Integrating supervision annotators

[hardikdava]

This PR introduces converting predicitons to supervision.Detections format which can be directly useful to create different visual annotations and unleash full potentials of supervision. This will also reduces annotations responsibility to supervision. Further detections can be directly useful for some of the post process steps e.g. combining multiple predictions and some of the dataset utilities.

Ref:
https://supervision.roboflow.com/latest/detection/core/
https://supervision.roboflow.com/latest/detection/annotators/

Sample output:

[henrytsui000]

Let me take some time to review the pull request—it looks like it could be really helpful!

[hardikdava]

@henrytsui000 @madmike2010 I have made changes that you requested. Please review again.

[hardikdava]

@henrytsui000 A side question, is there any list of open tasks? That will help external contributes to manage task.

[henrytsui000]

@henrytsui000 A side question, is there any list of open tasks? That will help external contributes to manage task.

Yes, I do have a list of tasks waiting for implementation. However, the top priority right now is refining the training strategy. Our model converges slower after the 10th epoch compared to other YOLO models, which is currently the most annoying issue.

[hardikdava]

@henrytsui000 Thanks for reply. If possible please include your open task list to read me. Also please merge this PR.