the parameters of the mano_layer

facebookresearch / InterWild

Official PyTorch implementation of "Bringing Inputs to Shared Domains for 3D Interacting Hands Recovery in the Wild", CVPR 2023

Other

159 stars 16 forks source link

the parameters of the mano_layer #25

Open luckyday2022 opened 6 months ago

luckyday2022 commented 6 months ago

When the model is trained, are the parameters of the mano_layer layer updated?

mks0601 commented 6 months ago

No, it is not included in the trainable module https://github.com/facebookresearch/InterWild/blob/b570c122e506936e646013bba077b3afb51dae4b/main/model.py#L35

luckyday2022 commented 6 months ago

The following is my understanding of coordinates. Can you help me answer it?

with torch.no_grad(): output = mano.layer[hand_type](betas=shape, hand_pose=hand_pose, global_orient=root_pose, transl=trans) Is the output here the coordinates in the camera coordinate system? Why are these two processes different?

mks0601 commented 6 months ago

The upper one is for data loader to make groundtruth. The second one is for the forward function.