mks0601 / Hand4Whole_RELEASE

Official PyTorch implementation of "Accurate 3D Hand Pose Estimation for Whole-Body 3D Human Mesh Estimation", CVPRW 2022 (Oral.)
MIT License
305 stars 30 forks source link

Question about SMPLX hand only prediction #51

Closed buaacyw closed 1 year ago

buaacyw commented 1 year ago

Hi! Thanks for your work. I want to ask if it is possible to perform hand-only prediction for SMPLX models. I have seen the hand-only prediction for MANO models. I wonder if it is possible to translate the MANO prediction to SMPLX prediction (only the hand part of SMPLX). Thanks!

mks0601 commented 1 year ago

What do you mean by only the hand part of SMPLX? As hands are connected with body in SMPLX, kinematic chain of hands is connected to the body. That means, ancestor joints (body joints) should be supervised as well for hand prediction. For example, given two hand/finger positions, you can randomly place other body parts such as elbow or lower body parts

buaacyw commented 1 year ago

Thanks for your quick reply! Yes. That is exactly what I want. I don't care about other body parts of the SMPLX. I only need the hand parts (just one hand) to fit the image. I found in SMPLX repo that MANO vertices can be mapped to the SMPLX vertices. But I need the whole SMPLX model. So Is this possible?

mks0601 commented 1 year ago

You can make rotations of other parts to zero and make 3D wrist local rotation to the 3D global hand rotation.

buaacyw commented 1 year ago

You can make rotations of other parts to zero and make 3D wrist local rotation to the 3D global hand rotation.

Thanks! I think this is feasible.