G3Splat: Geometrically Consistent Generalizable Gaussian Splatting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hosseinzadeh, Mehdi, Chng, Shin-Fang, Xu, Yi, Lucey, Simon, Reid, Ian, Garg, Ravi
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914210355609600
author Hosseinzadeh, Mehdi
Chng, Shin-Fang
Xu, Yi
Lucey, Simon
Reid, Ian
Garg, Ravi
author_facet Hosseinzadeh, Mehdi
Chng, Shin-Fang
Xu, Yi
Lucey, Simon
Reid, Ian
Garg, Ravi
contents 3D Gaussians have recently emerged as an effective scene representation for real-time splatting and accurate novel-view synthesis, motivating several works to adapt multi-view structure prediction networks to regress per-pixel 3D Gaussians from images. However, most prior work extends these networks to predict additional Gaussian parameters -- orientation, scale, opacity, and appearance -- while relying almost exclusively on view-synthesis supervision. We show that a view-synthesis loss alone is insufficient to recover geometrically meaningful splats in this setting. We analyze and address the ambiguities of learning 3D Gaussian splats under self-supervision for pose-free generalizable splatting, and introduce G3Splat, which enforces geometric priors to obtain geometrically consistent 3D scene representations. Trained on RE10K, our approach achieves state-of-the-art performance in (i) geometrically consistent reconstruction, (ii) relative pose estimation, and (iii) novel-view synthesis. We further demonstrate strong zero-shot generalization on ScanNet, substantially outperforming prior work in both geometry recovery and relative pose estimation. Code and pretrained models are released on our project page (https://m80hz.github.io/g3splat/).
format Preprint
id arxiv_https___arxiv_org_abs_2512_17547
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle G3Splat: Geometrically Consistent Generalizable Gaussian Splatting
Hosseinzadeh, Mehdi
Chng, Shin-Fang
Xu, Yi
Lucey, Simon
Reid, Ian
Garg, Ravi
Computer Vision and Pattern Recognition
3D Gaussians have recently emerged as an effective scene representation for real-time splatting and accurate novel-view synthesis, motivating several works to adapt multi-view structure prediction networks to regress per-pixel 3D Gaussians from images. However, most prior work extends these networks to predict additional Gaussian parameters -- orientation, scale, opacity, and appearance -- while relying almost exclusively on view-synthesis supervision. We show that a view-synthesis loss alone is insufficient to recover geometrically meaningful splats in this setting. We analyze and address the ambiguities of learning 3D Gaussian splats under self-supervision for pose-free generalizable splatting, and introduce G3Splat, which enforces geometric priors to obtain geometrically consistent 3D scene representations. Trained on RE10K, our approach achieves state-of-the-art performance in (i) geometrically consistent reconstruction, (ii) relative pose estimation, and (iii) novel-view synthesis. We further demonstrate strong zero-shot generalization on ScanNet, substantially outperforming prior work in both geometry recovery and relative pose estimation. Code and pretrained models are released on our project page (https://m80hz.github.io/g3splat/).
title G3Splat: Geometrically Consistent Generalizable Gaussian Splatting
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.17547