Research map / Detection and segmentation / 3D detection and driving
RangeDet: In Defense of Range View for LiDAR-Based 3D Object Detection
Lue Fan, Xuan Xiong, 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 |
|---|---|---|---|
| add Meta-Kernel convolution (ablation) | RangeDet baseline (A1) · WOD vehicle | 3D AP: 53.39 -> 56.58 | Improved |
| add Range Conditioned Pyramid (ablation) | RangeDet baseline (A1) · WOD vehicle | 3D AP: 53.39 -> 58.37 | Improved |
| add IoU prediction head (ablation) | RangeDet (Meta-Kernel + RCP) · WOD vehicle | 3D AP: 61.05 -> 64.61 | Improved |
| add weighted NMS (ablation) | RangeDet (A5) · WOD vehicle | 3D AP: 64.61 -> 69.00 | Improved |
| add range-view data augmentation (ablation) | RangeDet (A6) · WOD vehicle | 3D AP: 69.00 -> 72.85 | Improved |
| add Meta-Kernel (ablation) | RangeDet · WOD pedestrian | 3D AP: 69.06 -> 74.16 | Improved |
| aggregate by concatenation instead of max-pooling (ablation) | Meta-Kernel · WOD vehicle | 3D AP: 63.47 -> 67.37 | Improved |
| aggregate by concatenation instead of summation (ablation) | Meta-Kernel · WOD vehicle | 3D AP: 63.52 -> 67.37 | Improved |
| generate kernel weights from relative pixel coordinates only (ablation) | Meta-Kernel · WOD vehicle | 3D AP: 63.57 -> 63.87 | No clear effect |
| generate kernel weights from relative Cartesian coordinates (ablation) | Meta-Kernel · WOD vehicle | 3D AP: 63.57 -> 67.00 | Improved |
| add relative range to relative Cartesian coordinates as meta-data (ablation) | Meta-Kernel · WOD vehicle | 3D AP: 67.00 -> 67.37 | No clear effect |
| place Meta-Kernel at stride 1 instead of stride 8 (ablation) | RangeDet · WOD vehicle | 3D AP: 63.80 -> 67.37 | Improved |
Later papers that compared with this method
2 later accepted papers report RangeDet as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- RangePerception: Taming LiDAR Range View for Efficient and Accurate 3D Object Detection (NeurIPS 2023 · better on 6 of 6 shared results)
- Enhancing Multi-modal Features Using Local Self-Attention for 3D Object Detection (ECCV 2022 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- LaserNet — LaserNet: An Efficient Probabilistic 3D Object Detector for Autonomous Driving (CVPR 2019) · better on 2 of 2 shared results
- Voxel-RCNN — Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detection (AAAI 2021) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- RangePerception: Taming LiDAR Range View for Efficient and Accurate 3D Object Detection (NeurIPS 2023 · similarity 0.93)
- Fully Convolutional One-Stage 3D Object Detection on LiDAR Range Images (NeurIPS 2022 · similarity 0.93)
- RangeIoUDet: Range Image Based Real-Time 3D Object Detector Optimized by Intersection Over Union (CVPR 2021 · similarity 0.93)
- LaserNet: An Efficient Probabilistic 3D Object Detector for Autonomous Driving (CVPR 2019 · similarity 0.90)
- Rethinking Range View Representation for LiDAR Segmentation (ICCV 2023 · similarity 0.90)
- Far3D: Expanding the Horizon for Surround-View 3D Object Detection (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.
