Get Started
Home
Topics
Search
Library
Research questionHow can 3D occupancy models learn from noisy 2D pseudo-labels without 3D annotations?2D pseudo-labels may contain both depth errors and semantic mistakes. Projecting these imperfect targets into 3D can propagate inaccuracies through the predicted occupancy field.
AI
Computer Vision
Image & Video Processing
Machine Learning
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.LetOccVote: Learning Weakly Supervised 3D Occupancy through ConsensusThe evidence concerns a Gaussian-based 3D occupancy framework trained solely with 2D pseudo-labels and cross-frame observations. Results are reported on Occ3D-nuScenes, including 53.27 IoU and 20.39 mIoU, against methods using 2D pseudo-label supervision; no 3D occupancy annotations are used.research paper · Sep 7, 2026
Related questions
How can one 3D semantic occupancy model handle indoor and outdoor scenes with varying cameras, scales, and label taxonomies?How can we predict local 3D occupancy from a single underwater RGB image despite unreliable visual geometry?How can we complete unseen 3D scene regions from sparse, unconstrained views without dense 3D supervision?How can we correct semantic labels in completed aerial 3D voxel maps while preserving geometry and occupancy?