TensorRT-5.1.5.0-SSD

文档：

https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/tensorrt-513

https://docs.nvidia.com/deeplearning/sdk/tensorrt-api/c_api/index.html

安装：

https://docs.nvidia.com/deeplearning/sdk/tensorrt-install-guide/index.html

https://docs.nvidia.com/deeplearning/sdk/tensorrt-install-guide/index.html#installing-tar

export CUDA_HOME=/usr/local/cuda
export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/boyun/NVIDIA/TensorRT-5.1.5.0/lib

sudo ./cuda_9.0.176_384.81_linux.run
sudo ./cuda_9.0.176.1_linux.run
sudo ./cuda_9.0.176.2_linux.run
sudo ./cuda_9.0.176.3_linux.run
sudo ./cuda_9.0.176.4_linux.run
cp  cudnn-9.0-linux-x64-v7.5.1.10.solitairetheme8 cudnn-9.0-linux-x64-v7.5.1.10.tgz
sudo cp cuda/include/cudnn.h /usr/local/cuda-9.0/include/
sudo cp cuda/lib64/libcudnn* /usr/local/cuda-9.0/lib64/
cd /usr/local/cuda-9.0/lib64/
sudo rm -rf libcudnn.so.7
sudo rm -rf libcudnn.so
sudo ln -s libcudnn.so.7.5.1 libcudnn.so.7
sudo ln -s libcudnn.so.7 libcudnn.so

sudo gedit ~/.bashrc

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/boyun/NVIDIA/TensorRT-5.1.5.0/lib

cd /home/boyun/NVIDIA/TensorRT-5.1.5.0/python
sudo pip install tensorrt-5.1.5.0-cp27-none-linux_x86_64.whl

一些插件，比如使用TensorFlow的时候，所以我们都安装上：
cd /home/boyun/NVIDIA/TensorRT-5.1.5.0/uff
sudo pip install uff-0.6.3-py2.py3-none-any.whl 

cd /home/boyun/NVIDIA/TensorRT-5.1.5.0/graphsurgeon
sudo pip install graphsurgeon-0.4.1-py2.py3-none-any.whl

测试安装：

import tensorrt

 编译：

cd /home/boyun/NVIDIA/TensorRT-5.1.5.0/samples/
make -20
cd /home/boyun/NVIDIA/TensorRT-5.1.5.0/bin/

跑例子
./sample_googlenet

&&&& RUNNING TensorRT.sample_googlenet # ./sample_googlenet
[I] Building and running a GPU inference engine for GoogleNet
[I] Ran ./sample_googlenet with: 
[I] Input(s): data 
[I] Output(s): prob 
&&&& PASSED TensorRT.sample_googlenet # ./sample_googlenet

测试时间
./trtexec

&&&& RUNNING TensorRT.trtexec # ./trtexec

Mandatory params:
  --deploy=<file>          Caffe deploy file
  OR --uff=<file>          UFF file
  OR --onnx=<file>         ONNX Model file
  OR --loadEngine=<file>   Load a saved engine

Mandatory params for UFF:
  --uffInput=<name>,C,H,W Input blob name and its dimensions for UFF parser (can be specified multiple times)
  --output=<name>      Output blob name (can be specified multiple times)

Mandatory params for Caffe:
  --output=<name>      Output blob name (can be specified multiple times)

Optional params:
  --model=<file>          Caffe model file (default = no model, random weights used)
  --batch=N               Set batch size (default = 1)
  --device=N              Set cuda device to N (default = 0)
  --iterations=N          Run N iterations (default = 10)
  --avgRuns=N             Set avgRuns to N - perf is measured as an average of avgRuns (default=10)
  --percentile=P          For each iteration, report the percentile time at P percentage (0<=P<=100, with 0 representing min, and 100 representing max; default = 99.0%)
  --workspace=N           Set workspace size in megabytes (default = 16)
  --safe                  Only test the functionality available in safety restricted flows.
  --fp16                  Run in fp16 mode (default = false). Permits 16-bit kernels
  --int8                  Run in int8 mode (default = false). Currently no support for ONNX model.
  --verbose               Use verbose logging (default = false)
  --saveEngine=<file>     Save a serialized engine to file.
  --loadEngine=<file>     Load a serialized engine from file.
  --calib=<file>          Read INT8 calibration cache file.  Currently no support for ONNX model.
  --useDLACore=N          Specify a DLA engine for layers that support DLA. Value can range from 0 to n-1, where n is the number of DLA engines on the platform.
  --allowGPUFallback      If --useDLACore flag is present and if a layer can't run on DLA, then run on GPU. 
  --useSpinWait           Actively wait for work completion. This option may decrease multi-process synchronization time at the cost of additional CPU usage. (default = false)
  --dumpOutput            Dump outputs at end of test. 
  -h, --help              Print usage
&&&& FAILED TensorRT.trtexec # ./trtexec

测试时间程序：请使用绝对路径 

./trtexec --deploy=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.prototxt --output=prob --model=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.caffemodel

 

