Vis4GS: A Visual Analytic Tool for 3D Gaussian Splatting Reconstruction
Authors
Kai-Yuan Lin (National Tsing Hua University), Aryabima Mandala Putra (National Tsing Hua University), Jui-Chi Lee (National Tsing Hua University), Shih-Hsuan Hung (National Tsing Hua University)
Presentation
- Session
- Big Data, Bigger Physics
- Time
- Friday, Nov 13, 08:00 – 08:09 (US/Eastern) · session 08:00 – 09:30
- Location
- Hall America south
Keywords
3D Gaussian Splatting, Visual analytics, Artifact diagnosis, Multiscale visualization
Abstract
3D Gaussian Splatting (3DGS) supports fast training and real-time rendering, but its optimization process remains difficult to inter- pret. Existing viewers mainly expose the final reconstructed scene and offer limited support for explaining how Gaussian properties contribute to visible artifacts or evolve during training. We present Vis4GS, a multi-view visual analytics tool for primitive-level diag- nosis of 3DGS reconstruction artifacts. Built on the original 3DGS viewer and training framework, Vis4GS links rendered artifacts to Gaussian properties, View Coverage, training progress, and Gaus- sian genealogy through four linked views: an interactive Gaussian analysis view, a property timeline view, a Gaussian densification tree view, and a log and control panel. The system supports Gaus- sian selection, blur and needle-like artifact scoring, View Cover- age analysis, and multiscale genealogy exploration of clone, split, prune, and clone-split events. By connecting scene-level artifacts with primitive-level evidence and optimization history, Vis4GS en- ables a structured workflow for diagnosing reconstruction failures beyond final-image inspection and global metrics. A user study also shows that Vis4GS provides stronger support for usability and artifact understanding than the original 3DGS viewer.
For Practitioners
What type of practitioners would be interested in reading this paper? This paper is primarily for 3D graphics engineers who use 3D Gaussian Splatting (3DGS) pipeline to design, optimize, or deploy neural scene representations. Other practitioners, from fields like computer vision, machine learning, or visual analytics, may also be interested in reading this paper. How could practitioners apply what they learn from this paper to their work? Practitioners can apply the insights from Vis4GS to adopt a structured, primitive-level workflow for diagnosing 3D scene artifacts, instead of focusing more on global metrics such as PSNR or SSIM. By utilizing our artifact scoring system and multi-view genealogy mapping, developers can trace visible rendering errors directly to the issues in optimization process, initializations, and view coverage. Overall, these diagnostic techniques allow practitioners to replace trial-and-error tuning with more informed decisions to improve the 3D scene they are working on.