Bayesian Scene Change Detection via 3D Gaussian Splatting
October 20, 2026
This method enables online scene change detection by learning from automatic 3D Gaussian Splatting (3DGS) edits. The approach uses these edits to generate labeled training views, allowing for effective pixel prior and response statistics estimation without manual annotation.
HOW THIS AFFECTS YOU
●
builderThis could improve real-time scene reconstruction and monitoring in robotics or AR applications.
●
researcherThe use of 3DGS edits for unsupervised supervision is a novel way to handle online Bayesian inference.