Rust FFI bindings for shouxieai/tensorRT_Pro/example-simple_yolo. This is an example of creating an rust wrapper for shouxieai/tensorRT_Pro. More idiomatic rust bindings could then be developed on top of this.
libyolo is modified from shouxieai/tensorRT_Pro/example-simple_yolo.
You can use this lib to call yolov5-tensorrt from rust.
Build your docker environment under the instructions of https://hub.docker.com/r/hopef/tensorrt-pro.
Modify libyolo/CMakeLists.txt according to your own environment.
Set the correct CUDA_GEN_CODE with your own gpu. If you don't know you cuda gencode, you can look for it at https://developer.nvidia.com/zh-cn/cuda-gpus#compute. The gpu compute capability for my NVIDIA TITAN X is 6.1.
set(CUDA_GEN_CODE "-gencode=arch=compute_61,code=sm_61")
Set the correct paths of opencv, cuda, cudnn and tensorrt that you downloaded under the instructions of https://hub.docker.com/r/hopef/tensorrt-pro.
set(OpenCV_DIR "/nfs/users/chenquan/packages/tensorrt_pro/data/lean/opencv-4.2.0/include/opencv4")
set(CUDA_DIR "/nfs/users/chenquan/packages/tensorrt_pro/data/lean/cuda-11.2")
set(CUDNN_DIR "/nfs/users/chenquan/packages/tensorrt_pro/data/lean/cudnn8.2.2.26")
set(TENSORRT_DIR "/nfs/users/chenquan/packages/tensorrt_pro/data/lean/TensorRT-8.0.3.4.cuda11.3.cudnn8.2")
Change the opencv path in build.rs used by clang_arg.
Then, simply run
cargo build
Modify the model and image paths in src/ib.rs.
Yolov5 onnx file are obtained under the instuctions of shouxieai/tensorRT_Pro.
Now run the following command to compile the tensorrt engine
RUST_BACKTRACE=1 cargo test test_compile_tensorrt_engine --lib -- --nocapture
If you met a cudnn path not found error, export the path of cudnn in LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/nfs/users/chenquan/packages/tensorrt_pro/data/lean/cudnn8.2.2.26/lib:$LD_LIBRARY_PATH
And run the following command to run the tensorrt engine
RUST_BACKTRACE=1 cargo test test_run_engine --lib -- --nocapture