[Feature] how to differentiate keypoints of different humans in multi-person video?

[iPsych]

What is the feature?

Some example has 'get_track_id' in earlier version, but recent demos doesn't have the option. When the examples are run, visual output shows successful tracking of the keypoints, but only (randomly selected?) first person's keypoints are extracted.

Is there any way to get all people's keypoints json output in body3d_two_stage_video_demo.py?

I need to separate keypoints from 2 (at least) people, But when I set num_instances to 2, only very first frame is processed. (num_instances -1 works, but does not provide person id at all.)

Any other context?

No response

[iPsych]

Usually track_id is supposed to work, but the videos work with only "num_instances=-1" and with ""num_instances=-1" "track_id" is not working.