No Adaptation Without Observation: Observability-Constrained Test-Time Prompt Tuning for LiDAR Semantic Segmentation

2026-06-29

AI Summary

This work addresses the performance degradation of LiDAR semantic segmentation under shifting perceptual conditions during deployment and the scarcity of newly annotated data. To tackle this challenge, the authors propose a geometry-constrained test-time prompt tuning framework that estimates spatial observability via depth-consistent ray termination and neighborhood support, enabling reliability-weighted supervision signals. The approach freezes the backbone network and updates only a lightweight prompt adapter, while incorporating a spatiotemporally smoothed prototype alignment strategy to stabilize online adaptation. Requiring no additional annotations, the method significantly improves cross-domain segmentation performance and adaptation stability on standard LiDAR benchmarks.

Abstract

LiDAR semantic segmentation often degrades under real-world deployment due to evolving sensing conditions, while collecting new annotations for retraining is impractical. Test-time adaptation (TTA) updates model parameters online using pseudo-label supervision, but directly applying standard TTA strategies to LiDAR data is challenging. Because pseudo-label reliability is spatially heteroscedastic under range-dependent sparsity and occlusion, uniform updates on globally shared parameters can inject unstable gradients and destabilize adaptation. We propose a geometry-constrained test-time prompt tuning framework for LiDAR semantic segmentation. Our method estimates per-location sensing reliability from depth-consistent beam terminations and neighborhood support, and uses it to reweight spatial supervision. Adaptation is confined to lightweight prompt adapters inserted into a frozen backbone, with spatial gating to prevent unreliable regions from perturbing globally shared representations. A temporally smoothed prototype alignment strategy further stabilizes online updates by accumulating reliable semantic evidence over time. Experiments on standard LiDAR benchmarks demonstrate improved adaptation stability and segmentation performance under deployment variations without additional annotations.

Problem

• LiDAR semantic segmentation
• test-time adaptation
• pseudo-label reliability
• spatial heteroscedasticity
• deployment shift

Innovation

• test-time adaptation
• prompt tuning
• observability
• LiDAR semantic segmentation
• spatial reliability