This is a fork of yolov5 at v3.1. Under active development and selectively including features from v4.0

1. YoloV5 Requires Python >= 3.8 and PyTorch >= 1.7.0. While NVIDIA makes PyTorch binaries available, these are for Python 3.6.9

   1. I've built a PyTorch Wheel for Python 3.8. Building on Jetson takes 12+ hours. Given it's size (250MB), it's available under Releases:

   2. See PyTorch 1.7.0, Python 3.8 for Jetson Nano (aarch64)

2. YoloV5 is not installable. This limited the useability for purposes outside of the source code. I.e. deployment with Flask

3. I found using pre-built models straightforward and wanted to add convenience methods for reusing models trained on custom data

4. The performance insights required additional dependencies.