Is your feature request related to a problem? Please describe.
Not related to a problem, but it could be a useful feature to have to expand the use of pose estimation datasets to other computer vision tasks (mainly detection).

Describe the solution you'd like
Given a set of pose estimation annotations, generate corresponding bounding boxes, keeping animal ID. This could be done using convex hull algorithms or simply min/max x,y coordinates (in pixel space).

Describe alternatives you've considered
N/A

Additional context
Could this feature be relevant not only to expand the use of pose estimation datasets, but also from the analysis point of view? For example, you may want to define a "region of influence" around an animal to inspect when an animal comes close to another animal. Going forward, maybe we could explore the option of deriving panoptic segmentation masks from pose data (using the keypoints as "initial points" in SAM for example?)