&&&& RUNNING TensorRT.trtexec # ./trtexec --deploy=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.prototxt --output=prob --model=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.caffemodel
[I] deploy: /home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.prototxt
[I] output: prob
[I] model: /home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.caffemodel
[I] Input "data": 3x224x224
[I] Output "prob": 1000x1x1
[I] Average over 10 runs is 1.42812 ms (host walltime is 1.55343 ms, 99% percentile time is 1.43962).
[I] Average over 10 runs is 1.42616 ms (host walltime is 1.51039 ms, 99% percentile time is 1.43974).
[I] Average over 10 runs is 1.42445 ms (host walltime is 1.60433 ms, 99% percentile time is 1.43638).
[I] Average over 10 runs is 1.42804 ms (host walltime is 1.5044 ms, 99% percentile time is 1.43565).
[I] Average over 10 runs is 1.42922 ms (host walltime is 1.49007 ms, 99% percentile time is 1.43565).
[I] Average over 10 runs is 1.42908 ms (host walltime is 1.59964 ms, 99% percentile time is 1.44486).
[I] Average over 10 runs is 1.43004 ms (host walltime is 1.47835 ms, 99% percentile time is 1.43974).
[I] Average over 10 runs is 1.4258 ms (host walltime is 1.60077 ms, 99% percentile time is 1.43667).
[I] Average over 10 runs is 1.42793 ms (host walltime is 1.48041 ms, 99% percentile time is 1.43872).
[I] Average over 10 runs is 1.4263 ms (host walltime is 1.618 ms, 99% percentile time is 1.43155).
&&&& PASSED TensorRT.trtexec # ./trtexec --deploy=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.prototxt --output=prob --model=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.caffemodel

将推理出的TensorRT的模型保存

mkdir saveEngine

./trtexec --deploy=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.prototxt --output=prob --model=/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/googlenet/googlenet.caffemodel --saveEngine=/home/boyun/NVIDIA/TensorRT-5.1.5.0/saveEngine/googlenet

注释：

sampleGoogleNet.cpp:

/*
 * Copyright 1993-2019 NVIDIA Corporation.  All rights reserved.
 *
 * NOTICE TO LICENSEE:
 *
 * This source code and/or documentation ("Licensed Deliverables") are
 * subject to NVIDIA intellectual property rights under U.S. and
 * international Copyright laws.
 *
 * These Licensed Deliverables contained herein is PROPRIETARY and
 * CONFIDENTIAL to NVIDIA and is being provided under the terms and
 * conditions of a form of NVIDIA software license agreement by and
 * between NVIDIA and Licensee ("License Agreement") or electronically
 * accepted by Licensee.  Notwithstanding any terms or conditions to
 * the contrary in the License Agreement, reproduction or disclosure
 * of the Licensed Deliverables to any third party without the express
 * written consent of NVIDIA is prohibited.
 *
 * NOTWITHSTANDING ANY TERMS OR CONDITIONS TO THE CONTRARY IN THE
 * LICENSE AGREEMENT, NVIDIA MAKES NO REPRESENTATION ABOUT THE
 * SUITABILITY OF THESE LICENSED DELIVERABLES FOR ANY PURPOSE.  IT IS
 * PROVIDED "AS IS" WITHOUT EXPRESS OR IMPLIED WARRANTY OF ANY KIND.
 * NVIDIA DISCLAIMS ALL WARRANTIES WITH REGARD TO THESE LICENSED
 * DELIVERABLES, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY,
 * NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR PURPOSE.
 * NOTWITHSTANDING ANY TERMS OR CONDITIONS TO THE CONTRARY IN THE
 * LICENSE AGREEMENT, IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY
 * SPECIAL, INDIRECT, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, OR ANY
 * DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
 * WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
 * ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
 * OF THESE LICENSED DELIVERABLES.
 *
 * U.S. Government End Users.  These Licensed Deliverables are a
 * "commercial item" as that term is defined at 48 C.F.R. 2.101 (OCT
 * 1995), consisting of "commercial computer software" and "commercial
 * computer software documentation" as such terms are used in 48
 * C.F.R. 12.212 (SEPT 1995) and is provided to the U.S. Government
 * only as a commercial end item.  Consistent with 48 C.F.R.12.212 and
 * 48 C.F.R. 227.7202-1 through 227.7202-4 (JUNE 1995), all
 * U.S. Government End Users acquire the Licensed Deliverables with
 * only those rights set forth herein.
 *
 * Any use of the Licensed Deliverables in individual and commercial
 * software must include, in the user documentation and internal
 * comments to the code, the above Disclaimer and U.S. Government End
 * Users Notice.
 */

//!
//! sampleGoogleNet.cpp
//! This file contains the implementation of the GoogleNet sample. It creates the network using
//! the GoogleNet caffe model.
//! It can be run with the following command line:
//! Command: ./sample_googlenet [-h or --help] [-d=/path/to/data/dir or --datadir=/path/to/data/dir]
//!

#include "argsParser.h"
#include "buffers.h"
#include "logger.h"
#include "common.h"
#include "NvCaffeParser.h"
#include "NvInfer.h"
#include <cuda_runtime_api.h>

#include <cstdlib>
#include <fstream>
#include <iostream>
#include <sstream>

//yangninghua
#include "time.h"

const std::string gSampleName = "TensorRT.sample_googlenet";

//!
//! \brief  The SampleGoogleNet class implements the GoogleNet sample
//!
//! \details It creates the network using a caffe model
//!
class SampleGoogleNet
{
    template <typename T>
    using SampleUniquePtr = std::unique_ptr<T, samplesCommon::InferDeleter>;

public:
    SampleGoogleNet(const samplesCommon::CaffeSampleParams& params)
        : mParams(params)
    {
    }

    //!
    //! \brief Function builds the network engine
    //!
    //构建网络引擎
    bool build();

    //!
    //! \brief This function runs the TensorRT inference engine for this sample
    //!
    //运行TensorRT推理引擎
    bool infer();

    //!
    //! \brief This function can be used to clean up any state created in the sample class
    //!
    //清空状态
    bool teardown();

    samplesCommon::CaffeSampleParams mParams;

private:
    //用于运行网络的TensorRT引擎
    std::shared_ptr<nvinfer1::ICudaEngine> mEngine = nullptr; //!< The TensorRT engine used to run the network

