Hello everyone, I am coming from the old Pytorch3d Library and I used a lot the differentiable chamfer distance loss fucntion from there. I see that here Chamfer Distance is only a metric with no backward method implemented.

The main point is that through differentiation we can obtain deformation fields to match surfaces, as in this tutorial. I have a paper about this in which I show that smooth deformation fields can be obtained from the chamfer distance gradients using some form of regularization.

Do you think this can be implemented? I would like to port my software to kaolin instead of pytorch3d.