SSA-3DGS: Unsupervised Removal of Screen-Space Artifacts for 3D Gaussian Splatting

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Expert, extended

Summary

SSA-3DGS is an unsupervised framework designed to remove screen-space artifacts from 3D Gaussian Splatting (3DGS) reconstructions. Traditional Novel View Synthesis (NVS) methods struggle with issues like sensor defects, lens obstructions, or digital watermarks, which are erroneously baked into 3D geometry as "floaters." SSA-3DGS addresses this by jointly optimizing a clean 3D scene and a learnable 2D overlay representing the corrupting artifacts. It leverages motion parallax to disentangle static artifacts from the 3D scene without supervision, using sparsity and total-variation regularization. The method improves reconstruction fidelity by up to 9 dB PSNR over 3DGS trained on corrupted inputs, demonstrated across diverse synthetic corruptions and a self-captured real-world dataset, while faithfully preserving the artifact.

Key takeaway

For Machine Learning Engineers or 3D content creators working with real-world multi-view captures, screen-space artifacts like watermarks or lens obstructions can severely degrade 3D Gaussian Splatting quality. You should consider SSA-3DGS to automatically disentangle these static 2D artifacts from your 3D scene geometry. This unsupervised framework significantly improves reconstruction fidelity, reducing the need for laborious manual data cleaning and ensuring higher quality 3D assets from imperfect input data.

Key insights

Jointly optimizing a 3D scene and a 2D overlay, leveraging motion parallax, disentangles screen-space artifacts from 3D Gaussian Splatting.

Principles

• Screen-space artifacts are static relative to the camera.
• Motion parallax enables unsupervised artifact separation.
• Regularization prevents overlay from absorbing scene details.

Method

SSA-3DGS jointly optimizes 3DGS parameters and a learnable 2D overlay, using sparsity and total-variation regularization to separate artifacts from the 3D scene.

In practice

• Clean 3DGS inputs corrupted by watermarks or UI elements.
• Restore scenes from cameras with mud or rain on the lens.
• Improve 3D reconstruction from dashcam footage.

Topics

• 3D Gaussian Splatting
• Novel View Synthesis
• Screen-Space Artifacts
• Unsupervised Learning
• Image Restoration
• Multi-View Stereo

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Structure-Aware Gaussian Splatting for Large-Scale Scene Reconstruction — SIG improves large-scale 3D Gaussian Splatting by synchronizing image supervision with Gaussian frequencies, preventing uncontrolled densification.
• NG-GS: NeRF-Guided 3D Gaussian Splatting Segmentation — NG-GS enhances 3D Gaussian Splatting segmentation by integrating NeRF-guided continuous feature fields to resolve boundary artifacts.
• PixGS: Pixel-Space Diffusion for Direct 3D Gaussian Splat Generation — PixGS directly generates high-quality 3D Gaussian Splats using pixel-space diffusion, bypassing latent compression artifacts.
• OrbitForge: Text-to-3D Scene Generation via Reconstruction-Anchored Video Synthesis — OrbitForge converts text-generated videos into consistent 3D Gaussian Splatting scenes by anchoring reconstruction.
• WildSplatter: Feed-forward 3D Gaussian Splatting with Appearance Control from Unconstrained Images — WildSplatter enables fast, pose-free 3DGS from unconstrained images by disentangling geometry and appearance via learned embeddings.
• Three-dimensional Damage Visualization of Civil Structures via Gaussian Splatting-enabled Digital Twins — Gaussian Splatting enhances 3D damage visualization in civil infrastructure digital twins, improving accuracy and efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.