RePL: Pseudo-label Refinement for Semi-supervised LiDAR Semantic Segmentation

2026-04-08

AI Summary

This work addresses the challenges of error propagation and confirmation bias caused by noisy pseudo-labels in semi-supervised LiDAR semantic segmentation. To mitigate these issues, we propose RePL, a pseudo-label refinement framework that leverages a mask reconstruction mechanism to identify and correct errors in pseudo-labels, integrated within a purpose-designed semi-supervised training strategy to enhance label quality. We provide theoretical guarantees for the effectiveness of RePL and demonstrate that its required conditions are readily satisfied in practice. Experimental results on the nuScenes-lidarseg and SemanticKITTI datasets show that RePL significantly improves pseudo-label quality and achieves state-of-the-art performance.

Abstract

Semi-supervised learning for LiDAR semantic segmentation often suffers from error propagation and confirmation bias caused by noisy pseudo-labels. To tackle this chronic issue, we introduce RePL, a novel framework that enhances pseudo-label quality by identifying and correcting potential errors in pseudo-labels through masked reconstruction, along with a dedicated training strategy. We also provide a theoretical analysis demonstrating the condition under which the pseudo-label refinement is beneficial, and empirically confirm that the condition is mild and clearly met by RePL. Extensive evaluations on the nuScenes-lidarseg and SemanticKITTI datasets show that RePL improves pseudo-label quality a lot and, as a result, achieves the state of the art in LiDAR semantic segmentation.

Problem

• semi-supervised learning
• LiDAR semantic segmentation
• pseudo-labels
• error propagation
• confirmation bias

Innovation

• pseudo-label refinement
• masked reconstruction
• semi-supervised learning
• LiDAR semantic segmentation
• error correction