Mano parameters

[anilesec]

@samarth-robo issue continuing from an email conversation

Hi Samarth,
​
I have one more question, it would be a great help if you could clarify it as well. Thanks in Advance! 
1. I see there are only 6 mano_fits_**.json given for each sequence. How can we obtain the mano parameters of hand in each frame? More precisely, Is there a way to get mano axis angle parameters for each frame in the contactpoose dataset?
​
Looking forward to hearing from you!

Hi Anil,
 
Those 6 json files represent 6 different MANO parameter sizes. The MANO model allows you to represent the hand pose with different parameter sizes through PCA.
 
They don't correspond to frame numbers.
 
Regardless of which one you choose from those 6, you can get that hand model for each frame in the sequence. Please see the [mano_meshes()](https://github.com/facebookresearch/ContactPose/blob/main/utilities/dataset.py#L300) function of the ContactPose dataset.
 
These meshes are in the object coordinate frame. Then you can use [ContactPose.object_pose()](https://github.com/facebookresearch/ContactPose/blob/main/utilities/dataset.py#L278) to transform them into the camera coordinate frame.
 
You can see a working example of all this in [this demo notebook](https://github.com/facebookresearch/ContactPose/blob/main/rendering.ipynb), where the posed MANO meshes are used to render hand masks in the image.
 
Please use GitHub issues for these questions, so the answers are publicly documented and others can see them later if they have the same questions.

​

[anilesec]

Thanks, @samarth-robo for the response.
I think I understand what you mean. This means that MANO pose and shape parameters are the same for all the frames (as hand grasp is fixed) however to get the global orientation of hand in each frame, we will have to use the rotation component of the ContactPose.object_pose().

[anilesec]

I understood how to get the meshes(of hand) for each frame. But, I am interested in mano parameters for each frame( the meshes of each frame are trivial for my application). Also, in mano_fits_10.json there are 13 parameters where the last three parameters correspond to the global orientation of the hand. Does this global orientation correspond to the orientation of the hand in the first frame of the sequence or is it random orientation?

Thank you!

[samarth-robo]

Thanks, @samarth-robo for the response. I think I understand what you mean. This means that MANO pose and shape parameters are the same for all the frames (as hand grasp is fixed) however to get the global orientation of hand in each frame, we will have to use the rotation component of the ContactPose.object_pose().

correct

[samarth-robo]

I understood how to get the meshes(of hand) for each frame. But, I am interested in mano parameters for each frame( the meshes of each frame are trivial for my application). Also, in mano_fits_10.json there are 13 parameters where the last three parameters correspond to the global orientation of the hand. Does this global orientation correspond to the orientation of the hand in the first frame of the sequence or is it random orientation?

Thank you!

no it is not a random rotation. It is a necessary output of the optimizer that optimizes the MANO params for L2 distance of joint 3D locations from ground truth.

Here is the chain of transforms if you are curious:

Denote the output dict of ContactPose.mano_params() by mp, and start with a hand vertex m0_p. The coordinate system m0 corresponds to MANO's PCA vertex regressor (internal detail you can understand if you read the MANO paper).

1. rotation by 3 parameters in mp['pose']) -> hand vertex in MANO coordinates m_p
2. apply mp['hTm'] -> hand vertex in hand coordinates h_p. mp['hTm'] is the inverse of mTc and mTc is a rigid body transform I remove before constructing the optimization objective, to make the optimization easier.
3. apply ContactPose._oTh -> hand vertex in object coordinates o_p. oTh is usually identity, but it can be different if the hand is dynamic w.r.t. object, which happens in some double-handed grasps
4. apply cTo, the output of ContactPose.object_pose() -> hand vertex in camera coordinates c_p

This is just FYI. You don't need to worry about applying steps 1, 2, and 3: ContactPose.mano_meshes() does that for you.

[anilesec]

Thanks for elucidating! I will use the per-frame MANO parameters(with object pose as the global orientation of each frame) and see how it works.

[gs-ren]

@anilesec Hi, could you share how to get the mano pose for each frame?

[anilesec]

@gs-ren you can find the info here #16 (comment)

[gs-ren]

@gs-ren you can find the info here #16 (comment)
Could you describe step 4 process in detail please? I want to get the pose, 3d points and vertex in camera coordinate. Thank you! @anilesec