search

davrempe / humor

Code for ICCV 2021 paper "HuMoR: 3D Human Motion Model for Robust Pose Estimation"
MIT License
511 stars 69 forks source link

Does HuMoR support separate intrinsics for each frame? #49

Closed nnop closed 2 months ago

nnop commented 2 months ago

Hi, thanks for the great motion model! I have a different situation. My data is a sequence of croped person from large images. So each image has its own size and camera center (cx, cy). Could I modify HuMoR to feed each image its own intrinsics? I wonder that is ok in theory.

davrempe commented 2 months ago

HuMoR will not work well in this setting. It is designed to recover the global motion of the person, so it assumes the camera is static. If you only have a crop from a larger image and the crop is changing at each frame, it won't be able to recover the global motion. If you have the original video with the full image size, I would recommend using this as input.