RoHM icon indicating copy to clipboard operation
RoHM copied to clipboard

The official PyTorch code for RoHM: Robust Human Motion Reconstruction via Diffusion.

Results 9 RoHM issues
Sort by recently updated
recently updated
newest added

Thanks for releasing such an outstanding work, I wonder how to test it on Internet video? Now i can get the init_motion with WHAM project (contains :world_pose_root\cam_pose_root、body_pose、transl_cam\trans_world、openpose25 joints),and i want...

I'm running the whole project on a Linux remote server. And I launched Xvfb for rendering. But when I run eval_amass_full.py, when it runs into r = pyrender.OffscreenRenderer(viewport_width=W, viewport_height=H, point_size=1.0),...

Hi, Great work! If I just want to test the code (without train), do I need to download all the datasets (AMASS, PROX, EGOBODY) ?

Good work! I'm confused about the setting of num_step=1000000000 in the config file. How long does it take to train RoHM and what GPU do you use?

The preprocessing script `preprocessing_amass.py` does not generate any preprocessed data for certain AMASS subsets (CNRS, EKUT, KIT). For all data in these subsets, it consistently shows error messages like `frame...

According to this screen shot, I'm in AMASS already, but I can't find the right SMPL-X N to download and work with your project. There is no mentioned in this...

In the paper, it is mentioned "excluding first derivatives to avoid global drifting caused by inaccurate velocities." (i.e. repr_abs_only=True) With this setting, it is impossible to reconstruct the whole root...

Thanks for your great work! Why is the **motion_repr_clean** data involved in the test process, and how should it be handled during the video demo inference process without GT ?...

How to employ RoHM to smooth smplx sequence predicted by image-based HMR methods. Thanks