huuthieu / pytorch-yolov4-tiny

YOLOv4 Tiny Pytorch & ONNX & TensorRT

---

Tutorial

https://github.com/bubbliiiing/yolov4-tiny-pytorch/blob/master/README.md

Convert .pth file to ONNX

python convert_onnx.py --model_name [...]

How to run with TensorRT

Environment

You can run in docker or in your local enviroment. Recomended run with docker.
Install NVIDIA docker for Tensorrt

docker pull nvcr.io/nvidia/tensorrt:21.07-py3 

Run container

docker run -it --runtime=nvidia -v $(pwd):/yolo-tiny --name yolov4-tiny nvcr.io/nvidia/tensorrt:21.07-py3
docker exec yolov4-tiny -it yolov4-tiny bash

Build TRT serialized file

cd /yolo-tiny
python3 build_engine.py -m yolo-tiny

Inference

python3 infer_trt.py

Reference

https://github.com/bubbliiiing/yolov4-tiny-pytorch