AI Paper Atlas

Research map / 3D vision / Gaussian splatting

EDGS: Eliminating Densification for Efficient Convergence of 3DGS

Dmytro Kotovenko, Olga Grebenkova, Björn Ommer

CVPR 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
replace incremental densification with dense triangulated-correspondence initialization (EDGS)3DGS · Mip-NeRF 360 (9 scenes)LPIPS: 0.215 -> 0.141Improved
add densification to EDGS (ablation)EDGS · Mip-NeRF 360PSNR: 28.02 -> 28.08No clear effect
remove densification from plain 3DGS (ablation)3DGS · Mip-NeRF 360PSNR: 27.49 -> 25.60Hurt
use EDGS as initialization for adaptive densification methods (Taming 3DGS)Taming 3DGS · Mip-NeRF 360LPIPS: 0.207 -> 0.179Improved
use EDGS as initialization for 3DGS-MCMC3DGS-MCMC · Mip-NeRF 360LPIPS: 0.176 -> 0.159Improved
add spherical harmonics initialization (ablation)EDGS · Mip-NeRF 360LPIPS: 0.175 -> 0.141Improved
add reprojection-error sampling term p_proj (ablation)EDGS · Mip-NeRF 360LPIPS: 0.179 -> 0.175Improved
use uniform sampling over confidence-thresholded matches instead of confidence-proportional sampling (ablation)EDGS · Mip-NeRF 360LPIPS: 0.197 -> 0.179Improved
use RoMa matcher instead of RAFT (ablation)EDGS · Mip-NeRF 360PSNR: 26.90 -> 28.02Improved
use LoFTR or DKM instead of RoMa (ablation)EDGS · Mip-NeRF 360PSNR: 27.79/27.81 vs 28.02No clear effect
duplicate COLMAP points 10x or 50x with noise as a denser init (ablation)3DGS · Mip-NeRF 360PSNR: 27.49 -> 27.07 (x50, sigma 1e-3)Hurt
replace COLMAP init with random init (ablation)3DGS · Mip-NeRF 360PSNR: 27.49 -> 22.19Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 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