Spatially Stratified Distillation for Heterogeneous Radar Place Recognition

2026-06-17

AI Summary

This work addresses the performance degradation in place recognition caused by data sparsity and field-of-view discrepancies between low-cost 4D automotive radar and high-density rotating radar. To tackle this challenge, the authors propose Spatially Stratified Distillation (SSD), a method that assigns non-uniform distillation weights based on the spatial overlap of physical radar echoes. Specifically, SSD enforces strong feature alignment in overlapping regions while applying weak supervision in sparse areas, thereby achieving asymmetric alignment of heterogeneous radar features. Unlike conventional uniform distillation strategies, SSD effectively mitigates performance degradation in multi-session scenarios. Experimental results on dynamic sequences from the HeRCULES dataset demonstrate that the proposed approach significantly outperforms existing methods, establishing a new state of the art in heterogeneous radar-based place recognition.

Abstract

Scalable, all-weather place recognition increasingly relies on heterogeneous radar place recognition to bridge diverse hardware platforms. A notable application is matching queries from cost-effective 4D automotive radars against high-fidelity reference maps built by dense spinning radars. This process is fundamentally limited by the extreme sparsity (and narrow field-of-view) of the 4D sensor, which captures only a fraction of the structural density present in the spinning radar database. Prior efforts address this issue by unifying different radar signals. That is, projecting both signals into a common representational space. Yet, they suffer performance degradation in multi-session environments. In this paper, we propose spatially-stratified distillation (SSD); a strategy that replaces standard uniform distillation with an asymmetric spatial alignment derived directly from physical radar returns. In regions where both radars exhibit overlapping returns, SSD enforces strong feature alignment. Crucially, in sparse regions where the 4D student lacks returns but the teacher contains valid structure within the shared field of view, SSD applies heavily discounted distillation weights. Extensive evaluations of the recent HeRCULES dataset demonstrate that SSD significantly outperforms prior place recognition methods, achieving state-of-the-art results on its challenging dynamic sequences.

Problem

• heterogeneous radar
• place recognition
• 4D automotive radar
• sparsity
• multi-session

Innovation

• spatially-stratified distillation
• heterogeneous radar place recognition
• feature alignment
• radar distillation
• 4D automotive radar