    //!
    //! \brief This function parses a Caffe model for GoogleNet and creates a TensorRT network
    //!
    //此功能解析GoogleNet的Caffe模型并创建TensorRT网络
    void constructNetwork(SampleUniquePtr<nvinfer1::IBuilder>& builder, SampleUniquePtr<nvinfer1::INetworkDefinition>& network, SampleUniquePtr<nvcaffeparser1::ICaffeParser>& parser);
};

//!
//! \brief This function creates the network, configures the builder and creates the network engine
//!
//! \details This function creates the GoogleNet network by parsing the caffe model and builds
//!          the engine that will be used to run GoogleNet (mEngine)
//!
//! \return Returns true if the engine was created successfully and false otherwise
//!
bool SampleGoogleNet::build()
{
    //一个名为的全局TensorRT API方法 createInferBuilder（gLogger） 用于创建类型的对象 IBuilder
    //创建IBuilder同iLogger作为输入参数
    auto builder = SampleUniquePtr<nvinfer1::IBuilder>(nvinfer1::createInferBuilder(gLogger.getTRTLogger()));
    if (!builder)
        return false;

    //一种叫做的方法 createNetwork 为iBuilder定义用于创建类型的对象 iNetworkDefinition'
    //createNetwork()用于创建网络
    auto network = SampleUniquePtr<nvinfer1::INetworkDefinition>(builder->createNetwork());
    if (!network)
        return false;

    //创建一个可用的解析器（Caffe，ONNX或UFF）
    //ONNX:  auto parser = nvonnxparser::createParser(*network, gLogger);
    //Caffe: auto parser = nvcaffeparser1::createCaffeParser();
    //UFF:   auto parser = nvuffparser::createUffParser();
    auto parser = SampleUniquePtr<nvcaffeparser1::ICaffeParser>(nvcaffeparser1::createCaffeParser());
    if (!parser)
        return false;

    //加载caffe模型
    //解析模型文件
    constructNetwork(builder, network, parser);

    //一种叫做的方法buildCudaEngine()的IBuilder被调用来创建一个对象ICudaEngine类型
    //创建TensorRT引擎
    //可以选择将引擎序列化并转储到文件中。
    mEngine = std::shared_ptr<nvinfer1::ICudaEngine>(builder->buildCudaEngine(*network), samplesCommon::InferDeleter());
    if (!mEngine)
        return false;

    return true;
}

//!
//! \brief This function uses a caffe parser to create the googlenet Network and marks the
//!        output layers
//!
//! \param network Pointer to the network that will be populated with the googlenet network
//!
//! \param builder Pointer to the engine builder
//!
void SampleGoogleNet::constructNetwork(SampleUniquePtr<nvinfer1::IBuilder>& builder, SampleUniquePtr<nvinfer1::INetworkDefinition>& network, SampleUniquePtr<nvcaffeparser1::ICaffeParser>& parser)
{
    //对象iParser调用parse方法读取模型文件并填充TensorRT网络
    //params.dataDirs.push_back("data/googlenet/");
    //params.dataDirs.push_back("data/samples/googlenet/");
    //params.prototxtFileName = "googlenet.prototxt";
    //params.weightsFileName = "googlenet.caffemodel";
    const nvcaffeparser1::IBlobNameToTensor* blobNameToTensor = parser->parse(
        locateFile(mParams.prototxtFileName, mParams.dataDirs).c_str(),
        locateFile(mParams.weightsFileName, mParams.dataDirs).c_str(),
        *network,
        nvinfer1::DataType::kFLOAT);

    //params.outputTensorNames.push_back("prob");
    for (auto& s : mParams.outputTensorNames)
        network->markOutput(*blobNameToTensor->find(s.c_str()));

    //params.batchSize = 4;
    //params.dlaCore = args.useDLACore;
    builder->setMaxBatchSize(mParams.batchSize);
    builder->setMaxWorkspaceSize(16_MB);

    //builder->setFp16Mode(true);
    //builder->setInt8Mode(true);

    samplesCommon::enableDLA(builder.get(), mParams.dlaCore);
}

//!
//! \brief This function runs the TensorRT inference engine for this sample
//!
//! \details This function is the main execution function of the sample. It allocates the buffer,
//!          sets inputs and executes the engine.
//!
bool SampleGoogleNet::infer()
{
    // Create RAII buffer manager object
    samplesCommon::BufferManager buffers(mEngine, mParams.batchSize);

    //IExecutionContext用于执行推理引擎
    auto context = SampleUniquePtr<nvinfer1::IExecutionContext>(mEngine->createExecutionContext());
    if (!context)
        return false;

    // Fetch host buffers and set host input buffers to all zeros
    //获取主机缓冲区并将主机输入缓冲区设置为全零
    for (auto& input : mParams.inputTensorNames)
    {
        const auto bufferSize = buffers.size(input);
        if (bufferSize == samplesCommon::BufferManager::kINVALID_SIZE_VALUE)
        {
            gLogError << "input tensor missing: " << input << "\n";
            exit(EXIT_FAILURE);
        }
        memset(buffers.getHostBuffer(input), 0, bufferSize);
    }

    // Memcpy from host input buffers to device input buffers
    //Memcpy从主机输入缓冲区到设备输入缓冲区
    buffers.copyInputToDevice();

    clock_t start, finish;
    double  duration;
    start = clock();
    bool status = context->execute(mParams.batchSize, buffers.getDeviceBindings().data());
    if (!status)
        return false;
    finish = clock();
    duration = (double)(finish - start) / CLOCKS_PER_SEC;
    cout<<"前向推理时间"<<duration*1000<<"ms"<<endl;;

