search

facebookresearch / InterHand2.6M

Official PyTorch implementation of "InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image", ECCV 2020
Other
676 stars 92 forks source link

MANO or SMPLX ? #96

Closed zeakey closed 2 years ago

zeakey commented 2 years ago

I input the 48D pose parameters in your dataset to a MANO model https://github.com/otaheri/MANO to construct 3d mesh. However, hand shape is quit weird:

1641987917(1)

In your code https://github.com/facebookresearch/InterHand2.6M/blob/2b8061d2c8e762aa6fcb8e6f5d18f8a9e83bfd0c/tool/MANO_world_to_camera/convert.py it seems that you are using SMPLX, a predecessor of MANO, not MANO for hand mesh construction.

So what should I use if I want to construct 3D mesh using your pose parameters ?

mks0601 commented 2 years ago

'smplx' contains all 3D human models including smpl, smplx, mano, and flame. The example is using only mano. Maybe you are missing 'flat_hand_mean=False'

zeakey commented 2 years ago

I see. Setting flat_hand_mean=False solved my problem.