chris-r-u / onnx-tensorrt-libtorch

deploy onnx models with TensorRT and LibTorch

Deploying Deep Learning Models with ONNX+TensorRT+LibTorch

This project is based on https://github.com/dusty-nv/jetson-inference#api-reference.

With pretrained model exported as .onnx model, runtime engines are created and deployed along with TensorRT and LibTorch on NVIDIA Geforce or embedded Jetson platforms.

Setup

modify build.cmake to specify the path of TRT_INCLUDE_DIR and TORCH_LIB_PATH (if TensorRT was pre-installed like in Jetson, TRT_INCLUDE_DIR should not be specified), then run the commands below:

$ mkdir build

$ cp build.cmake build

$ cd build

$ cmake .. -DCMAKE_BUILD_TYPE=Release

$ make

Object Detection

Network	runtime precision	Object classes	resolution	Jetson NX
YOLOV5s	FP16	4	1920*1080	100FPS