3dgs-sc: a controlled static screen-content benchmark for 3d gaussian splatting

Shicheng Cai; Hao Zhang; Dong Dai; Xuerui Ma; Ying Hu; Tao Zhang

cs.CVSep 23, 2026

3dgs-sc: a controlled static screen-content benchmark for 3d gaussian splatting

Authors: Shicheng Cai, Hao Zhang, Dong Dai, Xuerui Ma, Ying Hu, Tao Zhang

Organizations: Central South University, Changsha, China · Hunan Malanshan Audio and Video Laboratory, China

Abstract

Does high image fidelity imply readable screen content in 3D Gaussian Splatting (3DGS)? We introduce 3DGS-SC, a controlled static screen-content dataset and benchmark for examining this mismatch. Ten procedural scenes provide fixed multi-view splits, exact cameras, screen masks, text boxes, and transcripts. The protocol separates whole-image fidelity, screen-region fidelity, OCR readability, and edge preservation. In the reported five-method comparison, LightGaussian exceeds Mip-Splatting in screen PSNR by only 0.08 dB, yet trails it in OCR accuracy by 12.7 percentage points. Across all ten method pairs, screen-PSNR and OCR orderings disagree in seven cases. These aggregate results expose a method-selection failure of fidelity-only evaluation. Complementing prior text-aware 3DGS research, 3DGS-SC targets controlled monitor interfaces and exact annotations; scene-wise robustness and acquisition effects remain open validation questions.

Figures & tables

Figure 1: Screen fidelity versus readability on 3DGS-SC. Each point is one method’s reported aggregate in Table 2 . LightGaussian and Mip-Splatting differ by only 0.08 dB in screen PSNR but by 12.7 percentage points in OCR accuracy. The dashed segment connects this comparison, not a fitted trend. Axes show the observed score range; per-scene uncertainty is unavailable.

Figure 2: Representative scenes and annotations in 3DGS-SC. The top two rows show diverse screen-content scenes, including reading, code editing, spreadsheet, dashboard, meeting, and chat layouts. The bottom row visualizes benchmark annotations, including screen masks, text bounding boxes, aligned transcripts, and a monitor-setup example, demonstrating the screen-aware annotation package released with the benchmark.

Dataset	Domain	Acquisition	Evaluation emphasis
NeRF-Synthetic [ 13 ]	Objects	Synthetic	Multi-view image fidelity
Mip-NeRF 360 [ 2 ]	Natural scenes	Captured	Multi-view image fidelity
SCID [ 16 ]	Screen images	2D image database	Screen image quality
STRinGS-360 [ 17 ]	Text-rich objects	Captured	Text-aware 3DGS / OCR-CER
3DGS-SC	Static monitor interfaces	Procedural	Screen fidelity, OCR, edges

Table 1: Scope of representative evaluation datasets. STRinGS-360 already studies text-rich 3DGS; the proposed distinction is controlled screen content and exact procedural annotations, not the first use of OCR for 3DGS.

Method	Global Fidelity		SC-aware Quality			Structure
Method	PSNR $\uparrow$	SSIM $\uparrow$	PSNR ${}_{\text{screen}}\uparrow$	OCR Acc. $\uparrow$	CER $\downarrow$	Edge-F1 $\uparrow$
3DGS	11.67	0.79	19.88	78.9	12.8	0.781
Mip-Splatting	11.18	0.84	20.63	86.2	7.4	0.842
AbsGS	13.41	0.74	17.54	83.6	8.7	0.834
Scaffold-GS	28.96	0.84	17.41	84.8	8.1	0.828
LightGaussian	11.94	0.92	20.71	73.5	17.2	0.744

Table 2: Overall benchmark results on 3DGS-SC. Best values are bold; second-best values are underlined (including ties). OCR accuracy and CER are expressed as percentages; PSNR is in dB. Values are inherited from the manuscript aggregates. The table shows that global fidelity and SC-oriented quality are not equivalent.

3D Gaussian Splatting

3D Gaussian Splatting represents scenes as optimizable 3D Gaussian primitives for fast, photorealistic rendering from new viewpoints.

See all 1,027 papers

Drop-In Perceptual Optimization for 3D Gaussian Splatting
Mar 23, 2026Ezgi Ozyilkan, Zhiqi Chen, Oren Rippel +23D Gaussian Splatting Perceptual Loss
Apple · New York University Tandon School of Engineering
SPLATIFY: Reproduce, Discover, Innovate! From Papers and Ideas to Trainable 3DGS Code
Oct 6, 2026Seemandhar Jain, Keshav Gupta, Manmohan ChandrakerCode Generation 3D Gaussian Splatting
University of California, San Diego
When 3D Gaussian Splatting Recovers Real Surfaces
Aug 30, 2026Songhe Wang, David Johnathan Miller3D Gaussian Splatting 3D Reconstruction
School of EECS, Penn State, University Park, PA, USA

Mar 23, 2026J/K move · Enter open · S save

Drop-In Perceptual Optimization for 3D Gaussian Splatting

Ezgi Ozyilkan, Zhiqi Chen, Oren Rippel, Jona Ballé+1

Apple · New York University Tandon School of Engineering

3D Gaussian Splatting Perceptual Loss Wasserstein Distance 3DGS Training

Open paper AskPDF arXiv

Despite their output being ultimately consumed by human viewers, 3D Gaussian Splatting (3DGS) methods often rely on ad-hoc combinations of pixel-level losses, resulting in blurry renderings. To address this, we systematically explore perceptual optimization strategies for 3DGS by searching over a diverse set of distortion losses. We conduct the first-of-its-kind large-scale human subjective study on 3DGS, involving 39,320 pairwise ratings across several datasets and 3DGS frameworks. A regularized version of Wasserstein Distortion, which we call WD-R, emerges as the clear winner, excelling at recovering fine textures without incurring a higher splat count. WD-R is preferred by raters more than

2.3\times

over the original 3DGS loss, and

1.5\times

over the current best method Perceptual-GS. WD-R also consistently achieves state-of-the-art LPIPS, DISTS, and FID scores across various datasets, and generalizes across recent frameworks, such as Mip-Splatting and Scaffold-GS, where replacing the original loss with WD-R consistently enhances perceptual quality within a similar resource budget (number of splats for Mip-Splatting, model size for Scaffold-GS), and leads to reconstructions being preferred by human raters

1.8\times

and

3.6\times

, respectively. We also find that this carries over to the task of 3DGS scene compression, with

\approx 50\%

bitrate savings for comparable perceptual metric performance.

Abstract

3D Gaussian Splatting

Explore similar work

Drop-In Perceptual Optimization for 3D Gaussian Splatting

SPLATIFY: Reproduce, Discover, Innovate! From Papers and Ideas to Trainable 3DGS Code

When 3D Gaussian Splatting Recovers Real Surfaces