Closed russelldj closed 4 months ago
I'm pretty sure that #73 could be used as an alternative to this. Rather than using a specific function to check which face a pixel corresponds to, we could simply render a texture that encodes the face ID. This might require doing multiple renders if there are more than 256^3
faces, but I think that's workable.
This could be an alternative to pytorch3d pixel-to-face correspondence computation. I'm not sure how the runtimes would compare, or if it does occlusion checking properly. But it would reduce the complexity by not depending on
pytorch3d
, which is difficult to install.