    // Memcpy from device output buffers to host output buffers
    //从设备输出缓冲区到主机输出缓冲区的Memcpy
    buffers.copyOutputToHost();

    return true;
}

//!
//! \brief This function can be used to clean up any state created in the sample class
//!
bool SampleGoogleNet::teardown()
{
    //! Clean up the libprotobuf files as the parsing is complete
    //! \note It is not safe to use any other part of the protocol buffers library after
    //! ShutdownProtobufLibrary() has been called.
    nvcaffeparser1::shutdownProtobufLibrary();
    return true;
}

//!
//! \brief This function initializes members of the params struct using the command line args
//!
samplesCommon::CaffeSampleParams initializeSampleParams(const samplesCommon::Args& args)
{
    samplesCommon::CaffeSampleParams params;
    if (args.dataDirs.size() != 0) //!< Use the data directory provided by the user
        params.dataDirs = args.dataDirs;
    else //!< Use default directories if user hasn't provided directory paths
    {
        params.dataDirs.push_back("data/googlenet/");
        //params.dataDirs.push_back("data/samples/googlenet/");
    }
    params.prototxtFileName = "googlenet.prototxt";
    params.weightsFileName = "googlenet.caffemodel";
    params.inputTensorNames.push_back("data");
    params.batchSize = 4;
    params.outputTensorNames.push_back("prob");
    params.dlaCore = args.useDLACore;

    return params;
}
//!
//! \brief This function prints the help information for running this sample
//!
void printHelpInfo()
{
    std::cout << "Usage: ./sample_googlenet [-h or --help] [-d or --datadir=<path to data directory>] [--useDLACore=<int>]\n";
    std::cout << "--help          Display help information\n";
    std::cout << "--datadir       Specify path to a data directory, overriding the default. This option can be used multiple times to add multiple directories. If no data directories are given, the default is to use data/samples/googlenet/ and data/googlenet/" << std::endl;
    std::cout << "--useDLACore=N  Specify a DLA engine for layers that support DLA. Value can range from 0 to n-1, where n is the number of DLA engines on the platform." << std::endl;
}

int main(int argc, char** argv)
{
    samplesCommon::Args args;
    bool argsOK = samplesCommon::parseArgs(args, argc, argv);
    if (args.help)
    {
        printHelpInfo();
        return EXIT_SUCCESS;
    }
    if (!argsOK)
    {
        gLogError << "Invalid arguments" << std::endl;
        printHelpInfo();
        return EXIT_FAILURE;
    }

    auto sampleTest = gLogger.defineTest(gSampleName, argc, const_cast<const char**>(argv));
    
    //函数reportTestEnd与其对应
    //&&&& RUNNING TensorRT.sample_googlenet # ./sample_googlenet
    gLogger.reportTestStart(sampleTest);

    //参数初始化
    samplesCommon::CaffeSampleParams params = initializeSampleParams(args);
    SampleGoogleNet sample(params);
    gLogInfo << "Building and running a GPU inference engine for GoogleNet" << std::endl;




    if (!sample.build())
    {
        return gLogger.reportFail(sampleTest);
    }
    if (!sample.infer())
    {
        return gLogger.reportFail(sampleTest);
    }
    if (!sample.teardown())
    {
        return gLogger.reportFail(sampleTest);
    }

    //[I] Ran ./sample_googlenet with: 
    gLogInfo << "Ran " << argv[0] << " with: " << std::endl;

    std::stringstream ss;

    //可能有多个输入，进行遍历
    //[I] Input(s): data 
    ss << "Input(s): ";
    for (auto& input : sample.mParams.inputTensorNames)
        ss << input << " ";
    gLogInfo << ss.str() << std::endl;

    ss.str(std::string());

    //可能有多个输出，进行遍历
    //[I] Output(s): prob 
    ss << "Output(s): ";
    for (auto& output : sample.mParams.outputTensorNames)
        ss << output << " ";
    gLogInfo << ss.str() << std::endl;

    //reportTestEnd函数在reportPass中被调用
    //&&&& PASSED TensorRT.sample_googlenet # ./sample_googlenet
    return gLogger.reportPass(sampleTest);
}

 

接着修改SSD的例子：

https://github.com/NVIDIA/TensorRT/tree/release/5.1/samples/opensource/sampleSSD

---

下载SSD的权重和deploy.prototxt还有labelmap_voc.prototxt

https://drive.google.com/file/d/0BzKzrI_SkD1_WVVTSmQxU0dVRzA/view

里面有VGG_VOC0712_SSD_300x300_iter_120000.caffemodel和deploy.prototxt

https://github.com/intel/caffe/blob/master/data/VOC0712/labelmap_voc.prototxt

可以找到labelmap_voc.prototxt

mv deploy.prototxt  /data/ssd/ssd.prototxt

然后修改这个ssd.prototxt

首先：param，weight_filler，bias_filler参数全部不要(只做推理，要这没用)但是不去也行，我懒了抱歉

更改为：

 

所有的Flatten改为Reshape这样

Update the detection_out layer to add the keep_count output as expected by the TensorRT DetectionOutput Plugin. top: "keep_count"

&&&& RUNNING TensorRT.sample_ssd # ./sample_ssd
[I] Begin parsing model...
[I] FP32 mode running...
[I] End parsing model...
[I] Begin building engine...
[I] End building engine...
[I] *** deserializing
[I]  Image name:../../../data/ssd/bus.ppm, Label: car, confidence: 96.0588 xmin: 4.14485 ymin: 117.443 xmax: 244.102 ymax: 241.829
&&&& PASSED TensorRT.sample_ssd # ./sample_ssd

