search

NVlabs / nvdiffrast

Nvdiffrast - Modular Primitives for High-Performance Differentiable Rendering
Other
1.29k stars 139 forks source link

Backpropagation of nvdiffrast.torch.rasterize #112

Closed jiaxiangshang closed 1 year ago

jiaxiangshang commented 1 year ago

Dear groupers,

Thanks again for your great job

When I use the return of nvdiffrast.torch.rasterize (u, v, z/w, triangle_id), will triangle_id support backpropagation?

Best Jiaxiang

s-laine commented 1 year ago

No, triangle_id is a discrete variable so there is no meaningful gradient.

If you're thinking of geometry gradients related to visibility/occlusion in general, those are obtained via the antialiasing op, as explained in the documentation.