AI Paper Atlas

Research map / 3D vision / Gaussian splatting

Pushing Rendering Boundaries: Hard Gaussian Splatting

Qingshan Xu, Jiequan Cui, Xuanyu Yi, Yuxuan Wang, Yuan Zhou, Yew-Soon Ong, Hanwang Zhang

AAAI 2026 · 3D vision › Gaussian splatting · approach: Regularization and primitive design for 3DGS · 3 citations (OpenAlex)

Read the paper

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 positional gradient driven HGS (PGHGS) (ablation)Scaffold-GS · Tanks&TemplesLPIPS: 0.170 -> 0.129Improved
add rendering error guided HGS (REHGS) (ablation)Scaffold-GS · Tanks&TemplesLPIPS: 0.170 -> 0.155Improved
add REHGS on top of PGHGS (ablation)Model-A · Tanks&TemplesLPIPS: 0.129 -> 0.126Improved
add Effi-HGS to existing Gaussian methods3DGS, Pixel-GS*, Mip-Splatting*, HAC · Tanks&TemplesLPIPS: HAC 0.173 -> 0.150; 3DGS 0.178 -> 0.163Improved
replace Scaffold-GS growing with full HGSScaffold-GS · Mip-NeRF360PSNR: 27.98 -> 28.30Improved
replace Scaffold-GS growing with full HGSScaffold-GS · Mip-NeRF360SSIM: 0.825 -> 0.837No clear effect
replace Scaffold-GS growing with full HGSScaffold-GS · Tanks&TemplesLPIPS: 0.170 -> 0.126Improved
replace Scaffold-GS growing with full HGSScaffold-GS · Deep BlendingLPIPS: 0.245 -> 0.225Improved
use HGS instead of Scaffold-GS growing (efficiency cost)Scaffold-GS · Tanks&Templestraining time and FPS: 20.4m -> 66.2m; 125 -> 42 FPSHurt
use Effi-HGS instead of full HGSHGS · Tanks&Templestraining time and FPS: 66.2m -> 28.6m; 42 -> 87 FPSImproved
lower k in PGHGS (3 -> 2) (ablation)HGS · Tanks&TemplesPSNR vs LPIPS: PSNR 26.36 -> 26.34; LPIPS 0.126 -> 0.123Mixed
increase k in PGHGS (3 -> 4) (ablation)HGS · Tanks&TemplesPSNR, LPIPS: 26.36 -> 26.26; 0.126 -> 0.131Hurt

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