C++ Triton YoloV4 client
Developed to infer the model deployed in Nvidia Triton Server like in Isarsoft yolov4-triton-tensorrt repo release v1.3.0, inference part based on Wang-Xinyu tensorrtx Yolov4 code and communication with server based on Triton image client example
Build client libraries
https://github.com/triton-inference-server/client/tree/r21.05
Dependencies
- Nvidia Triton Inference Server container pulled from NGC(Tested Release 21.05)
- Triton client libraries
- Protobuf, Grpc++, Rapidjson(versions according to the ones used within Triton server project. I used libraries built inside Triton Client third party folder)
- Cuda(Tested 11.3)
- Opencv4(Tested 4.2.0)
Build and compile
- mkdir build
- cd build
- cmake -DCMAKE_BUILD_TYPE=Release ..
- make
How to run
- ./yolov4-triton-cpp-client --video=/path/to/video/videoname.format
- ./yolov4-triton-cpp-client --help for all available parameters
Realtime inference test on video
- Inference test ran from VS Code: https://youtu.be/IUdbplJlspg
- other video inference test: https://youtu.be/VsENXGMNlhA