Research map / Detection and segmentation / 3D detection and driving
Camera-radar fusion detection
Fuse radar with camera or LiDAR for robust 3D vehicle detection.
36 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in 3D detection and driving, keyword cluster “bev · autonomous · radar”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 4 tracked benchmarks.
10 papers in 2023–24 and 18 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 6.3% to 11.8%.
Open questions the data points to
- nuScenes val (NDS): best went from 57.3 (up to 2024) to 63 in 2025 — RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion (CVPR 2025 · 19 citations)
- Papers report CRN on nuScenes val, Rainy (mAP) anywhere from 47.3 to 57: comparisons against it depend on whose number you take.
- Papers report Baseline on nuScenes val (mAP) anywhere from 37.8 to 43.2: comparisons against it depend on whose number you take.
What looks settled
- nuScenes test (NDS): best up to 2024 71.7, best in 2025–26 70.2 from 2 papers, at most 0.2% better, if at all.
- nuScenes test (mAP): best up to 2024 68.9, best in 2025–26 64.9 from 2 papers, at most 0.2% better, if at all.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2020 · Seeing Through Fog Without Seeing Fog: Deep Multimodal Sensor Fusion in Unseen Adverse Weather (CVPR 2020 · 610 citations)
- 2021 · Robust Multimodal Vehicle Detection in Foggy Weather Using Complementary Lidar and Radar Signals (CVPR 2021 · 199 citations)
- 2022 · TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection With Transformers (CVPR 2022 · 939 citations)
- 2023 · CRAFT: Camera-Radar 3D Object Detection with Spatio-Contextual Fusion Transformer (AAAI 2023 · 122 citations)
- 2024 · RCBEVDet: Radar-camera Fusion in Bird's Eye View for 3D Object Detection (CVPR 2024 · 133 citations)
- 2025 · L4DR: LiDAR-4DRadar Fusion for Weather-Robust 3D Object Detection (AAAI 2025 · 27 citations)
- 2026 · Indoor Multi-View Radar Object Detection via 3D Bounding Box Diffusion (AAAI 2026 · 1 citation)
Earliest papers here
- Seeing Through Fog Without Seeing Fog: Deep Multimodal Sensor Fusion in Unseen Adverse Weather (CVPR 2020 · 610 citations)
- Robust Multimodal Vehicle Detection in Foggy Weather Using Complementary Lidar and Radar Signals (CVPR 2021 · 199 citations)
- TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection With Transformers (CVPR 2022 · 939 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
nuScenes test · NDS (higher is better)
8 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 71.7 | TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection With Transformers (CVPR 2022 · 939 citations) |
nuScenes test · mAP (higher is better)
8 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 68.9 | TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection With Transformers (CVPR 2022 · 939 citations) |
nuScenes val · mAP (higher is better)
5 papers, 2023–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 62 | Bi-LRFusion: Bi-Directional LiDAR-Radar Fusion for 3D Dynamic Object Detection (CVPR 2023 · 48 citations) |
nuScenes val · NDS (higher is better)
5 papers, 2024–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 57.3 | CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge Distillation (CVPR 2024 · 35 citations) |
| 2025 | 63 | RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion (CVPR 2025 · 19 citations) |
What has been tried
28 papers here report 340 measured changes (a component added, removed or swapped, compared with a baseline): helped 81%, no clear effect 3%, hurt 14%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add temporal features | nuScenes val · NDS | 47.5 -> 51.9 (+4.4) | Helped | RCBEVDet: Radar-camera Fusion in Bird's Eye View for 3D Object Detection (CVPR 2024 · 133 citations) |
