Modify the code for oriented bounding box

[ramdhan1989]

Hi Chen, Is it possible to extend the solution for oriented bounding box? do you have suggestion how can I start the modification.

Thanks

[WindVChen]

Hi there,

It is a good idea. Since the degraded function/branch does not rely on the BBOX angle, I believe it is feasible and easy to extend to the oriented task.

You may just refer to existing oriented methods like Yolov5-OBB for the modifications of label processing part, data augmentation part (e.g., random rotate transformation), rotated loss design (angle loss and rotated IOU loss?). The intermediate model structure can keep fixed.

While I'm not currently available to work on this function at the moment, if you achieve the successful adaption, I would be more than happy to receive a PR from you and see how we can integrate it.

[ramdhan1989]

Thanks for the guidance. I'll try to modify it.

thanks