Learned 3DGS edits power online Bayesian scene change detection
Original title: Learning from Automatic 3D Gaussian Splatting Edits for Online Bayesian Scene Change Detection
Authors report 67.64% mIoU on PASLCD and 75.82% on CCS3D, with the full PASLCD pipeline at 17.2 FPS.
The method learns scene-specific evidence models offline from automatic edits to a reference 3D Gaussian Splatting reconstruction, so labelled views do not need manual annotation. Online prediction fuses new image evidence with historical support, and closed-form local Bayes factors update an accumulated 3D Gaussian field. Object-guided refinement produces the change masks. On PASLCD the pipeline is described as the fastest among the online methods evaluated.
Unconfirmed: Accuracy and speed figures are author-reported on PASLCD and CCS3D.
- Training views come from automatic 3DGS edits plus unchanged reference images.
- Pixel prior and response statistics are estimated per scene.
- Online updates use closed-form local Bayes factors.
- Final change masks follow object-guided refinement.
From the original article
1 min read at huggingface.co
Online scene change detection requires integrating incoming observations to determine which regions have changed relative to a reference scene. We update change beliefs as new images arrive using generalised Bayesian evidence fusion. However, the prior probability of change and evidence models are unknown beforehand, and numerical Bayesian inference can be costly.
By Changrui Zhu, Simon Julier · huggingface.co
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.