Research map / Point clouds / Segmentation and detection
Fully Convolutional One-Stage 3D Object Detection on LiDAR Range Images
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use 5-round MRV instead of 1 round (ablation) | FCOS-LiDAR multi-frame · nuScenes val | mAP (%): 55.02 -> 57.08 | Helped |
| increase MRV rounds from 5 to 7 (ablation) | FCOS-LiDAR multi-frame · nuScenes val | mAP (%): 57.08 -> 56.99 | No clear effect |
| MRV rounds 1 to 3 in single-frame (ablation) | FCOS-LiDAR single-frame · nuScenes val | mAP (%): 53.14 -> 53.76 | Helped |
| prioritize current-frame points on collision (ablation) | FCOS-LiDAR multi-frame · nuScenes val | mAP (%): 54.29 -> 57.08 | Helped |
| per-channel modality-wise conv instead of full conv (ablation) | FCOS-LiDAR · nuScenes val | mAP (%): 56.35 -> 57.08 | Helped |
| group same-type channels across frames (ablation) | FCOS-LiDAR · nuScenes val | mAP (%): 56.83 -> 57.08 | Helped |
| multi-dilation modality-wise branches (1,3,6) vs single (ablation) | FCOS-LiDAR · nuScenes val | mAP (%): 55.87 -> 57.08 | Helped |
| untie detection head weights across FPN levels (ablation) | FCOS-LiDAR · nuScenes val | mAP (%): 56.44 -> 57.08 | Helped |
| range view with FCOS-LiDAR vs BEV CenterPoint, single frame | FCOS-LiDAR · nuScenes val single-frame | mAP (%): 52.83 -> 53.42 | Helped |
| range view vs BEV CenterPoint, multi-frame | FCOS-LiDAR · nuScenes val multi-frame | mAP (%): 60.40 -> 57.08 | Hurt |
| wider detection heads (64 to 128 channels) | FCOS-LiDAR · nuScenes val | mAP (%): about 0.6% | Helped |
Closest papers by meaning
- RangePerception: Taming LiDAR Range View for Efficient and Accurate 3D Object Detection (NeurIPS 2023 · similarity 0.92)
- RangeIoUDet: Range Image Based Real-Time 3D Object Detector Optimized by Intersection Over Union (CVPR 2021 · similarity 0.92)
- LaserNet: An Efficient Probabilistic 3D Object Detector for Autonomous Driving (CVPR 2019 · similarity 0.91)
- Fully Sparse 3D Object Detection (NeurIPS 2022 · similarity 0.91)
- GBlobs: Explicit Local Structure via Gaussian Blobs for Improved Cross-Domain LiDAR-based 3D Object Detection (CVPR 2025 · similarity 0.89)
- BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · similarity 0.88)
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.