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
675 stars 92 forks source link

How to transform 3D joints set to MANO angle axis representation? #123

Closed Maradowei closed 1 year ago

Maradowei commented 1 year ago

I want to know How to transform 3D joints set to MANO angle axis representation? thanks.

mks0601 commented 1 year ago

As I said in the earlier issue, you cannot directly transform as it is an ill-posed problem. This is called inverse kinematics. Please read my paper (NeuralAnnot), which is used to obtain MANO parameters of this dataset.

Maradowei commented 1 year ago

Thanks very much.

Maradowei commented 1 year ago

I think I don't know what the 45 hand pose parameters (angle axis) mean. I compute each angle axis of 3 points from 21 3D joints following the kinematic tree. For example, I compute the angle axis of Vector(J0->J1) and Vector(J1->J2), and I stacked the 15 angle axis as the mano parameter. And I set random shape parameters, but the resulting mesh looks weird.

mks0601 commented 1 year ago

You'd better study 6 dof. You cannot get twist rotation from points

Maradowei commented 1 year ago

Tanks

Maradowei commented 1 year ago

Very sorry to interrupt again. I have got close 3D coordinates in the camera coordinate system somehow, and I want to use the joints to manually do something on mano parameters rather than just feed them in the network, I have used the root location to the trans parameter of mano and it seems good. Is there any way to use these joints manually? Thanks very much.