gaudi-xl.md

GAUDI-XL

i've been thinking about decoupling pose and scene for a while, this is basically a demonstration of that idea.

fit a separate model as a prior over the (scene) latent. maybe an adaptor layer/module/network over a frozen TTI mode like stable diffusion or maybe even CLIP. I bet noised CLIP could work. simple way to regularize over the semantic content of a whole scene? that might be a separate idea to explore.

https://github.com/apple/ml-gaudi

throw some of these in the pot too, baby we got a stew goin!

• Fast rendering/training

  • https://github.com/NVlabs/instant-ngp
  • https://nv-tlabs.github.io/nglod/
  • https://arxiv.org/pdf/2107.02791.pdf
  • https://github.com/sunset1995/DirectVoxGO

• few-input learning

  • https://github.com/hbb1/PREF
  • https://github.com/cwchenwang/NeRF-SR

• efficient nerf

  • https://nvlabs.github.io/eg3d/
  • https://github.com/nv-tlabs/vqad

• fancy nerf

  • https://hypernerf.github.io/

from the citations

• https://github.com/apple/ml-gsn