Open kevinYitshak opened 1 year ago
Hi, the orientation loss here is to regularize the normal attributes attached to the neural points (not gradient computed normal). The design of this loss is inspired by the Ref-NeRF (Eq. 11), which minimize the cosine distance between gradient-computed normal and discrete normal vectors attached to points.
Thanks for the explanation!
Hi, I was trying to understand orientation loss, which is being used. Can you explain the intuition behind it and what exactly it's optimizing for?
Thanks!