yolov5模型转TensorRT踩坑记录

按照 https://codechina.csdn.net/mirrors/enazoe/yolo-tensorrt?utm_source=csdn_github_accelerator 转换遇到的问题记录。

问题1

找不到tensorRT，报错找不到 NvInfer.h
解决：
如果tensorRT为下载解压安装的，需要修改CMakeLists.txt：

# tensorrt
include_directories(/home/jiang/TensorRT-7.0.0.11/include)
link_directories(/home/jiang/TensorRT-7.0.0.11/lib)

问题2

报错：

/home/jiang/TensorRT-7.0.0.11/include/NvInfer.h:3290:22: note: declared here
 class TRT_DEPRECATED IPluginLayer : public ILayer
                      ^~~~~~~~~~~~
[100%] Linking CXX executable yolov5
libmyplugins.so：对‘cv::String::deallocate()’未定义的引用
collect2: error: ld returned 1 exit status
CMakeFiles/yolov5.dir/build.make:177: recipe for target 'yolov5' failed
make[2]: *** [yolov5] Error 1
CMakeFiles/Makefile2:104: recipe for target 'CMakeFiles/yolov5.dir/all' failed
make[1]: *** [CMakeFiles/yolov5.dir/all] Error 2
Makefile:83: recipe for target 'all' failed
make: *** [all] Error 2

解决：
增加OpenCV的include路径。

find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})

问题3

运行sudo ./yolov5 -s 时出现 libmyelin.so.1找不到的问题。
解决：

./yolov5: error while loading shared libraries: libmyelin.so.1: cannot open shared object file: No such file or directory

以下不知道是否有效果，参考。也可以直接忽略，去掉sudo运行。

locate libmyelin.so.1
sudo gedit ~/.bashrc  #加入上面的文件夹到路径
#如：export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/jiang/TensorRT-7.0.0.11/targets/x86_64-linux-gnu/lib
source  ~/.bashrc
sudo ldconfig

问题4

运行转换出现：

Yolov5_DeepSort_Pytorch-master/tensorrtx/yolov5/yolov5.cpp:426: void APIToModel(unsigned int, nvinfer1::IHostMemory**): Assertion `engine != nullptr' failed.
已放弃 (核心已转储)

解决：参考
我的问题是类别数没有修改，yololayer.h中 修改static constexpr int CLASS_NUM = 1;默认是80。
然后再make，运行。