生成的图保存在：

附上修改后的ssd.prototxt

name: "VGG_VOC0712_SSD_300x300_deploy"
input: "data"
input_shape {
  dim: 1
  dim: 3
  dim: 300
  dim: 300
}
layer {
  name: "conv1_1"
  type: "Convolution"
  bottom: "data"
  top: "conv1_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_1"
  type: "ReLU"
  bottom: "conv1_1"
  top: "conv1_1"
}
layer {
  name: "conv1_2"
  type: "Convolution"
  bottom: "conv1_1"
  top: "conv1_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_2"
  type: "ReLU"
  bottom: "conv1_2"
  top: "conv1_2"
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1_2"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "conv2_1"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2_1"
  type: "ReLU"
  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: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2_2"
  type: "ReLU"
  bottom: "conv2_2"
  top: "conv2_2"
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2_2"
  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: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_1"
  type: "ReLU"
  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: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_2"
  type: "ReLU"
  bottom: "conv3_2"
  top: "conv3_2"
}
layer {
  name: "conv3_3"
  type: "Convolution"
  bottom: "conv3_2"
  top: "conv3_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_3"
  type: "ReLU"
  bottom: "conv3_3"
  top: "conv3_3"
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "conv3_3"
  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: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_1"
  type: "ReLU"
  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: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_2"
  type: "ReLU"
  bottom: "conv4_2"
  top: "conv4_2"
}
layer {
  name: "conv4_3"
  type: "Convolution"
  bottom: "conv4_2"
  top: "conv4_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_3"
  type: "ReLU"
  bottom: "conv4_3"
  top: "conv4_3"
}
layer {
  name: "pool4"
  type: "Pooling"
  bottom: "conv4_3"
  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: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    dilation: 1
  }
}
layer {
  name: "relu5_1"
  type: "ReLU"
  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: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    dilation: 1
  }
}
layer {
  name: "relu5_2"
  type: "ReLU"
  bottom: "conv5_2"
  top: "conv5_2"
}
layer {
  name: "conv5_3"
  type: "Convolution"
  bottom: "conv5_2"
  top: "conv5_3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    dilation: 1
  }
}
layer {
  name: "relu5_3"
  type: "ReLU"
  bottom: "conv5_3"
  top: "conv5_3"
}
layer {
  name: "pool5"
  type: "Pooling"
  bottom: "conv5_3"
  top: "pool5"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 1
    pad: 1
  }
}
layer {
  name: "fc6"
  type: "Convolution"
  bottom: "pool5"
  top: "fc6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 1024
    pad: 6
    kernel_size: 3
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    dilation: 6
  }
}
layer {
  name: "relu6"
  type: "ReLU"
  bottom: "fc6"
  top: "fc6"
}
layer {
  name: "fc7"
  type: "Convolution"
  bottom: "fc6"
  top: "fc7"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 1024
    kernel_size: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu7"
  type: "ReLU"
  bottom: "fc7"
  top: "fc7"
}
layer {
  name: "conv6_1"
  type: "Convolution"
  bottom: "fc7"
  top: "conv6_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv6_1_relu"
  type: "ReLU"
  bottom: "conv6_1"
  top: "conv6_1"
}
layer {
  name: "conv6_2"
  type: "Convolution"
  bottom: "conv6_1"
  top: "conv6_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 512
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv6_2_relu"
  type: "ReLU"
  bottom: "conv6_2"
  top: "conv6_2"
}
layer {
  name: "conv7_1"
  type: "Convolution"
  bottom: "conv6_2"
  top: "conv7_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv7_1_relu"
  type: "ReLU"
  bottom: "conv7_1"
  top: "conv7_1"
}
layer {
  name: "conv7_2"
  type: "Convolution"
  bottom: "conv7_1"
  top: "conv7_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv7_2_relu"
  type: "ReLU"
  bottom: "conv7_2"
  top: "conv7_2"
}
layer {
  name: "conv8_1"
  type: "Convolution"
  bottom: "conv7_2"
  top: "conv8_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv8_1_relu"
  type: "ReLU"
  bottom: "conv8_1"
  top: "conv8_1"
}
layer {
  name: "conv8_2"
  type: "Convolution"
  bottom: "conv8_1"
  top: "conv8_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv8_2_relu"
  type: "ReLU"
  bottom: "conv8_2"
  top: "conv8_2"
}
layer {
  name: "conv9_1"
  type: "Convolution"
  bottom: "conv8_2"
  top: "conv9_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv9_1_relu"
  type: "ReLU"
  bottom: "conv9_1"
  top: "conv9_1"
}
layer {
  name: "conv9_2"
  type: "Convolution"
  bottom: "conv9_1"
  top: "conv9_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv9_2_relu"
  type: "ReLU"
  bottom: "conv9_2"
  top: "conv9_2"
}
layer {
  name: "conv4_3_norm"
  type: "Normalize"
  bottom: "conv4_3"
  top: "conv4_3_norm"
  norm_param {
    across_spatial: false
    scale_filler {
      type: "constant"
      value: 20
    }
    channel_shared: false
  }
}
layer {
  name: "conv4_3_norm_mbox_loc"
  type: "Convolution"
  bottom: "conv4_3_norm"
  top: "conv4_3_norm_mbox_loc"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 16
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv4_3_norm_mbox_loc_perm"
  type: "Permute"
  bottom: "conv4_3_norm_mbox_loc"
  top: "conv4_3_norm_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv4_3_norm_mbox_loc_flat"
