封装ResNet27的TensorRT外围接口

平台：TX2 ，JetPack3.3

最近在使用TensorRT封装27层的ResNet，封装过程中遇到以下问题：

1.网络结构中的Prelu类型的GPU代码实现;

2.在比对网络结构fc5的输出的时候，输出特征参数只有一半是正确，有一般为0;

首先分享一下网络结构的改造：

 

原始网络结构：

name: "face_res27net"
input: "data"
input_dim: 1
input_dim: 3
input_dim: 128
input_dim: 128
layer {
  name: "conv1a"
  type: "Convolution"
  bottom: "data"
  top: "conv1a"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1a"
  type: "PReLU"
  bottom: "conv1a"
  top: "conv1a"
}
layer {
  name: "conv1b"
  type: "Convolution"
  bottom: "conv1a"
  top: "conv1b"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1b"
  type: "PReLU"
  bottom: "conv1b"
  top: "conv1b"
}
layer {
  name: "pool1b"
  type: "Pooling"
  bottom: "conv1b"
  top: "pool1b"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2 
  }
}
layer {
  name: "conv2_1"
  type: "Convolution"
  bottom: "pool1b"
  top: "conv2_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2_1"
  type: "PReLU"
  bottom: "conv2_1"
  top: "conv2_1"
}
layer {
  name: "conv2_2"
  type: "Convolution"
  bottom: "conv2_1"
  top: "conv2_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2_2"
  type: "PReLU"
  bottom: "conv2_2"
  top: "conv2_2"
}
layer {
  name: "res2_2"
  type: "Eltwise"
  bottom: "pool1b"
  bottom: "conv2_2"
  top: "res2_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "res2_2"
  top: "conv2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2"
  type: "PReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2 
  }
}
layer {
  name: "conv3_1"
  type: "Convolution"
  bottom: "pool2"
  top: "conv3_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_1"
  type: "PReLU"
  bottom: "conv3_1"
  top: "conv3_1"
}
layer {
  name: "conv3_2"
  type: "Convolution"
  bottom: "conv3_1"
  top: "conv3_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_2"
  type: "PReLU"
  bottom: "conv3_2"
  top: "conv3_2"
}
layer {
  name: "res3_2"
  type: "Eltwise"
  bottom: "pool2"
  bottom: "conv3_2"
  top: "res3_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv3_3"
  type: "Convolution"
  bottom: "res3_2"
  top: "conv3_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_3"
  type: "PReLU"
  bottom: "conv3_3"
  top: "conv3_3"
}
layer {
  name: "conv3_4"
  type: "Convolution"
  bottom: "conv3_3"
  top: "conv3_4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_4"
  type: "PReLU"
  bottom: "conv3_4"
  top: "conv3_4"
}
layer {
  name: "res3_4"
  type: "Eltwise"
  bottom: "res3_2"
  bottom: "conv3_4"
  top: "res3_4"
  eltwise_param { 
    operation: 1
  }
}

