chungyiweng / humannerf

HumanNeRF turns a monocular video of moving people into a 360 free-viewpoint video.
MIT License
786 stars 86 forks source link

New Prepare_Dataset.py for Custom Monocular training #72

Closed Dipankar1997161 closed 10 months ago

Dipankar1997161 commented 1 year ago

@chungyiweng,

I read the file prepare_wild.py and I realized that the Th and Rh values are taken from the smpl directly.

   _, tpose_joints = smpl_model(np.zeros_like(poses), betas)
    # get global Rh, Th
    pelvis_pos = tpose_joints[0].copy()
    Th = pelvis_pos
    Rh = poses[:3].copy()

However, if I have my frame-specific Rh and Th values, can I use them for implementation instead of the above method? I did not use SPIN, VIBE or ROMP for smpl generation, but rather used another method

Kindly let me know, then I can create a new prepare_dataset.py for my own case.