DATENeRF: Depth-Aware Text-Based Editing of NeRFs

Sara Rojas*, Julien Philip, Kai Zhang, Sai Bi, Fujun Luan, Bernard Ghanem, Kalyan Sunkavalli

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Recent diffusion models have demonstrated impressive capabilities for text-based 2D image editing. Applying similar ideas to edit a NeRF scene [31] remains challenging as editing 2D frames individually does not produce multiview-consistent results. We make the key observation that the geometry of a NeRF scene provides a way to unify these 2D edits. We leverage this geometry in depth-conditioned ControlNet [57] to improve the consistency of individual 2D image edits. Furthermore, we propose an inpainting scheme that uses the NeRF scene depth to propagate 2D edits across images while staying robust to errors and resampling issues. We demonstrate that this leads to more consistent, realistic and detailed editing results compared to previous state-of-the-art text-based NeRF editing methods.

Original languageEnglish (US)
Title of host publicationComputer Vision – ECCV 2024 - 18th European Conference, Proceedings
EditorsAleš Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, Gül Varol
PublisherSpringer Science and Business Media Deutschland GmbH
Pages267-284
Number of pages18
ISBN (Print)9783031732461
DOIs
StatePublished - 2025
Event18th European Conference on Computer Vision, ECCV 2024 - Milan, Italy
Duration: Sep 29 2024Oct 4 2024

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume15069 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference18th European Conference on Computer Vision, ECCV 2024
Country/TerritoryItaly
CityMilan
Period09/29/2410/4/24

Keywords

  • 3D Scene Editing
  • Diffusion Models
  • Neural Rendering

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'DATENeRF: Depth-Aware Text-Based Editing of NeRFs'. Together they form a unique fingerprint.

Cite this