CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge Distillation

Jingyu Song; Katherine A. Skinner; Lingjun Zhao

RECORD

CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge Distillation

Lingjun Zhao, Jingyu Song, Katherine A. Skinner

Computer Vision and Pattern Recognition · Submitted 28 Mar 2024

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim We present extensive evaluations on the nuScenes dataset to demonstrate the effectiveness of the proposed CRKD framework.

how it is stated · asserted 0.99 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

arXiv ID2403.19104 (version 1)

Pith IDM6CIZZ5S, Pith's permanent record of this paper and its review

CitationJingyu Song, Katherine A. Skinner, Lingjun Zhao. "CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge Distillation." arXiv:2403.19104 (2024). https://pith.science/paper/M6CIZZ5S

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 6,171 input tokens.

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.11 · human participants 0.02 · controlled comparison 0.70 · randomized assignment 0.01 · states quantitative result 0.04 · compares with prior results 0.76 · mentions code or data availability 0.51.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; show every role.

In the field of 3D object detection for autonomous driving, LiDAR-Camera (LC) fusion is the top-performing sensor configuration. Still, LiDAR is relatively high cost, which hinders adoption of this technology for consumer automobiles. Alternatively, camera and radar are commonly deployed on vehicles already on the road today, but performance of Camera-Radar (CR) fusion falls behind LC fusion. In this work, we propose Camera-Radar Knowledge Distillation (CRKD) to bridge the performance gap between LC and CR detectors with a novel cross-modality KD framework. We use the Bird's-Eye-View (BEV) representation as the shared feature space to enable effective knowledge distillation. To accommodate the unique cross-modality KD path, we propose four distillation losses to help the student learn crucial features from the teacher model. We present extensive evaluations on the nuScenes dataset to demonstrate the effectiveness of the proposed CRKD framework. The project page for CRKD is https://song-jingyu.github.io/CRKD.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	In the field of 3D object detection for autonomous driving, LiDAR-Camera (LC) fusion is the top-performing sensor configuration.	background 1.00	0.07	0.03	asserted 1.00
s1	Still, LiDAR is relatively high cost, which hinders adoption of this technology for consumer automobiles.	background 0.95	0.06	0.02	asserted 1.00
s2	Alternatively, camera and radar are commonly deployed on vehicles already on the road today, but performance of Camera-Radar (CR) fusion falls behind LC fusion.	background 0.80	0.07	0.03	asserted 1.00
s3	In this work, we propose Camera-Radar Knowledge Distillation (CRKD) to bridge the performance gap between LC and CR detectors with a novel cross-modality KD framework.	contribution 1.00	0.90	0.24	asserted 0.99
s4	We use the Bird's-Eye-View (BEV) representation as the shared feature space to enable effective knowledge distillation.	method 0.98	0.88	0.04	asserted 0.99
s5	To accommodate the unique cross-modality KD path, we propose four distillation losses to help the student learn crucial features from the teacher model.	method 0.81	0.92	0.04	asserted 0.97
s6	We present extensive evaluations on the nuScenes dataset to demonstrate the effectiveness of the proposed CRKD framework.	contribution 0.54	0.95	0.29	asserted 0.99
s7	The project page for CRKD is https://song-jingyu.github.io/CRKD.	contribution 0.57	0.18	0.01	asserted 0.63

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.89.

What was examined, and what was not

sources read

arXiv listing abstract v1

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

In the field of 3D object detection for autonomous driving, LiDAR-Camera (LC) fusion is the top-performing sensor configuration. Still, LiDAR is relatively high cost, which hinders adoption of this technology for consumer automobiles. Alternatively, camera and radar are commonly deployed on vehicles already on the road today, but performance of Camera-Radar (CR) fusion falls behind LC fusion. In this work, we propose Camera-Radar Knowledge Distillation (CRKD) to bridge the performance gap between LC and CR detectors with a novel cross-modality KD framework. We use the Bird's-Eye-View (BEV) representation as the shared feature space to enable effective knowledge distillation. To accommodate the unique cross-modality KD path, we propose four distillation losses to help the student learn crucial features from the teacher model. We present extensive evaluations on the nuScenes dataset to demonstrate the effectiveness of the proposed CRKD framework. The project page for CRKD is https://song-jingyu.github.io/CRKD.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

1 arXiv paper cite this in 1 sentence

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it

Cites it as background 1 paper

CRKD [ref] employs a method to transfer the knowledge possessed by the LiDAR-Camera fusion detector to the Camera-Radar fusion detector using the Cross-modality Knowledge Distillation technique.

CRT-Fusion: Camera, Radar, Temporal Fusion Using Motion Information for 3D Object Detection · Camera-Radar 3D Object Detection · 0.89

What it cites · 11 sentences without a confident label

Uses the method or tool of 1 paper

We add a gated network [ref] to BEVFusion [ref] to enable the model to learn to generate attention weights on the single-modality feature maps to fuse the complementary modalities adaptively.

LiRaFusion: Deep Adaptive LiDAR-Radar Fusion for 3D Object Detection · Model Architecture Refinement · 0.92

Cites as support 2 papers

LiDAR-Camera (LC) fusion has been widely demonstrated as the top-performing sensor configuration for 3D object detection [ref] .

EA-LSS: Edge-aware Lift-splat-shot Framework for 3D BEV Object Detection · Introduction · 0.96
Meanwhile, more works have demonstrated that the boundary region of the foreground can also contribute to effective KD [ref] .

SimDistill: Simulated Multi-modal Distillation for BEV 3D Object Detection · Mask-Scaling Feature Distillation (MSFD) · 0.97

Cites as background 5 papers

Typical paths include LiDAR-to-Camera (L2C) [ref] and Camera-to-LiDAR (C2L) [ref] .

TiG-BEV: Multi-view BEV 3D Object Detection via Target Inner-Geometry Learning · Cross-modality Knowledge Distillation · 0.95
Specifically, in the field of 3D object detection, a group of KD methods requires that the teacher and student models use the same modality such as LO-to-LO (L2L) [ref] and CO-to-CO (C2C) [ref] .

Structured Knowledge Distillation Towards Efficient and Compact Multi-View 3D Detection · Cross-modality Knowledge Distillation · 0.84
Perception is an important module for achieving safe and effective autonomous driving [ref] . 3D object detection is an essential task in perception as it is of great significance for subsequent tasks [ref] .

LiRaFusion: Deep Adaptive LiDAR-Radar Fusion for 3D Object Detection · Introduction · 0.82

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.