RCTDistill: Cross-Modal Knowledge Distillation Framework for Radar-Camera 3D Object Detection with Temporal Fusion

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Bang, Geonho, Seong, Minjae, Kim, Jisong, Baek, Geunju, Oh, Daye, Kim, Junhyung, Koh, Junho, Choi, Jun Won
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915506636718080
author Bang, Geonho
Seong, Minjae
Kim, Jisong
Baek, Geunju
Oh, Daye
Kim, Junhyung
Koh, Junho
Choi, Jun Won
author_facet Bang, Geonho
Seong, Minjae
Kim, Jisong
Baek, Geunju
Oh, Daye
Kim, Junhyung
Koh, Junho
Choi, Jun Won
contents Radar-camera fusion methods have emerged as a cost-effective approach for 3D object detection but still lag behind LiDAR-based methods in performance. Recent works have focused on employing temporal fusion and Knowledge Distillation (KD) strategies to overcome these limitations. However, existing approaches have not sufficiently accounted for uncertainties arising from object motion or sensor-specific errors inherent in radar and camera modalities. In this work, we propose RCTDistill, a novel cross-modal KD method based on temporal fusion, comprising three key modules: Range-Azimuth Knowledge Distillation (RAKD), Temporal Knowledge Distillation (TKD), and Region-Decoupled Knowledge Distillation (RDKD). RAKD is designed to consider the inherent errors in the range and azimuth directions, enabling effective knowledge transfer from LiDAR features to refine inaccurate BEV representations. TKD mitigates temporal misalignment caused by dynamic objects by aligning historical radar-camera BEV features with current LiDAR representations. RDKD enhances feature discrimination by distilling relational knowledge from the teacher model, allowing the student to differentiate foreground and background features. RCTDistill achieves state-of-the-art radar-camera fusion performance on both the nuScenes and View-of-Delft (VoD) datasets, with the fastest inference speed of 26.2 FPS.
format Preprint
id arxiv_https___arxiv_org_abs_2509_17712
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RCTDistill: Cross-Modal Knowledge Distillation Framework for Radar-Camera 3D Object Detection with Temporal Fusion
Bang, Geonho
Seong, Minjae
Kim, Jisong
Baek, Geunju
Oh, Daye
Kim, Junhyung
Koh, Junho
Choi, Jun Won
Computer Vision and Pattern Recognition
Radar-camera fusion methods have emerged as a cost-effective approach for 3D object detection but still lag behind LiDAR-based methods in performance. Recent works have focused on employing temporal fusion and Knowledge Distillation (KD) strategies to overcome these limitations. However, existing approaches have not sufficiently accounted for uncertainties arising from object motion or sensor-specific errors inherent in radar and camera modalities. In this work, we propose RCTDistill, a novel cross-modal KD method based on temporal fusion, comprising three key modules: Range-Azimuth Knowledge Distillation (RAKD), Temporal Knowledge Distillation (TKD), and Region-Decoupled Knowledge Distillation (RDKD). RAKD is designed to consider the inherent errors in the range and azimuth directions, enabling effective knowledge transfer from LiDAR features to refine inaccurate BEV representations. TKD mitigates temporal misalignment caused by dynamic objects by aligning historical radar-camera BEV features with current LiDAR representations. RDKD enhances feature discrimination by distilling relational knowledge from the teacher model, allowing the student to differentiate foreground and background features. RCTDistill achieves state-of-the-art radar-camera fusion performance on both the nuScenes and View-of-Delft (VoD) datasets, with the fastest inference speed of 26.2 FPS.
title RCTDistill: Cross-Modal Knowledge Distillation Framework for Radar-Camera 3D Object Detection with Temporal Fusion
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.17712