Fast Darknet YOLO v3/v4 inference by means of TensorRT and PyTorch (optionally)
This project consists of 3 main parts:
- YOLO to ONNX;
- ONNX to TensorRT;
- Main processing class.
Darknet YOLO to ONNX conversion based on NVidia example. You may have seen such way of conversion in some repos, but mainly authors doesn't change anything in this script. The most important, they leave "yolo"-layer unsupported, and hence not converted. In this repository I have added "yolo"-layer to ONNX graph. Also, it's able to parse cfg-file deeper then original script, and support YOLO v4.