Learning compact 3D Gaussians via feed-forward point fusion

Conference item

We present Splatt3RFusion, a feed-forward neural network that, given a set of unposed and uncalibrated images, directly reconstructs a compact and high-quality 3D Gaussian Splat representation of a scene. Unlike prior pixel-aligned feed-forward methods that typically predict one 3D Gaussian primitive per pixel in each image - producing severe redundancy, duplication, and ghosting on one physical surface - our approach efficiently fuses points in 3D space through a multi-scale octree structure, yielding a compact and coherent representation. Built upon VGGT, a foundation model for pose-free 3D geometry prediction, Splatt3RFusion introduces a Gaussian prediction branch that infers primitive parameters using only photometric supervision. We also introduce the ability to control the number of 3D Gaussians generated at test-time, allowing for a controllable tradeoff between PSNR and the number of 3D Gaussian primitives used. The model is efficient, reducing both memory usage and rendering cost, while achieving state-of-the-art results on RealEstate10k and ScanNet++.

Authors: Smart, B; Zheng, C; Laina, I; Prisacariu, VA

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: IEEE
• Host title: 2026 International Conference on 3D Vision (3DV)
• Pages: 554-564
• Publication date: 2026-05-27
• Event title: 13th International Conference on 3D Vision (3DV 2026)
• Event location: Vancouver, BC, Canada
• Event website: https://3dvconf.github.io/2026/
• Event start date: 2026-03-20
• Event end date: 2026-03-23
• DOI: 10.1109/3dv69130.2026.00059
• EISSN: 2475-7888
• ISSN: 2378-3826
• EISBN: 9798331573126
• ISBN: 9798331573133
• Language: English
• Keywords: networking and information technology r&d (nitrd)