Clarification on fragments.zbuf

[yuya737]

A simple clarification: the output from somerasterizer.forward() (i.e. the fragment in pytorch3d language) and in particular the fragment.zbuf tensor should be differentiable with respect to the input parameters like cameras correct?

A beginner here....

[bottler]

Yes, all the rasterizers in PyTorch3D except MeshRasterizerOpenGL are differentiable, and for them you should see gradients flowing back from the depths to the cameras.