TensorFlow YOLOR

An attempt at implementing YOLOR in TensorFlow

Credit

This library is heavily derived from the original repo: https://github.com/WongKinYiu/yolor

Paper Link: https://arxiv.org/abs/2105.04206

Disclaimer

No guarantees that this will work properly. I have only tested it on a single dataset with limited success.

Quickstart

pip install git+https://github.com/jacoblubecki/tf-yolor.git

And then to create the YolorP6 model from the paper:

from tfyolor.model import YolorP6

n_classes = 1  # Or however many classes you need to detect.
model = YolorP6(n_classes)

TODO (Coming Soon)

• Loss Function
• Label Utilities
• NMS

Out of Scope

• Data Loading
• Full Mixed Precision Support
  • This was sort of working previously, but it was not stable.
• Multi-GPU Support
• ONNX Support
• Probably a lot of other things as well...

I don't have a ton of spare time for this project and it was mostly for fun. This was also my first TensorFlow project, so to be honest there are still a lot of APIs I haven't gotten a hang of yet, or just don't know about.

These are things I couldn't think of a nice way to port or tried to port, but ran into difficulties. If there is enough interest, I would consider tackling some of these. Feel free to open a ticket if you have a suggestion for how to solve these problems, but please be as clear as possible.

Contributing

I will consider some pull requests, but I still need to set up some CI stuff for code style, testing, and documentation. Until that other stuff is setup, I will probably take a bit to address any PRs so that I can review them more carefully.