search

lucidrains / pi-GAN-pytorch

Implementation of π-GAN, for 3d-aware image synthesis, in Pytorch
MIT License
115 stars 14 forks source link

3D coordinates missing #5

Open RoniPaiss opened 3 years ago

RoniPaiss commented 3 years ago

Hi,

Thank you for the implementation. I noticed you used the pixel coordinates (coors in Generator module), while in the paper they used 3D coordinates in space x = (x, y, z). Is there a reason for this difference?

lucidrains commented 3 years ago

@RoniPaiss this is like my second repository that touches on NERF and the like, so I am still learning as I go. it's quite probable that I have an error here and there!

lucidrains commented 3 years ago

@RoniPaiss embarrassingly enough, I mistook the implementation in the paper to be that of a siren 2d (R ^ 2 [x y] -> R ^ 3 [r g b]), when in reality they combined Siren with neural radiance fields, which I don't have a full understanding of. 😞 That is why I did not have the z in there

Let me read up on NERFs and understand the algorithm completely before I circle back and bring this project to completion