#  type: "Flatten"
#  bottom: "conv4_3_norm_mbox_loc_perm"
#  top: "conv4_3_norm_mbox_loc_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv4_3_norm_mbox_loc_flat"
  type: "Reshape"
  bottom: "conv4_3_norm_mbox_loc_perm"
  top: "conv4_3_norm_mbox_loc_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv4_3_norm_mbox_conf"
  type: "Convolution"
  bottom: "conv4_3_norm"
  top: "conv4_3_norm_mbox_conf"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 84
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv4_3_norm_mbox_conf_perm"
  type: "Permute"
  bottom: "conv4_3_norm_mbox_conf"
  top: "conv4_3_norm_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv4_3_norm_mbox_conf_flat"
#  type: "Flatten"
#  bottom: "conv4_3_norm_mbox_conf_perm"
#  top: "conv4_3_norm_mbox_conf_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv4_3_norm_mbox_conf_flat"
  type: "Reshape"
  bottom: "conv4_3_norm_mbox_conf_perm"
  top: "conv4_3_norm_mbox_conf_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv4_3_norm_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv4_3_norm"
  bottom: "data"
  top: "conv4_3_norm_mbox_priorbox"
  prior_box_param {
    min_size: 30.0
    max_size: 60.0
    aspect_ratio: 2
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    step: 8
    offset: 0.5
  }
}
layer {
  name: "fc7_mbox_loc"
  type: "Convolution"
  bottom: "fc7"
  top: "fc7_mbox_loc"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 24
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "fc7_mbox_loc_perm"
  type: "Permute"
  bottom: "fc7_mbox_loc"
  top: "fc7_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "fc7_mbox_loc_flat"
#  type: "Flatten"
#  bottom: "fc7_mbox_loc_perm"
#  top: "fc7_mbox_loc_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "fc7_mbox_loc_flat"
  type: "Reshape"
  bottom: "fc7_mbox_loc_perm"
  top: "fc7_mbox_loc_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "fc7_mbox_conf"
  type: "Convolution"
  bottom: "fc7"
  top: "fc7_mbox_conf"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 126
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "fc7_mbox_conf_perm"
  type: "Permute"
  bottom: "fc7_mbox_conf"
  top: "fc7_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "fc7_mbox_conf_flat"
#  type: "Flatten"
#  bottom: "fc7_mbox_conf_perm"
#  top: "fc7_mbox_conf_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "fc7_mbox_conf_flat"
  type: "Reshape"
  bottom: "fc7_mbox_conf_perm"
  top: "fc7_mbox_conf_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "fc7_mbox_priorbox"
  type: "PriorBox"
  bottom: "fc7"
  bottom: "data"
  top: "fc7_mbox_priorbox"
  prior_box_param {
    min_size: 60.0
    max_size: 111.0
    aspect_ratio: 2
    aspect_ratio: 3
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    step: 16
    offset: 0.5
  }
}
layer {
  name: "conv6_2_mbox_loc"
  type: "Convolution"
  bottom: "conv6_2"
  top: "conv6_2_mbox_loc"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 24
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv6_2_mbox_loc_perm"
  type: "Permute"
  bottom: "conv6_2_mbox_loc"
  top: "conv6_2_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv6_2_mbox_loc_flat"
#  type: "Flatten"
#  bottom: "conv6_2_mbox_loc_perm"
#  top: "conv6_2_mbox_loc_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv6_2_mbox_loc_flat"
  type: "Reshape"
  bottom: "conv6_2_mbox_loc_perm"
  top: "conv6_2_mbox_loc_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv6_2_mbox_conf"
  type: "Convolution"
  bottom: "conv6_2"
  top: "conv6_2_mbox_conf"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 126
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv6_2_mbox_conf_perm"
  type: "Permute"
  bottom: "conv6_2_mbox_conf"
  top: "conv6_2_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv6_2_mbox_conf_flat"
#  type: "Flatten"
#  bottom: "conv6_2_mbox_conf_perm"
#  top: "conv6_2_mbox_conf_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv6_2_mbox_conf_flat"
  type: "Reshape"
  bottom: "conv6_2_mbox_conf_perm"
  top: "conv6_2_mbox_conf_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv6_2_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv6_2"
  bottom: "data"
  top: "conv6_2_mbox_priorbox"
  prior_box_param {
    min_size: 111.0
    max_size: 162.0
    aspect_ratio: 2
    aspect_ratio: 3
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    step: 32
    offset: 0.5
  }
}
layer {
  name: "conv7_2_mbox_loc"
  type: "Convolution"
  bottom: "conv7_2"
  top: "conv7_2_mbox_loc"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 24
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv7_2_mbox_loc_perm"
  type: "Permute"
  bottom: "conv7_2_mbox_loc"
  top: "conv7_2_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv7_2_mbox_loc_flat"
#  type: "Flatten"
#  bottom: "conv7_2_mbox_loc_perm"
#  top: "conv7_2_mbox_loc_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv7_2_mbox_loc_flat"
  type: "Reshape"
  bottom: "conv7_2_mbox_loc_perm"
  top: "conv7_2_mbox_loc_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv7_2_mbox_conf"
  type: "Convolution"
  bottom: "conv7_2"
  top: "conv7_2_mbox_conf"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 126
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv7_2_mbox_conf_perm"
  type: "Permute"
  bottom: "conv7_2_mbox_conf"
  top: "conv7_2_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv7_2_mbox_conf_flat"
#  type: "Flatten"
#  bottom: "conv7_2_mbox_conf_perm"
#  top: "conv7_2_mbox_conf_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv7_2_mbox_conf_flat"
  type: "Reshape"
  bottom: "conv7_2_mbox_conf_perm"
  top: "conv7_2_mbox_conf_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv7_2_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv7_2"
  bottom: "data"
  top: "conv7_2_mbox_priorbox"
