Research map / Detection and segmentation / 3D detection and driving
Ray Denoising: Depth-aware Hard Negative Sampling for Multi-view 3D Object Detection
Feng Liu, Tengteng Huang, Qianjing Zhang, Haotian Yao, Chi Zhang, Fang Wan, Qixiang Ye, Yanzhao Zhou
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 |
|---|---|---|---|
| Add Ray Denoising (depth-aware hard negative queries along camera rays) | StreamPETR · nuScenes val (ResNet50) | mAP: 45.0 -> 46.9 | Improved |
| Add Ray Denoising | StreamPETR on Argoverse 2 · Argoverse 2 val | mAP: 20.3 -> 22.3 | Improved |
| Use uniform distribution (Beta 1,1) for ray query sampling (ablation) | RayDN · nuScenes val ablation | mAP: 39.1 -> 41.3 | Improved |
| Tune Beta distribution to lambda=8, mu=2 instead of uniform (ablation) | RayDN · nuScenes val ablation | mAP: 41.3 -> 41.9 | Improved |
| Increase ray denoising queries from 5 to 7 (ablation) | RayDN · nuScenes val ablation | mAP: 41.9 -> 41.1 | Hurt |
| Set ray radius k=3 instead of 2 or 4 (ablation) | RayDN · nuScenes val ablation | mAP: 41.1 / 41.9 / 41.3 for k=2/3/4 | Mixed |
| Add RayDN to CMT single-frame | CMT · nuScenes val (single frame) | mAP: 30.6 -> 32.1 | Improved |
| Add RayDN to multi-modal CMT | CMT multi-modal · nuScenes val (multi-modal) | mAP: 67.6 -> 68.6 | Improved |
| Add RayDN on top of 3DPPE depth supervision (ablation) | StreamPETR + 3DPPE · nuScenes val ablation | mAP: 42.0 -> 43.9 | Improved |
| Add RayDN (training only) versus adding 3DPPE (ablation) | StreamPETR · nuScenes training cost | training time: 7 h -> 7.5h (3DPPE: 8.5 h); FPS 10.4 unchanged | Improved |
Closest papers by meaning
- Far3D: Expanding the Horizon for Surround-View 3D Object Detection (AAAI 2024 · similarity 0.88)
- GO-N3RDet: Geometry Optimized NeRF-enhanced 3D Object Detector (CVPR 2025 · similarity 0.87)
- MonoDGP: Monocular 3D Object Detection with Decoupled-Query and Geometry-Error Priors (CVPR 2025 · similarity 0.86)
- Cameras as Rays: Pose Estimation via Ray Diffusion (ICLR 2024 · similarity 0.86)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.86)
- MonoCD: Monocular 3D Object Detection with Complementary Depths (CVPR 2024 · similarity 0.86)
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.
