tensorrt安装_TensorRT_C++:加载onnx模型，序列化和反序列化

0、参考

官网：

https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#deploy_opt_model

参考代码： https://github.com/haoxurt/tensorRT_save_serialization/blob/master/tensorRT_save_engine.cpp

or

https://github.com/faedtodd/Tensorrt-yolov3-win10/blob/master/yolov3_trt.cpp

1、环境准备

我是在jetson-nano上面跑的,版本信息如下:sudo jtop 后按5

jetpack自带了tensorRT，样例代码在/usr/src/tensorrt/samples

由于加载onnx需要额外的包，这里提供一下安装指南：

Jetson Xavier上tensorRT环境安装​www.zybuluo.com

就看第四点就好

2、目录结构

.
├── build
├── CMakeLists.txt  -------------①
├── src
│   ├── CMakeLists.txt ----------②
│   └── main.cpp
└── weights
    └── yolov3-mytiny_98_0.96_warehouse.onnx

3、代码

• 1）CMakeLists.txt ------①

PROJECT(TRT)
ADD_SUBDIRECTORY(src)

• 2）CMakeLists.txt ------② CMakeLists写的比较丑…

SET(common_dir /usr/src/tensorrt/samples/common)
INCLUDE_DIRECTORIES(${common_dir})

SET(cuda_dir /usr/local/cuda-10.0/targets/aarch64-linux/include)
INCLUDE_DIRECTORIES(${cuda_dir})

set(ONNX_PARSE your_path/onnx-tensorrt-release-6.0/)
INCLUDE_DIRECTORIES(${ONNX_PARSE})

SET(LOG_CPP /usr/src/tensorrt/samples/common/logger.cpp)

ADD_EXECUTABLE(trt_test main.cpp ${LOG_CPP})

find_library(LIBONNX_PATH nvonnxparser your_path/onnx-tensorrt-release-6.0/build)
TARGET_LINK_LIBRARIES(trt_test ${LIBONNX_PATH})

find_library(LIBNVINFER nvinfer /usr/lib/aarch64-linux-gnu/)
TARGET_LINK_LIBRARIES(trt_test ${LIBNVINFER})

find_library(LIBCUDART cudart /usr/local/cuda-10.0/lib64/)
TARGET_LINK_LIBRARIES(trt_test ${LIBCUDART})

• 3）main.cpp

#include <iostream>
#include "argsParser.h"
#include "buffers.h"
#include "common.h"
#include "logger.h"
#include "BatchStream.h"
#include "EntropyCalibrator.h"
#include "NvOnnxParser.h"
#include "NvInfer.h"

using namespace nvinfer1;
using namespace nvonnxparser;


int main() {
        samplesCommon::Args args;

        // 1 加载onnx模型
        IBuilder* builder = createInferBuilder(gLogger); 
        nvinfer1::INetworkDefinition* network = builder->createNetwork();

        nvonnxparser::IParser* parser = nvonnxparser::createParser(*network, gLogger);

        const char* onnx_filename="./weights/yolov3-mytiny_98_0.96_warehouse.onnx";
        parser->parseFromFile(onnx_filename, static_cast<int>(Logger::Severity::kWARNING));
        for (int i = 0; i < parser->getNbErrors(); ++i)
        {
                std::cout << parser->getError(i)->desc() << std::endl;
        }
        std::cout << "successfully load the onnx model" << std::endl;

        // 2、build the engine
        unsigned int maxBatchSize=1;
        builder->setMaxBatchSize(maxBatchSize);
        IBuilderConfig* config = builder->createBuilderConfig();
        config->setMaxWorkspaceSize(1 << 20);
        ICudaEngine* engine = builder->buildEngineWithConfig(*network, *config);

        // 3、serialize Model
        IHostMemory *gieModelStream = engine->serialize();
        std::string serialize_str;
        std::ofstream serialize_output_stream;
        serialize_str.resize(gieModelStream->size());   
        memcpy((void*)serialize_str.data(),gieModelStream->data(),gieModelStream->size());
        serialize_output_stream.open("./serialize_engine_output.trt");
        serialize_output_stream<<serialize_str;
        serialize_output_stream.close();


        // 4、deserialize model
        IRuntime* runtime = createInferRuntime(gLogger);
        std::string cached_path = "./build/serialize_engine_output.trt";
        std::ifstream fin(cached_path);
        std::string cached_engine = "";
        while (fin.peek() != EOF){ 
                std::stringstream buffer;
                buffer << fin.rdbuf();
                cached_engine.append(buffer.str());
        }
        fin.close();
        ICudaEngine* re_engine = runtime->deserializeCudaEngine(cached_engine.data(), cached_engine.size(), nullptr);

        std::cout << "Hello, World!" << std::endl;
        return 0;
}

4、异常

• 1）nvinfer1::INetworkDefinition* network = builder->createNetwork();

// 官网给的创建NetWork的方式,会报下面那个错误
const auto explicitBatch = 1U << static_cast<uint32_t>(NetworkDefinitionCreationFlag::kEXPLICIT_BATCH);  
INetworkDefinition* network = builder->createNetworkV2(explicitBatch);
// 别人的代码中的，就不会报错
nvinfer1::INetworkDefinition* network = builder->createNetwork();

错误信息

ERROR: /home/webank/tqiu/env/onnx-tensorrt-release-6.0/builtin_op_importers.cpp:661 In function importBatchNormalization:
[6] Assertion failed: scale_weights.shape == weights_shape
Assertion failed: scale_weights.shape == weights_shape
successfully load the onnx model
[06/06/2020-19:17:23] [E] [TRT] Network must have at least one output

• 2）parser->parseFromFile(onnx_filename, ILogger::Severity::kWARNING);

// 官网写法会报如下错误
parser->parseFromFile(onnx_filename, ILogger::Severity::kWARNING);
// 正确写法
parser->parseFromFile(onnx_filename,static_cast<int>(Logger::Severity::kWARNING));

错误信息

error: no matching function for call to ‘nvonnxparser::IParser::parseFromFile(const char*&, nvinfer1::ILogger::Severity)’
         parser->parseFromFile(onnx_filename, ILogger::Severity::kWARNING);