  prior_box_param {
    min_size: 162.0
    max_size: 213.0
    aspect_ratio: 2
    aspect_ratio: 3
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    step: 64
    offset: 0.5
  }
}
layer {
  name: "conv8_2_mbox_loc"
  type: "Convolution"
  bottom: "conv8_2"
  top: "conv8_2_mbox_loc"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 16
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv8_2_mbox_loc_perm"
  type: "Permute"
  bottom: "conv8_2_mbox_loc"
  top: "conv8_2_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv8_2_mbox_loc_flat"
#  type: "Flatten"
#  bottom: "conv8_2_mbox_loc_perm"
#  top: "conv8_2_mbox_loc_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv8_2_mbox_loc_flat"
  type: "Reshape"
  bottom: "conv8_2_mbox_loc_perm"
  top: "conv8_2_mbox_loc_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv8_2_mbox_conf"
  type: "Convolution"
  bottom: "conv8_2"
  top: "conv8_2_mbox_conf"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 84
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv8_2_mbox_conf_perm"
  type: "Permute"
  bottom: "conv8_2_mbox_conf"
  top: "conv8_2_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv8_2_mbox_conf_flat"
#  type: "Flatten"
#  bottom: "conv8_2_mbox_conf_perm"
#  top: "conv8_2_mbox_conf_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv8_2_mbox_conf_flat"
  type: "Reshape"
  bottom: "conv8_2_mbox_conf_perm"
  top: "conv8_2_mbox_conf_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv8_2_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv8_2"
  bottom: "data"
  top: "conv8_2_mbox_priorbox"
  prior_box_param {
    min_size: 213.0
    max_size: 264.0
    aspect_ratio: 2
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    step: 100
    offset: 0.5
  }
}
layer {
  name: "conv9_2_mbox_loc"
  type: "Convolution"
  bottom: "conv9_2"
  top: "conv9_2_mbox_loc"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 16
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv9_2_mbox_loc_perm"
  type: "Permute"
  bottom: "conv9_2_mbox_loc"
  top: "conv9_2_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv9_2_mbox_loc_flat"
#  type: "Flatten"
#  bottom: "conv9_2_mbox_loc_perm"
#  top: "conv9_2_mbox_loc_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv9_2_mbox_loc_flat"
  type: "Reshape"
  bottom: "conv9_2_mbox_loc_perm"
  top: "conv9_2_mbox_loc_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv9_2_mbox_conf"
  type: "Convolution"
  bottom: "conv9_2"
  top: "conv9_2_mbox_conf"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 84
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv9_2_mbox_conf_perm"
  type: "Permute"
  bottom: "conv9_2_mbox_conf"
  top: "conv9_2_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
########################################
#layer {
#  name: "conv9_2_mbox_conf_flat"
#  type: "Flatten"
#  bottom: "conv9_2_mbox_conf_perm"
#  top: "conv9_2_mbox_conf_flat"
#  flatten_param {
#    axis: 1
#  }
#}
layer {
  name: "conv9_2_mbox_conf_flat"
  type: "Reshape"
  bottom: "conv9_2_mbox_conf_perm"
  top: "conv9_2_mbox_conf_flat"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "conv9_2_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv9_2"
  bottom: "data"
  top: "conv9_2_mbox_priorbox"
  prior_box_param {
    min_size: 264.0
    max_size: 315.0
    aspect_ratio: 2
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    step: 300
    offset: 0.5
  }
}
layer {
  name: "mbox_loc"
  type: "Concat"
  bottom: "conv4_3_norm_mbox_loc_flat"
  bottom: "fc7_mbox_loc_flat"
  bottom: "conv6_2_mbox_loc_flat"
  bottom: "conv7_2_mbox_loc_flat"
  bottom: "conv8_2_mbox_loc_flat"
  bottom: "conv9_2_mbox_loc_flat"
  top: "mbox_loc"
  concat_param {
    axis: 1
  }
}
layer {
  name: "mbox_conf"
  type: "Concat"
  bottom: "conv4_3_norm_mbox_conf_flat"
  bottom: "fc7_mbox_conf_flat"
  bottom: "conv6_2_mbox_conf_flat"
  bottom: "conv7_2_mbox_conf_flat"
  bottom: "conv8_2_mbox_conf_flat"
  bottom: "conv9_2_mbox_conf_flat"
  top: "mbox_conf"
  concat_param {
    axis: 1
  }
}
layer {
  name: "mbox_priorbox"
  type: "Concat"
  bottom: "conv4_3_norm_mbox_priorbox"
  bottom: "fc7_mbox_priorbox"
  bottom: "conv6_2_mbox_priorbox"
  bottom: "conv7_2_mbox_priorbox"
  bottom: "conv8_2_mbox_priorbox"
  bottom: "conv9_2_mbox_priorbox"
  top: "mbox_priorbox"
  concat_param {
    axis: 2
  }
}
layer {
  name: "mbox_conf_reshape"
  type: "Reshape"
  bottom: "mbox_conf"
  top: "mbox_conf_reshape"
  reshape_param {
    shape {
      dim: 0
      dim: -1
      dim: 21
    }
  }
}
layer {
  name: "mbox_conf_softmax"
  type: "Softmax"
  bottom: "mbox_conf_reshape"
  top: "mbox_conf_softmax"
  softmax_param {
    axis: 2
  }
}
########################################
#layer {
#  name: "mbox_conf_flatten"
#  type: "Flatten"
#  bottom: "mbox_conf_softmax"
# top: "mbox_conf_flatten"
#  flatten_param {
#    axis: 1
# }
#}
layer {
  name: "mbox_conf_flatten"
  type: "Reshape"
  bottom: "mbox_conf_softmax"
  top: "mbox_conf_flatten"
  reshape_param {
    shape {
        dim: 0
        dim: -1
        dim: 1
        dim: 1
    }
  }
}
########################################
layer {
  name: "detection_out"
  type: "DetectionOutput"
  bottom: "mbox_loc"
  bottom: "mbox_conf_flatten"
  bottom: "mbox_priorbox"
  top: "detection_out"
  top: "keep_count"
  include {
    phase: TEST
  }
  detection_output_param {
    num_classes: 21
    share_location: true
    background_label_id: 0
    nms_param {
      nms_threshold: 0.45
      top_k: 400
    }
    save_output_param {
      label_map_file: "/home/boyun/NVIDIA/TensorRT-5.1.5.0/data/ssd/labelmap_voc.prototxt"
    }
    code_type: CENTER_SIZE
    keep_top_k: 200
    confidence_threshold: 0.01
  }
}

