apple / ml-neuman

Official repository of NeuMan: Neural Human Radiance Field from a Single Video (ECCV 2022)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Question about method vertex_forward

buaacyw opened this issue · comments

Thanks for your great work.
I have a question about this method.
image

In this method, to get world_verts, da_pose_verts are first transformed by mesh_transf. But I think this equivalent to directly use pose_verts. In code :
mesh_transf @ da_pose_verts)
is equivalent to

    pose_verts, _= body_model(return_tensor=True,
                                               return_joints=True,
                                               poses=pose[None],
                                               betas=betas[None],
                                               transl=torch.zeros([1, 3]).float().to(device),
                                               )

I have checked this by debugging.
So I'm really confused, is this redundancy? Thanks!

Yes, I think you are right. The da_smpl here is actually just a T-pose. The body_model here should be enough.

Except you need to correctly handle the scale?

Except you need to correctly handle the scale?

Thanks for your quick reply!
But why do I need to handle the ‘scale’? I think it's fine to leave the code about scale unchanged.