Dear authors,

Thank you for this outstanding work! I was particularly impressed by how you visualized the results in a global coordinate system.

I recently experimented with my personal data, which features a person lying on a bed while the videographers walk around in a full 360° view, consistently keeping the subject within the frame.

However, I encountered two issues that I hope you can help me address:

(1) Although we used the SMPLest-X checkpoints, the SMPL inference results in WHAC do not seem as accurate as in the SMPLest project alone.

(2) In cases where the people in the video remain stationary, the inferred results indicate that they are moving (e.g., moving together with the camera in 3D space). Do you have any recommendations to mitigate this issue?

(3) If we know the ground truth camera trajectory from my camera, can I use it to improve the SMPL inference?

I appreciate your time and any tips you can provide to improve the results. Thank you!

Sincerely,
David