yohanshin / WHAM

MIT License
685 stars 74 forks source link

multiple person video input #30

Open linjiangya opened 9 months ago

linjiangya commented 9 months ago

Dear authors,

Does the current demo support videos with multiple people?

image

It seems only one person is visualized on the right side of the demo.

yohanshin commented 9 months ago

Hi @linjiangya

Currently, WHAM outputs world-grounded pose and shape of all people in the scene. However, their global trajectories are estimated individually, thus, we don't have such an implementation to constrain everyone on the same world coordinates.

We will implement a naive solution later that will anchor a single world coordinate on a representative individual and visualize all other people accordingly -- this is just for visualization purposes. But at the current stage, we only support to visualize the global-trajectory for only one person in the scene.

FrankBu0616 commented 9 months ago

Just want to follow up on this thread. Amazing work! Where can we find global coordinate information for each individual? Also, any thoughts on how to map the camera motion into the world coordinates as well?

WangJYao commented 6 months ago

I also want to know, hope you can let us know. Thanks!