay_mult: 1
  2891.   }
  2892.   param {
  2893.     lr_mult: 8.0
  2894.     decay_mult: 0
  2895.   }
  2896.   convolution_param {
  2897.     num_output: 128
  2898.     pad: 0
  2899.     kernel_size: 1
  2900.     weight_filler {
  2901.       type: "gaussian"
  2902.       std: 0.01
  2903.     }
  2904.     bias_filler {
  2905.       type: "constant"
  2906.     }
  2907.   }
  2908. }
  2909. layer {
  2910.   name: "Mrelu6_stage6_L2"
  2911.   type: "ReLU"
  2912.   bottom: "Mconv6_stage6_L2"
  2913.   top: "Mconv6_stage6_L2"
  2914. }
  2915. layer {
  2916.   name: "Mconv7_stage6_L1"
  2917.   type: "Convolution"
  2918.   bottom: "Mconv6_stage6_L1"
  2919.   top: "Mconv7_stage6_L1"
  2920.   param {
  2921.     lr_mult: 4.0
  2922.     decay_mult: 1
  2923.   }
  2924.   param {
  2925.     lr_mult: 8.0
  2926.     decay_mult: 0
  2927.   }
  2928.   convolution_param {
  2929.     num_output: 38
  2930.     pad: 0
  2931.     kernel_size: 1
  2932.     weight_filler {
  2933.       type: "gaussian"
  2934.       std: 0.01
  2935.     }
  2936.     bias_filler {
  2937.       type: "constant"
  2938.     }
  2939.   }
  2940. }
  2941. layer {
  2942.   name: "Mconv7_stage6_L2"
  2943.   type: "Convolution"
  2944.   bottom: "Mconv6_stage6_L2"
  2945.   top: "Mconv7_stage6_L2"
  2946.   param {
  2947.     lr_mult: 4.0
  2948.     decay_mult: 1
  2949.   }
  2950.   param {
  2951.     lr_mult: 8.0
  2952.     decay_mult: 0
  2953.   }
  2954.   convolution_param {
  2955.     num_output: 19
  2956.     pad: 0
  2957.     kernel_size: 1
  2958.     weight_filler {
  2959.       type: "gaussian"
  2960.       std: 0.01
  2961.     }
  2962.     bias_filler {
  2963.       type: "constant"
  2964.     }
  2965.   }
  2966. }
  2967. layer {
  2968.   name: "concat_stage7"
  2969.   type: "Concat"
  2970.   bottom: "Mconv7_stage6_L2"
  2971.   bottom: "Mconv7_stage6_L1"
  2972.   # top: "concat_stage7"
  2973.   top: "net_output"
  2974.   concat_param {
  2975.     axis: 1
  2976.   }
  2977. }
RAW Paste Data
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy. OK, I Understand
 
Top