layer {
  name: "conv3"
  type: "Convolution"
  bottom: "res3_4"
  top: "conv3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3"
  type: "PReLU"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "conv3"
  top: "pool3"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "conv4_1"
  type: "Convolution"
  bottom: "pool3"
  top: "conv4_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_1"
  type: "PReLU"
  bottom: "conv4_1"
  top: "conv4_1"
}
layer {
  name: "conv4_2"
  type: "Convolution"
  bottom: "conv4_1"
  top: "conv4_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_2"
  type: "PReLU"
  bottom: "conv4_2"
  top: "conv4_2"
}
layer {
  name: "res4_2"
  type: "Eltwise"
  bottom: "pool3"
  bottom: "conv4_2"
  top: "res4_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_3"
  type: "Convolution"
  bottom: "res4_2"
  top: "conv4_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_3"
  type: "PReLU"
  bottom: "conv4_3"
  top: "conv4_3"
}
layer {
  name: "conv4_4"
  type: "Convolution"
  bottom: "conv4_3"
  top: "conv4_4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_4"
  type: "PReLU"
  bottom: "conv4_4"
  top: "conv4_4"
}
layer {
  name: "res4_4"
  type: "Eltwise"
  bottom: "res4_2"
  bottom: "conv4_4"
  top: "res4_4"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_5"
  type: "Convolution"
  bottom: "res4_4"
  top: "conv4_5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_5"
  type: "PReLU"
  bottom: "conv4_5"
  top: "conv4_5"
}
layer {
  name: "conv4_6"
  type: "Convolution"
  bottom: "conv4_5"
  top: "conv4_6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_6"
  type: "PReLU"
  bottom: "conv4_6"
  top: "conv4_6"
}
layer {
  name: "res4_6"
  type: "Eltwise"
  bottom: "res4_4"
  bottom: "conv4_6"
  top: "res4_6"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_7"
  type: "Convolution"
  bottom: "res4_6"
  top: "conv4_7"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_7"
  type: "PReLU"
  bottom: "conv4_7"
  top: "conv4_7"
}
layer {
  name: "conv4_8"
  type: "Convolution"
  bottom: "conv4_7"
  top: "conv4_8"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_8"
  type: "PReLU"
  bottom: "conv4_8"
  top: "conv4_8"
}
layer {
  name: "res4_8"
  type: "Eltwise"
  bottom: "res4_6"
  bottom: "conv4_8"
  top: "res4_8"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_9"
  type: "Convolution"
  bottom: "res4_8"
  top: "conv4_9"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_9"
  type: "PReLU"
  bottom: "conv4_9"
  top: "conv4_9"
}
layer {
  name: "conv4_10"
  type: "Convolution"
  bottom: "conv4_9"
  top: "conv4_10"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_10"
  type: "PReLU"
  bottom: "conv4_10"
  top: "conv4_10"
}
layer {
  name: "res4_10"
  type: "Eltwise"
  bottom: "res4_8"
  bottom: "conv4_10"
  top: "res4_10"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4"
  type: "Convolution"
  bottom: "res4_10"
  top: "conv4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4"
  type: "PReLU"
  bottom: "conv4"
  top: "conv4"
}
layer {
  name: "pool4"
  type: "Pooling"
  bottom: "conv4"
  top: "pool4"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "conv5_1"
  type: "Convolution"
  bottom: "pool4"
  top: "conv5_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_1"
  type: "PReLU"
  bottom: "conv5_1"
  top: "conv5_1"
}
layer {
  name: "conv5_2"
  type: "Convolution"
  bottom: "conv5_1"
  top: "conv5_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_2"
  type: "PReLU"
  bottom: "conv5_2"
  top: "conv5_2"
}
layer {
  name: "res5_2"
  type: "Eltwise"
  bottom: "pool4"
  bottom: "conv5_2"
  top: "res5_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv5_3"
  type: "Convolution"
  bottom: "res5_2"
  top: "conv5_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_3"
  type: "PReLU"
  bottom: "conv5_3"
  top: "conv5_3"
}
layer {
  name: "conv5_4"
  type: "Convolution"
  bottom: "conv5_3"
  top: "conv5_4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_4"
  type: "PReLU"
  bottom: "conv5_4"
  top: "conv5_4"
}
layer {
  name: "res5_4"
  type: "Eltwise"
  bottom: "res5_2"
  bottom: "conv5_4"
  top: "res5_4"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv5_5"
  type: "Convolution"
  bottom: "res5_4"
  top: "conv5_5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_5"
  type: "PReLU"
  bottom: "conv5_5"
  top: "conv5_5"
}
layer {
  name: "conv5_6"
  type: "Convolution"
  bottom: "conv5_5"
  top: "conv5_6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_6"
  type: "PReLU"
  bottom: "conv5_6"
  top: "conv5_6"
}
layer {
  name: "res5_6"
  type: "Eltwise"
  bottom: "res5_4"
  bottom: "conv5_6"
  top: "res5_6"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "fc5"
  type: "InnerProduct"
  bottom: "res5_6"
  top: "fc5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

针对上面的原始网络结构，需要对其进行更改，更改Prelu层，为什么要对Prelu层进行修改，原因：Prelu是使用caffe中的gpu源码（尽管nvidia官网说，已经支持Prelu，但是我没有找到相应的使用方法;同时网上也有人说nvidia所支持的Prelu，实质是Lrelu，根本没有实现Prelu。所以自己在Plugin中加入了相应的Prelu的实现）。

