search

david-svitov / HAHA

HAHA: Highly Articulated Gaussian Human Avatars with Textured Mesh Prior
BSD 3-Clause "New" or "Revised" License
117 stars 14 forks source link

Rendering Novel Pose with Trained X-Human Data #15

Open yc4ny opened 4 months ago

yc4ny commented 4 months ago

Hi, thanks for the amazing work.

I have a question on generating an animation with an avatar trained with the X-Human Dataset with a AMASS sequence. In the .yaml config file that you have provided, I have added:

  animation_novel_poses:
    target: callbacks.animation_logger.AnimationLogger
    params:
      animation_sequence: data/AMASS/CMU-SMPLX-N/105/105_11_stageii.npz

but it doesn't seem to work and gets this error: RuntimeError: shape '[-1, 15, 3]' is invalid for input of size 12

It seems like a data size mismatch with the smplx_input, but I'm not sure how to fix this error.

Do you have any insight on this?

Thanks.

david-svitov commented 4 months ago

Hello! It's difficult to say what exactly the problem is. Could you please provide the entire stack trace of the error indicating which line and function it occurred on. Meanwhile, I can assume that the problem is num_pca_comps=12. Try to set use_pca_for_hands: False in the config.