测试时间：

#include "time.h"

    clock_t start, finish;
    double  duration;
    start = clock();

    context.enqueue(batchSize, buffers, stream, nullptr);

    finish = clock();
    duration = (double)(finish - start) / CLOCKS_PER_SEC;
    cout<<"前向推理时间"<<duration*1000<<"ms"<<endl;;

 

&&&& RUNNING TensorRT.sample_ssd # ./sample_ssd
[I] Begin parsing model...
[I] FP32 mode running...
[I] End parsing model...
[I] Begin building engine...
[I] End building engine...
[I] *** deserializing
前向推理时间0.332ms
[I]  Image name:../../../data/ssd/bus.ppm, Label: car, confidence: 96.0588 xmin: 4.14485 ymin: 117.443 xmax: 244.102 ymax: 241.829
&&&& PASSED TensorRT.sample_ssd # ./sample_ssd

如果改变一个位置： 

&&&& RUNNING TensorRT.sample_ssd # ./sample_ssd
[I] Begin parsing model...
[I] FP32 mode running...
[I] End parsing model...
[I] Begin building engine...
[I] End building engine...
[I] *** deserializing
前向推理时间11.626ms
[I]  Image name:../../../data/ssd/bus.ppm, Label: car, confidence: 96.0587 xmin: 4.14484 ymin: 117.443 xmax: 244.102 ymax: 241.829
&&&& PASSED TensorRT.sample_ssd # ./sample_ssd

我的显卡是GTX TiTAN X 12G不是Nvidia TiTan X  12G

所以1080ti可能比这个更快

caffe-cuda9.0-cudnn7.5

 

./build/tools/caffe time -model=/home/boyun/code/caffe-ssd/ssd/deploy.prototxt --weights=/home/boyun/code/caffe-ssd/ssd/VGG_VOC0712_SSD_300x300_iter_120000.caffemodel --iterations=200 -gpu 0

I0714 13:52:09.150904 14202 caffe.cpp:412] Average Forward pass: 30.3141 ms.
I0714 13:52:09.150909 14202 caffe.cpp:414] Average Backward pass: 26.9677 ms.
I0714 13:52:09.150918 14202 caffe.cpp:416] Average Forward-Backward: 57.478 ms.
I0714 13:52:09.150923 14202 caffe.cpp:418] Total Time: 11495.6 ms.
I0714 13:52:09.150928 14202 caffe.cpp:419] *** Benchmark ends ***

下载：

https://developer.nvidia.com/nvidia-tensorrt-5x-download 

一些示例：

Caffe版yolov3+tensorRT：

https://www.jianshu.com/p/e78c5c321248?tdsourcetag=s_pcqq_aiomsg

https://github.com/C-H-D/tensorRT-Caffe

在Caffe中调用TensorRT提供的MNIST model

https://blog.csdn.net/fengbingchun/article/details/78606228

mobileNet-ssd使用tensorRT部署

https://blog.csdn.net/qq_17278169/article/details/82971983

TensorRT基于caffe模型加速MobileNet SSD

https://blog.csdn.net/qq_22764813/article/details/84544409

贤者之路， Caffe转TensorRT

https://blog.csdn.net/chanzhennan/article/details/87085754

caffe-ssd网络模型 tensorRT加速

http://www.pianshen.com/article/4373405601/

TensorRT 的 C++ API 使用详解

https://blog.csdn.net/u010552731/article/details/89501819

Use TensorRT API to implement Caffe-SSD， SSD（channel pruning）， Mobilenet-SSD

https://github.com/chenzhi1992/TensorRT-SSD

TensorRT-Mobilenet-SSD can run 50fps on jetson tx2

https://github.com/Ghustwb/MobileNet-SSD-TensorRT

TensorRT 3.0 RC run SSD error in DetectionOutput layer

https://devtalk.nvidia.com/default/topic/1025153/gpu-accelerated-libraries/tensorrt-3-0-rc-run-ssd-error-in-detectionoutput-layer/post/5214393/#5214393

faster RCNN tensorRT代码

https://github.com/NVIDIA/TensorRT/tree/release/5.1/samples/opensource/sampleFasterRCNN

SSD tensorRT 代码
https://github.com/NVIDIA/TensorRT/tree/release/5.1/samples/opensource/sampleSSD

github:TensorRT-SSD

https://github.com/onnx/onnx-tensorrt

https://github.com/ardianumam/Tensorflow-TensorRT

视频：

https://www.youtube.com/watch?v=AIGOSz2tFP8&list=PLkRkKTC6HZMwdtzv3PYJanRtR6ilSCZ4f

https://github.com/onnx/models

---

https://devblogs.nvidia.com/speed-up-inference-tensorrt/

How to Speed Up Deep Learning Inference Using TensorRT

https://developer.nvidia.com/tensorrt

文档：

https://developer.download.nvidia.cn/compute/machine-learning/tensorrt/models/sampleMLP-notebook.html

---

https://docs.nvidia.com/deeplearning/sdk/pdf/TensorRT-Developer-Guide.pdf

---

https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/index.html

---

https://docs.nvidia.com/deeplearning/sdk/tensorrt-sample-support-guide/index.html#charRNN_sample