Spatially Stratified Distillation for Heterogeneous Radar Place Recognition

• URL: http://arxiv.org/abs/2606.18687v1
• Date: Wed, 17 Jun 2026 04:55:42 GMT
• Title: Spatially Stratified Distillation for Heterogeneous Radar Place Recognition
• Abstract summary: A notable application is matching queries from cost-effective 4D automotive radars against high-fidelity reference maps built by dense spinning radars.<n>This process is fundamentally limited by the extreme sparsity (and narrow field-of-view) of the 4D sensor.<n>We propose spatially-stratified distillation (SSD), a strategy that replaces standard uniform distillation with an asymmetric spatial alignment.
• Score: 6.215632013632239
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• 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.

Related papers

• DiffRadar: Differentiable Physics-Aware Radar SLAM with Gaussian Fields [8.442884076695789]
  We present DiffRadar, a real-time radar SLAM system that models radar observations as a differentiable, physics-aware Gaussian field.<n>DiffRadar represents the scene as anisotropic Gaussian primitives and renders radar measurements in range-azimuth and Doppler-azimuth spaces.<n>We implement DiffRadar on commodity FMCW radar hardware and evaluate it on both the public Radarize benchmark and a controlled stress-test suite.
  arXiv  Detail & Related papers  (2026-07-14T02:08:14Z)
• RadarSim: Simulating Single-Chip Radar via Multimodal Neural Fields [49.41438378775056]
  Radars are an ideal complement to cameras, with cameras offering fine angular resolution and radars providing metric depth and robustness under adverse weather.<n>RadarSim produces sharper geometry and Doppler range frames than radar-only reconstructions.
  arXiv  Detail & Related papers  (2026-05-25T21:03:40Z)
• Dr-BA: Separable Optimization for Direct Radar Bundle Adjustment & Localization [12.318856176227383]
  Dr-BA is a first-of-its-kind radar bundle adjustment framework that operates directly on 2D spinning radar intensity images.<n>Dr-BA achieves state-of-the-art radar-based BA and cross-session localization performance, demonstrated on more than 200 km of on-road data.
  arXiv  Detail & Related papers  (2026-05-07T23:41:16Z)
• RLPR: Radar-to-LiDAR Place Recognition via Two-Stage Asymmetric Cross-Modal Alignment for Autonomous Driving [14.25906590501402]
  Radar-to-LiDAR place recognition can localize radar scans within existing LiDAR maps.<n>We propose RLPR, a robust radar-to-LiDAR place recognition framework compatible with single-chip, scanning, and 4D radars.<n>Experiments on four datasets demonstrate that RLPR achieves recognition accuracy with strong zero-shot generalization capabilities.
  arXiv  Detail & Related papers  (2026-03-09T03:28:44Z)
• RadarGen: Automotive Radar Point Cloud Generation from Cameras [64.69976771710057]
  We present RadarGen, a diffusion model for synthesizing realistic automotive radar point clouds from multi-view camera imagery.<n>RadarGen adapts efficient image-latent diffusion to the radar domain by representing radar measurements in bird's-eye-view form.<n>We show that RadarGen captures characteristic radar measurement distributions and reduces the gap to perception models trained on real data.
  arXiv  Detail & Related papers  (2025-12-19T18:57:33Z)
• 4D-RaDiff: Latent Diffusion for 4D Radar Point Cloud Generation [10.945807584683726]
  We propose a novel framework to generate 4D radar point clouds for training and evaluating object detectors.<n>The proposed 4D-RaDiff converts unlabeled bounding boxes into high-quality radar annotations and transforms existing LiDAR point cloud data into realistic radar scenes.
  arXiv  Detail & Related papers  (2025-12-16T09:43:05Z)
• Enhanced 3D Object Detection via Diverse Feature Representations of 4D Radar Tensor [5.038148262901536]
  Raw 4D Radar (4DRT) offers richer spatial and Doppler information than conventional point clouds.<n>We propose a novel 3D object detection framework that maximizes the utility of 4DRT while preserving efficiency.<n>We show that our framework achieves improvements of 7.3% in AP_3D and 9.5% in AP_BEV over the baseline RTNH model when using extremely sparse inputs.
  arXiv  Detail & Related papers  (2025-02-10T02:48:56Z)
• TransRAD: Retentive Vision Transformer for Enhanced Radar Object Detection [6.163747364795787]
  We present TransRAD, a novel 3D radar object detection model.<n>We propose Location-Aware NMS to mitigate the common issue of duplicate bounding boxes in deep radar object detection.<n>Results demonstrate that TransRAD outperforms state-of-the-art methods in both 2D and 3D radar detection tasks.
  arXiv  Detail & Related papers  (2025-01-29T20:21:41Z)
• Radar Fields: Frequency-Space Neural Scene Representations for FMCW Radar [62.51065633674272]
  We introduce Radar Fields - a neural scene reconstruction method designed for active radar imagers. Our approach unites an explicit, physics-informed sensor model with an implicit neural geometry and reflectance model to directly synthesize raw radar measurements. We validate the effectiveness of the method across diverse outdoor scenarios, including urban scenes with dense vehicles and infrastructure.
  arXiv  Detail & Related papers  (2024-05-07T20:44:48Z)
• Echoes Beyond Points: Unleashing the Power of Raw Radar Data in Multi-modality Fusion [74.84019379368807]
  We propose a novel method named EchoFusion to skip the existing radar signal processing pipeline. Specifically, we first generate the Bird's Eye View (BEV) queries and then take corresponding spectrum features from radar to fuse with other sensors.
  arXiv  Detail & Related papers  (2023-07-31T09:53:50Z)
• Bi-LRFusion: Bi-Directional LiDAR-Radar Fusion for 3D Dynamic Object Detection [78.59426158981108]
  We introduce a bi-directional LiDAR-Radar fusion framework, termed Bi-LRFusion, to tackle the challenges and improve 3D detection for dynamic objects. We conduct extensive experiments on nuScenes and ORR datasets, and show that our Bi-LRFusion achieves state-of-the-art performance for detecting dynamic objects.
  arXiv  Detail & Related papers  (2023-06-02T10:57:41Z)
• Semantic Segmentation of Radar Detections using Convolutions on Point Clouds [59.45414406974091]
  We introduce a deep-learning based method to convolve radar detections into point clouds. We adapt this algorithm to radar-specific properties through distance-dependent clustering and pre-processing of input point clouds. Our network outperforms state-of-the-art approaches that are based on PointNet++ on the task of semantic segmentation of radar point clouds.
  arXiv  Detail & Related papers  (2023-05-22T07:09:35Z)
• RadarNet: Exploiting Radar for Robust Perception of Dynamic Objects [73.80316195652493]
  We tackle the problem of exploiting Radar for perception in the context of self-driving cars. We propose a new solution that exploits both LiDAR and Radar sensors for perception. Our approach, dubbed RadarNet, features a voxel-based early fusion and an attention-based late fusion.
  arXiv  Detail & Related papers  (2020-07-28T17:15:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.