search

TaatiTeam / MotionAGFormer

Official implementation of the paper "MotionAGFormer: Enhancing 3D Pose Estimation with a Transformer-GCNFormer Network" (WACV 2024).
Apache License 2.0
128 stars 15 forks source link

Train/Test for multiple people in the scene #19

Closed jagdishbhanushali closed 6 months ago

jagdishbhanushali commented 6 months ago

Thank you for amazing work. I have tested inference and it worked out of the box great. However I want to test this on in-the-wild videos where there are multiple people in the scene. After looking at the code I see you are always assuming there is only one person in the scene.

How to run inference as well as train for multiple person in the same frame? Do I have to separate each person's 2d pose based on tracking and then do the inference?

Thanks, Jagdish Bhanushali

SoroushMehraban commented 6 months ago

Thanks, @jagdishbhanushali. That's true when we have multiple people you have to do person detection/tracking first, crop the region containing a single person and then give it to the model.

jagdishbhanushali commented 6 months ago

Okay, Got it. Thank you