RoGUENeRF - A Robust Geometry-Consistent Universal Enhancer for NeRF

Sibi Catley-Chandar Richard Shaw Gregory Slabaugh Eduardo Pérez-Pellitero

Method

RoGUENeRF is a universal NeRF enhancer which leverages geometry-aware 3D alignment and 2D refinement to enhance any NeRF model while remaining view-consistent and being robust to inaccurate camera poses.

Method Overview

Results

If videos do not load, please try using Mozilla Firefox or download the video from the viewer.

MipNeRF360 - Treehill

Nerfacto - Garden

NeRF - Trex

View Consistency

Nerfacto - Room

Nerfacto (left) vs Nerfacto + Ours (right)

NeRFLiX - Room

Nerfacto + NeRFLiX (left) vs Nerfacto + Ours (right)

MipNeRF360 - Treehill

MipNeRF360 (left) vs MipNeRF360 + Ours (right)

NeRFLiX - Treehill

MipNeRF360 + NeRFLiX (left) vs MipNeRF360 + Ours (right)

Large Gaussian Camera Noise - Nerfacto

Nerfacto + Large Camera Noise (left) vs Nerfacto + Large Camera Noise + Ours (right)

Large Gaussian Camera Noise - NeRFLiX

Nerfacto + Large Camera Noise + NeRFLiX (left) vs Nerfacto + Large Camera Noise + Ours (right)

Medium Gaussian Camera Noise - Nerfacto

Nerfacto + Medium Camera Noise (left) vs Nerfacto + Medium Camera Noise + Ours (right)

Medium Gaussian Camera Noise - NeRFLiX

Nerfacto + Medium Camera Noise + NeRFLiX (left) vs Nerfacto + Medium Camera Noise + Ours (right)

@article{catleychandar2024roguenerf,
      title={RoGUENeRF: A Robust Geometry-Consistent Universal Enhancer for NeRF}, 
      author={Sibi Catley-Chandar and Richard Shaw and Gregory Slabaugh and Eduardo Perez-Pellitero},
      year={2024},
      eprint={2403.11909},
      journal={arXiv.org}
}