SpotLessSplats: Ignoring Distractors in 3D Gaussian Splatting
Three-dimensional Gaussian Splatting (3DGS) is a promising technique for 3D reconstruction, offering efficient training and rendering speeds, making it suitable for real-time applications. However, current methods require highly controlled environments–no moving people or wind-blown elements, and consistent lighting–to meet the interview consistency assumption of 3DGS. This makes reconstruction of real-world captures problematic. We present SpotLessSplats, an approach that leverages pre-trained and general-purpose features coupled with robust optimization to effectively ignore transient distractors. Our method achieves state-of-the-art reconstruction quality both visually and quantitatively, on casual captures.
Authors
- Georgios Kopanas (ORCID: https://orcid.org/0009-0002-5829-2192)
- Mark A. Matthews (ORCID: https://orcid.org/0009-0005-9725-5978)
- Lily Goli
- Sara Sabour (ORCID: https://orcid.org/0000-0001-7965-2113)
- David J. Fleet (ORCID: https://orcid.org/0000-0003-0734-7114)
- Andrea Tagliasacchi (ORCID: https://orcid.org/0000-0002-2209-7187)
- Alec Jacobson (ORCID: https://orcid.org/0000-0003-4603-7143)
- Leonidas Guibas (ORCID: https://orcid.org/0000-0002-8315-4886)
- Dmitry Lagun
Institutions
- Google (United States) (US)
- Simon Fraser University (CA)
- University of Toronto (CA)
- Google (United Kingdom) (GB)
- Google (Canada) (CA)
- Stanford University (US)
Publication Details
- Journal
- ACM Transactions on Graphics
- Published
- 2025-03-29
- DOI
- https://doi.org/10.1145/3727143
- Citations
- 14
- Primary Topic
- Computer Graphics and Visualization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 49.57