AI Paper Atlas

Research map / 3D vision / Gaussian splatting

GSFixer: Improving 3D Gaussian Splatting with Reference-Guided Video Diffusion Priors

Xingyilang Yin, Qi Zhang, Jiahao Chang, Ying Feng, Qingnan Fan, Xi Yang, Chi-Man Pun, Huaqi Zhang, Xiaodong Cun

ICML 2026 · 3D vision › Gaussian splatting · approach: Compact and efficient Gaussian splatting

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 VGGT 3D geometric tokens (ablation)GSFixer · DL3DV-ResPSNR: 16.36 -> 16.72Improved
add DINOv2 2D semantic tokens (ablation)GSFixer · DL3DV-ResPSNR: 16.48 -> 16.72Improved
add 3D tokens (ablation)GSFixer · DL3DV-ResLPIPS: 0.414 -> 0.399Improved
add 2D tokens (ablation)GSFixer · DL3DV-ResLPIPS: 0.409 -> 0.399Improved
use reference-guided trajectory instead of interpolation trajectory (ablation)GSFixer · Mip-NeRF 360 (3-view)PSNR: 14.50 -> 15.61Improved
use reference-guided trajectory instead of ellipse trajectory (ablation)GSFixer · Mip-NeRF 360 (3-view)PSNR: 15.46 -> 15.61Improved
use reference-guided trajectory instead of ellipse trajectory (ablation)GSFixer · Mip-NeRF 360 (3-view)LPIPS: 0.563 -> 0.559Improved
GSFixer restoration instead of GenFusionvideo restoration · DL3DV-ResPSNR: 14.56 -> 16.72Improved
apply GSFixer to 3DGS with 3 input views3DGS · DL3DV-Benchmark (3-view)PSNR: 13.72 -> 16.21Improved
GSFixer vs Difix3D+ with 9 views, LPIPSGSFixer · DL3DV-Benchmark (9-view)LPIPS: 0.259 -> 0.307Hurt
GSFixer vs ReconFusion (NeRF-based generative)GSFixer · Mip-NeRF 360 (3-view)PSNR: 15.50 -> 15.61Improved
Difix3D+ and GenFusion restoration versus unrestored frames on consistencyDifix3D+, GenFusion · DL3DV-ResI2V_SC: 0.9382 -> 0.9288Hurt

Later papers that compared with this method

1 later accepted paper reports GSFixer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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