| Add 3D-LRF | K-Radar, all weather · AP3D at IoU=0.5 | 35.9 (-1.9) -> 45.2 (+7.4) | Helped | Towards Robust 3D Object Detection with LiDAR and 4D Radar Fusion in Various Weather Conditions (CVPR 2024 · 44 citations) |
| Add CMA to PillarNet-18 | nuScenes validation · NDS | 27.3 -> 29.3 | Helped | RadarDistill: Boosting Radar-based Object Detection Performance via Knowledge Distillation from LiDAR Features (CVPR 2024 · 43 citations) |
| add a gated network for adaptive sensor fusion | nuScenes val · mAP | 43.2 -> 44.9 | Helped | CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge Distillation (CVPR 2024 · 35 citations) |
| Add MME to the LiDAR-only baseline | VoD-Fog, moderate difficulty, no fog · 3D mAP | 70.3 -> 77.1 | Helped | L4DR: LiDAR-4DRadar Fusion for Weather-Robust 3D Object Detection (AAAI 2025 · 27 citations) |
| Add RHGM to radar-camera fusion | VoD validation · EAA AP | +2.41% | Helped | HGSFusion: Radar-Camera Fusion with Hybrid Generation and Synchronization for 3D Object Detection (AAAI 2025 · 24 citations) |
| replace Vanilla-Unet denoising with diffusion denoising | V2X-R fog · 3D mAP@IoU=0.7 | 62.07 -> 68.08 | Helped | V2X-R: Cooperative LiDAR-4D Radar Fusion with Denoising Diffusion for 3D Object Detection (CVPR 2025 · 24 citations) |
| Add image-view feature sampling to query decoding | nuScenes val · mAP | 40.5 -> 43.6 | Helped | RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion (CVPR 2025 · 19 citations) |
| Use radar and camera modalities instead of a single modality | nuScenes val · NDS | ↑2.4 | Helped | RCTrans: Radar-Camera Transformer via Radar Densifier and Sequential Decoder for 3D Object Detection (AAAI 2025 · 15 citations) |
| Replace ground-truth supervision with SSOD | VoD, entire annotated area · mAP | 41.70 -> 43.70 | Helped | SCKD: Semi-Supervised Cross-Modality Knowledge Distillation for 4D Radar Object Detection (AAAI 2025 · 15 citations) |
| Replace a single decoder with multiple expert decoders, without query routing | nuScenes validation clean · mAP | 70.3 -> 33.4 | Hurt | Resilient Sensor Fusion Under Adverse Sensor Failures via Multi-Modal Expert Fusion (CVPR 2025 · 12 citations) |
| replace the learned RIC with a uniform distribution within object boundaries | nuScenes validation · range MAE (m) | 0.47 -> 0.67 | Hurt | RICCARDO: Radar Hit Prediction and Convolution for Camera-Radar 3D Object Detection (CVPR 2025 · 7 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| CRN | nuScenes val, Rainy | mAP | 47.3 – 57 | 2 papers, e.g. ICLR 2025 and NeurIPS 2024 |
| Baseline | nuScenes val | mAP | 37.8 – 43.2 | 2 papers, e.g. NeurIPS 2024 and CVPR 2024 |
| Baseline | nuScenes val | NDS | 47.4 – 54.1 | 2 papers, e.g. NeurIPS 2024 and CVPR 2024 |
What reviewers will expect
Datasets and baselines named in the 10 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- nuScenes · 5 of 10 papers
- VoD · 3 of 10 papers
- K-Radar · 3 of 10 papers
Baselines
- RCBEVDet · 6 of 10 papers
- BEVFusion · 4 of 10 papers
- RCFusion · 4 of 10 papers
- CRN · 4 of 10 papers
- PointPillars · 3 of 10 papers
- RadarPillarNet · 3 of 10 papers
- FUTR3D · 3 of 10 papers
- LXL · 3 of 10 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 28 papers; from OpenAlex and the papers' affiliations).
- Korea Advanced Institute of Science and Technology · 4 papers
- Hanyang University · 4 papers
- University of Science and Technology of China · 2 papers
- Seoul National University · 2 papers
Use it
Download BibTeX of 30 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- LiDAR perception for driving (75 papers)
- Multi-camera BEV detection (69 papers)
Elsewhere in the topic
- Unified LiDAR 3D detection (107 papers)
- Depth-based and label-free 3D detection (58 papers)
- 3D objectness and scene segmentation (57 papers)
- Monocular 3D object detection (56 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).