更改后是这样的：

name: "face_res27net"
input: "data"
input_dim: 1
input_dim: 3
input_dim: 128
input_dim: 128
layer {
  name: "conv1a"
  type: "Convolution"
  bottom: "data"
  top: "conv1a"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1a"
  type: "IPlugin"
  bottom: "conv1a"
  top: "conv1a"
}
layer {
  name: "conv1b"
  type: "Convolution"
  bottom: "conv1a"
  top: "conv1b"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1b"
  type: "IPlugin"
  bottom: "conv1b"
  top: "conv1b"
}
layer {
  name: "pool1b"
  type: "Pooling"
  bottom: "conv1b"
  top: "pool1b"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2 
  }
}
layer {
  name: "conv2_1"
  type: "Convolution"
  bottom: "pool1b"
  top: "conv2_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2_1"
  type: "IPlugin"
  bottom: "conv2_1"
  top: "conv2_1"
}
layer {
  name: "conv2_2"
  type: "Convolution"
  bottom: "conv2_1"
  top: "conv2_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2_2"
  type: "IPlugin"
  bottom: "conv2_2"
  top: "conv2_2"
}
layer {
  name: "res2_2"
  type: "Eltwise"
  bottom: "pool1b"
  bottom: "conv2_2"
  top: "res2_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "res2_2"
  top: "conv2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2"
  type: "IPlugin"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2 
  }
}
layer {
  name: "conv3_1"
  type: "Convolution"
  bottom: "pool2"
  top: "conv3_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_1"
  type: "IPlugin"
  bottom: "conv3_1"
  top: "conv3_1"
}
layer {
  name: "conv3_2"
  type: "Convolution"
  bottom: "conv3_1"
  top: "conv3_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_2"
  type: "IPlugin"
  bottom: "conv3_2"
  top: "conv3_2"
}
layer {
  name: "res3_2"
  type: "Eltwise"
  bottom: "pool2"
  bottom: "conv3_2"
  top: "res3_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv3_3"
  type: "Convolution"
  bottom: "res3_2"
  top: "conv3_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_3"
  type: "IPlugin"
  bottom: "conv3_3"
  top: "conv3_3"
}
layer {
  name: "conv3_4"
  type: "Convolution"
  bottom: "conv3_3"
  top: "conv3_4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_4"
  type: "IPlugin"
  bottom: "conv3_4"
  top: "conv3_4"
}
layer {
  name: "res3_4"
  type: "Eltwise"
  bottom: "res3_2"
  bottom: "conv3_4"
  top: "res3_4"
  eltwise_param { 
    operation: 1
  }
}

