Research map / Point clouds / Segmentation and detection
Echoes Beyond Points: Unleashing the Power of Raw Radar Data in Multi-modality Fusion
Yang Liu, Feng Wang, Naiyan Wang, Zhaoxiang Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fuse raw radar range-time data with camera via polar-aligned attention instead of FFTRadNet3D RD-only | EchoFusion · RADIal refined 3D boxes | BEV AP@0.7 overall: 57.26 -> 84.92 | Improved |
| add camera image to radar-only range-time input (ablation) | EchoFusion RT · RADIal refined 3D boxes | BEV AP@0.7 overall: 48.26 -> 84.92 | Improved |
| use raw radar (RT) instead of CFAR radar point cloud when fused with camera (ablation) | EchoFusion · RADIal refined 3D boxes | BEV AP@0.7 overall: 82.35 -> 84.92 | Improved |
| use range-azimuth map instead of range-time data (ablation) | EchoFusion C+radar · RADIal refined 3D boxes | BEV AP@0.7 overall: 84.92 -> 84.77 | No clear effect |
| decompose complex radar input into magnitude and phase instead of IQ (ablation) | EchoFusion · RADIal refined 3D boxes | BEV AP@0.7 overall: 82.64 -> 84.92 | Improved |
| use pre-trained image backbone (ablation) | EchoFusion · RADIal refined 3D boxes | BEV AP@0.7 overall: 74.34 -> 84.92 | Improved |
| replace polar with sphere BEV coordinate system (ablation) | EchoFusion · RADIal refined 3D boxes | BEV AP@0.7 overall: 84.92 -> 85.02 | No clear effect |
| use range-azimuth map + camera instead of radar point cloud + camera (ablation) | EchoFusion · KRadar-20 train split | BEV AP@0.7: 26.39 -> 29.65 | Improved |
| use EchoFusion with image and RA map instead of RTNH point cloud | KRadar-20 trainval · KRadar-20 test | BEV AP@0.7: 10.47 -> 33.07 | Improved |
| fuse radar RT with camera for 3D AP versus LiDAR | EchoFusion · RADIal 3D AP@0.5 | 3D AP@0.5 overall: 62.69 -> 39.81 | Hurt |
Closest papers by meaning
- HGSFusion: Radar-Camera Fusion with Hybrid Generation and Synchronization for 3D Object Detection (AAAI 2025 · similarity 0.85)
- RPGFusion: 4D Radar Prior-Guided Multi-Modal Fusion for 3D Detection (CVPR 2026 · similarity 0.85)
- Depth Estimation From Camera Image and mmWave Radar Point Cloud (CVPR 2023 · similarity 0.85)
- TacoDepth: Towards Efficient Radar-Camera Depth Estimation with One-stage Fusion (CVPR 2025 · similarity 0.83)
- Selection, Not Fusion: Radar-Modulated State Space Models for Radar-Camera Depth Estimation (NeurIPS 2026 · similarity 0.83)
- RadarMOSEVE: A Spatial-Temporal Transformer Network for Radar-Only Moving Object Segmentation and Ego-Velocity Estimation (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
