Efficient way to get ManoLayer's jacobian matrix

[milkcat0904]

Hello, thx for sharing your code, I got some issues about calculate ManoLayer's jacobian matrix in my project.
I use pytorch official api to get ManoLayer's jacobian but it takes too much time, about 0.2s each sample 1 Nvidia RTX 3090
Here is my code:

from torch.autograd.functional import jacobian
 j = jacobian(mano_layer.forward, (theta, beta)) 

btw,
I only calculate joints' jacobian matrix, if add verts' jacobian matrix, it cost a larger time.
Is there any way to get jacobian more efficient?
How could I get ManoLayer's analytical differential expression?

[hassony2]

Hi @milkcat0904,

Sorry for the late answer !
Computing vertices and joints in mano is done jointly and it is not easy (if even feasible) to reduce the code only to joint computations.
So unfortunately at this time no good solution for speeding up the computations comes to mind !

Best,

Yana