layer {
  name: "conv3"
  type: "Convolution"
  bottom: "res3_4"
  top: "conv3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3"
  type: "IPlugin"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "conv3"
  top: "pool3"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "conv4_1"
  type: "Convolution"
  bottom: "pool3"
  top: "conv4_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_1"
  type: "IPlugin"
  bottom: "conv4_1"
  top: "conv4_1"
}
layer {
  name: "conv4_2"
  type: "Convolution"
  bottom: "conv4_1"
  top: "conv4_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_2"
  type: "IPlugin"
  bottom: "conv4_2"
  top: "conv4_2"
}
layer {
  name: "res4_2"
  type: "Eltwise"
  bottom: "pool3"
  bottom: "conv4_2"
  top: "res4_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_3"
  type: "Convolution"
  bottom: "res4_2"
  top: "conv4_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_3"
  type: "IPlugin"
  bottom: "conv4_3"
  top: "conv4_3"
}
layer {
  name: "conv4_4"
  type: "Convolution"
  bottom: "conv4_3"
  top: "conv4_4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_4"
  type: "IPlugin"
  bottom: "conv4_4"
  top: "conv4_4"
}
layer {
  name: "res4_4"
  type: "Eltwise"
  bottom: "res4_2"
  bottom: "conv4_4"
  top: "res4_4"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_5"
  type: "Convolution"
  bottom: "res4_4"
  top: "conv4_5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_5"
  type: "IPlugin"
  bottom: "conv4_5"
  top: "conv4_5"
}
layer {
  name: "conv4_6"
  type: "Convolution"
  bottom: "conv4_5"
  top: "conv4_6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_6"
  type: "IPlugin"
  bottom: "conv4_6"
  top: "conv4_6"
}
layer {
  name: "res4_6"
  type: "Eltwise"
  bottom: "res4_4"
  bottom: "conv4_6"
  top: "res4_6"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_7"
  type: "Convolution"
  bottom: "res4_6"
  top: "conv4_7"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_7"
  type: "IPlugin"
  bottom: "conv4_7"
  top: "conv4_7"
}
layer {
  name: "conv4_8"
  type: "Convolution"
  bottom: "conv4_7"
  top: "conv4_8"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_8"
  type: "IPlugin"
  bottom: "conv4_8"
  top: "conv4_8"
}
layer {
  name: "res4_8"
  type: "Eltwise"
  bottom: "res4_6"
  bottom: "conv4_8"
  top: "res4_8"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4_9"
  type: "Convolution"
  bottom: "res4_8"
  top: "conv4_9"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_9"
  type: "IPlugin"
  bottom: "conv4_9"
  top: "conv4_9"
}
layer {
  name: "conv4_10"
  type: "Convolution"
  bottom: "conv4_9"
  top: "conv4_10"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_10"
  type: "IPlugin"
  bottom: "conv4_10"
  top: "conv4_10"
}
layer {
  name: "res4_10"
  type: "Eltwise"
  bottom: "res4_8"
  bottom: "conv4_10"
  top: "res4_10"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv4"
  type: "Convolution"
  bottom: "res4_10"
  top: "conv4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4"
  type: "IPlugin"
  bottom: "conv4"
  top: "conv4"
}
layer {
  name: "pool4"
  type: "Pooling"
  bottom: "conv4"
  top: "pool4"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "conv5_1"
  type: "Convolution"
  bottom: "pool4"
  top: "conv5_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_1"
  type: "IPlugin"
  bottom: "conv5_1"
  top: "conv5_1"
}
layer {
  name: "conv5_2"
  type: "Convolution"
  bottom: "conv5_1"
  top: "conv5_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_2"
  type: "IPlugin"
  bottom: "conv5_2"
  top: "conv5_2"
}
layer {
  name: "res5_2"
  type: "Eltwise"
  bottom: "pool4"
  bottom: "conv5_2"
  top: "res5_2"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv5_3"
  type: "Convolution"
  bottom: "res5_2"
  top: "conv5_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_3"
  type: "IPlugin"
  bottom: "conv5_3"
  top: "conv5_3"
}
layer {
  name: "conv5_4"
  type: "Convolution"
  bottom: "conv5_3"
  top: "conv5_4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_4"
  type: "IPlugin"
  bottom: "conv5_4"
  top: "conv5_4"
}
layer {
  name: "res5_4"
  type: "Eltwise"
  bottom: "res5_2"
  bottom: "conv5_4"
  top: "res5_4"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "conv5_5"
  type: "Convolution"
  bottom: "res5_4"
  top: "conv5_5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_5"
  type: "IPlugin"
  bottom: "conv5_5"
  top: "conv5_5"
}
layer {
  name: "conv5_6"
  type: "Convolution"
  bottom: "conv5_5"
  top: "conv5_6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5_6"
  type: "IPlugin"
  bottom: "conv5_6"
  top: "conv5_6"
}
layer {
  name: "res5_6"
  type: "Eltwise"
  bottom: "res5_4"
  bottom: "conv5_6"
  top: "res5_6"
  eltwise_param { 
    operation: 1
  }
}
layer {
  name: "fc5"
  type: "InnerProduct"
  bottom: "res5_6"
  top: "fc5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

然后进行相应的模型解析;

1.Prelu代码的实现，我参考的代码在这个位置：https://github.com/Goingqs/TensorRT-Prelu

同时我将代码也上传到这个位置了：https://download.csdn.net/download/qq_22764813/10727259（我的资源页）

2.在比对网络结构fc5的输出的时候，输出特征参数只有一半是正确，有一般为0，错误的原因是我这个代码使用错误：

cudaMemcpyAsync( data_cpu, data_gpu, sizeof(float)*count, cudaMemcpyDeviceToHost, stream);

这个函数中的第三个参数是所占用的字节byte数，但是在使用时，我只填入的count（数据的个数），所以在数据在内存拷贝的时候出现了问题，只拷贝了部分数据;