This repo here include:

• One TensorRT GIE generate tool;
• One TensorRT batch classification class;
• One performance evaluation program;
• The dockerfile for TensorRT development;
• (TODO) A TensorRT detection class.

Note:

• All the project is implement by C++
• All images are read by OpenCV

Build & Run:

Run in docker container is recommended. Or if you want to run in your own environment, you can follow the the scripts named "run_locally.sh" in each directory.

• Generate TensorRT GIE engine

cd ${your_repo}/app/trt_gen_tool
sudo ./run_docker.sh --flagfile=../you_flagfile.ff

• Run performance evaluation program

cd ${your_repo}/app/perf_eval
sudo ./run_docker.sh --flagfile=../your_flagfile.ff

Tips:

If you encounter with "cannot find TensorRT lib" bug when you run the program in your own environment, you can fix this by:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${your_tensorrt_lib_path}