ClearGS: Reliability-Aware Gaussian Splatting from Handheld Videos
ClearGS improves 3D Gaussian Splatting from handheld video using reliability weights and in-video restoration.
ClearGS adapts 3D Gaussian Splatting to handheld video with uneven viewpoints and mixed frame quality. Reliability-aware View Allocation assigns graded supervision weights from appearance reliability, degradation risk, and geometric utility, while Render-Guided In-Video Restoration uses the current render and a frozen no-reference restorer to recover lost detail. No-reference perceptual scores select among the render, restored observation, and a high-frequency fusion. On GS2E and GSOTM, the method reports state-of-the-art results with CLIP-IQA and MUSIQ gains and LPIPS reductions in most degradation settings, without paired sharp supervision.
- Graded reliability weights replace binary frame selection.
- Render-guided restoration repairs blur without a clean reference.
- Perceptual scores choose among render, restored, and fused frames.
- CLIP-IQA and MUSIQ rise and LPIPS falls on GS2E and GSOTM.
- AI models
- ClearGS
From the original article
1 min read at arxiv.orgWe present ClearGS for 3D Gaussian Splatting (3DGS) from handheld videos with uneven viewpoint coverage and mixed frame quality.
By Xuanzhi Liu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.