New benchmark and method tackle noisy labels in 3D semantic occupancy prediction

Researchers have introduced OccNL, a new benchmark designed to evaluate 3D semantic occupancy prediction models under noisy label conditions. They found that existing 2D label noise learning strategies perform poorly in sparse 3D voxel spaces. To address this, they developed DPR-Occ, a framework that uses dual-source partial label reasoning to create reliable supervision by leveraging temporal model memory and representation-level structural affinity. Experiments on SemanticKITTI showed that DPR-Occ maintains performance even with up to 90% label noise, significantly outperforming adapted baselines. AI

IMPACT This research could improve the reliability of robotic perception systems operating in real-world environments with imperfect data.

RANK_REASON This is a research paper introducing a new benchmark and a novel method for a specific computer vision task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• DPR-Occ
• Kailun Yang
• OccNL
• SemanticKITTI

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →