Get Started
Home
Topics
Search
Library
Research questionHow can we correct semantic labels in completed aerial 3D voxel maps while preserving geometry and occupancy?Errors from learned perception can persist in completed aerial 3D semantic maps, reducing their reliability for downstream autonomous tasks. Correcting labels directly in the finished map is difficult because semantic changes must remain consistent with local geometry and neighboring labels.
AI
Computer Vision
Machine Learning
Research Paper
Robotics
Technology
Latest papersRecent research connected to this question, newest first.VoxelFix: Post-Hoc Semantic Correction of Completed 3D Voxel MapsThe evidence concerns completed aerial voxel maps and a graph-based correction model using local geometry and neighboring semantic information. Training pairs are created by corrupting contiguous regions of annotated OccuFly maps according to upstream class confusions; evaluation uses maps from four independently trained 2D segmentation models and an independently reconstructed out-of-distribution aerial scene.research paper · Sep 4, 2026
Related questions
How can 3D occupancy models learn from noisy 2D pseudo-labels without 3D annotations?How can one 3D semantic occupancy model handle indoor and outdoor scenes with varying cameras, scales, and label taxonomies?How can multi-agent semantic segmentation reuse past guideline corrections instead of rediscovering the same errors?How can robots maintain geometrically grounded semantic maps as objects and spatial relations change in real time?