AI Paper Atlas

Research map / 3D vision / Gaussian splatting

GaussFusion: Improving 3D Reconstruction in the Wild with A Geometry-Informed Video Generator

Liyuan Zhu, Manjunath Narayana, Michal Stary, Will Hutchcroft, Gordon Wetzstein, Iro Armeni

CVPR 2026 · 3D vision › Gaussian splatting · approach: Deformable 4D Gaussian splatting · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add depth to RGB input in GP-Buffer (ablation)GaussFusion · DL3DV (1/3 data)FID: 15.451 -> 10.542Improved
add normal map (ablation)GaussFusion · DL3DV (1/3 data)PSNR: 19.289 -> 19.744Improved
add alpha map (ablation)GaussFusion · DL3DV (1/3 data)PSNR: 19.744 -> 19.958Improved
add inverse covariance (uncertainty) map (ablation)GaussFusion · DL3DV (1/3 data)PSNR: 19.958 -> 20.753Improved
use proposed artifact simulation instead of uniform downsampling and underfitting (ablation)GaussFusion · DL3DVPSNR: 21.087 -> 22.548Improved
add Geometry Adapter instead of convolutional fusion (ablation)GaussFusion · DL3DVPSNR: 20.904 -> 22.548Improved
joint training on all datasets and artifact types versus DL3DV-only (ablation)GaussFusion · DL3DVPSNR: 22.266 -> 22.548Improved
joint training on all datasets versus DL3DV-only (ablation)GaussFusion · RE10KPSNR: 24.754 -> 28.652Improved
distill to 4-step generatorGaussFusion · DL3DV and RE10KFID: FID 3.933 -> 7.383 (DL3DV); 4.3 -> 15.11 FPSMixed
refine DepthSplat output with Difix3D+ or ExploreGSDepthSplat · RE10KPSNR: 21.773 -> 21.415 (Difix3D+); -> 21.195 (ExploreGS)Hurt
refine DepthSplat output with GaussFusionDepthSplat · RE10KPSNR: 21.773 -> 22.802Improved
refine MVSplat output with GaussFusionMVSplat · RE10KPSNR: 18.744 -> 19.756Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy