pytorchx和tensorrtx编译tensorrt模型的一些bug问题

最新推荐文章于 2024-03-20 09:52:14 发布

code_pgf

最新推荐文章于 2024-03-20 09:52:14 发布

阅读量1.2k

点赞数

文章标签： visual studio code c++ c语言

本文链接：https://blog.csdn.net/m0_60827485/article/details/124473877

版权

1、fatal error: NvInfer.h: No such file or directory

NvInfer.h是TensorRT库中的一个头文件。如果在ubuntu系统tensorRT Deb版安装成功，该头文件所在目录/usr/include/x86_64-linux-gnu/。如果在jetson系统tensorRT Deb版安装成功，该头文件所在目录/usr/include/。如果安装tensorrt是TAR和ZIP文件，那么include_directories和link_directories路径在CMakeLists.txt.

通过dpkg -L可以打印动态库和静态库中的所有内容

$ dpkg -L libnvinfer-dev 
/.
/usr
/usr/lib
/usr/lib/x86_64-linux-gnu
/usr/lib/x86_64-linux-gnu/libnvinfer_static.a
/usr/lib/x86_64-linux-gnu/libmyelin_compiler_static.a
/usr/lib/x86_64-linux-gnu/libmyelin_executor_static.a
/usr/lib/x86_64-linux-gnu/libmyelin_pattern_library_static.a
/usr/lib/x86_64-linux-gnu/libmyelin_pattern_runtime_static.a
/usr/include
/usr/include/x86_64-linux-gnu
/usr/include/x86_64-linux-gnu/NvInfer.h
/usr/include/x86_64-linux-gnu/NvInferRuntime.h
/usr/include/x86_64-linux-gnu/NvInferRuntimeCommon.h
/usr/include/x86_64-linux-gnu/NvInferVersion.h
/usr/include/x86_64-linux-gnu/NvUtils.h
/usr/share
/usr/share/doc
/usr/share/doc/libnvinfer-dev
/usr/share/doc/libnvinfer-dev/copyright
/usr/share/doc/libnvinfer-dev/changelog.Debian
/usr/lib/x86_64-linux-gnu/libmyelin.so
/usr/lib/x86_64-linux-gnu/libnvinfer.so

2. fatal error: cuda_runtime_api.h: No such file or directory

头文件cuda_runtime_api.h是cuda-cudart的头文件。在/usr/local/cuda/CMakeLists.txt中查找include_directories和link_directories的路径，然后进入相应路径。

$ dpkg -L cuda-cudart-dev-10-2           #（10-2代表cuda版本为10.2） 
/.
/usr
/usr/local
/usr/local/cuda-10.0
/usr/local/cuda-10.0/targets
/usr/local/cuda-10.0/targets/x86_64-linux
/usr/local/cuda-10.0/targets/x86_64-linux/lib
/usr/local/cuda-10.0/targets/x86_64-linux/lib/libcudadevrt.a
/usr/local/cuda-10.0/targets/x86_64-linux/lib/libOpenCL.so.1.1
/usr/local/cuda-10.0/targets/x86_64-linux/lib/libculibos.a
/usr/local/cuda-10.0/targets/x86_64-linux/lib/libcudart_static.a
/usr/local/cuda-10.0/targets/x86_64-linux/include
/usr/local/cuda-10.0/targets/x86_64-linux/include/cuda_runtime_api.h
/usr/local/cuda-10.0/targets/x86_64-linux/include/cudart_platform.h
/usr/local/cuda-10.0/targets/x86_64-linux/include/cuda_device_runtime_api.h
/usr/local/cuda-10.0/targets/x86_64-linux/include/cuda_runtime.h
/usr/lib
/usr/lib/pkgconfig
/usr/lib/pkgconfig/cudart-10.0.pc
/usr/share
/usr/share/doc
/usr/share/doc/cuda-cudart-dev-10-0
/usr/share/doc/cuda-cudart-dev-10-0/changelog.Debian.gz
/usr/share/doc/cuda-cudart-dev-10-0/copyright
/usr/local/cuda-10.0/targets/x86_64-linux/lib/libOpenCL.so
/usr/local/cuda-10.0/targets/x86_64-linux/lib/libOpenCL.so.1
/usr/local/cuda-10.0/targets/x86_64-linux/lib/libcudart.so

3、.wts not prepared or not in the right directory

模型路径错误。模型默认加载路径为build.loadWeights()函数加载模型将报如下错误：

std::map<std::__cxx11::basic_string, nvinfer1::Weights> loadWeights(std::__cxx11::string): Assertion `input.is_open() && "Unable to load weight file."' failed.
Aborted (core dumped)

4. yolo -s failed, class_num not adapted

自己训练模型的类别错误，在yololayer.h修改类别。一般报如下错误：

[Convolution]: kernel weights has count xxx but xxx was expected
void APIToModel(unsigned int, nvinfer1::IHostMemory**): Assertion `engine != nullptr' failed.
Aborted (core dumped)