OASIS-Map keeps an object-level map consistent across repeated visits to a semi-static place, where objects appear, vanish, move or get replaced while the robot is away. It matches dense patch-level semantic features between old and new observations, which both shows where the scene changed and associates objects across revisits under partial views and imperfect segmentation. It reaches 0.783 F1 on change detection for car replacements in a car park and 0.667 F1 on moved-object association in 3RScan, and also runs on an outdoor market.
Relevance
Object association across visits is the weak link in most change detection, and swapped cars of the same model are a hard test of it. Read for long-term autonomy and dynamic scenes.
Links
- arXiv · project page
- Subfield: Semantic mapping, Perception