Research map / Detection and segmentation / 3D detection and driving
Graph R-CNN: Towards Accurate 3D Object Detection with Semantic-Decorated Local Graph
Honghui Yang, Zili Liu, Xiaopei Wu, Wenxiao Wang, Wei Qian, Xiaofei He, Deng Cai
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 RoI-graph pooling (RGP) (ablation) | CenterPoint-Voxel baseline · WOD val vehicle LEVEL 2 | 3D APH: 66.17 -> 69.48 | Improved |
| add RoI-graph pooling (RGP) (ablation) | CenterPoint-Voxel baseline · WOD val pedestrian LEVEL 2 | 3D APH: 62.67 -> 66.79 | Improved |
| add multi-level attentive fusion (MLAF) (ablation) | Graph R-CNN with RGP · WOD val vehicle LEVEL 2 | 3D APH: 69.48 -> 69.77 | Improved |
| add dynamic point aggregation (DPA) (ablation) | Graph R-CNN with RGP and MLAF · WOD val vehicle LEVEL 2 | 3D APH: 69.77 -> 70.83 | Improved |
| add dynamic point aggregation (DPA) (ablation) | Graph R-CNN with RGP and MLAF · WOD val vehicle LEVEL 2 (0-30m) | 3D APH: 90.25 -> 91.74 | Improved |
| add PointNet encoding of neighbor points (ablation) | Graph R-CNN · WOD val vehicle LEVEL 2 | 3D APH: 70.83 -> 71.04 | Improved |
| add two diagonal corners per graph node (ablation) | Graph R-CNN · WOD val vehicle LEVEL 2 | 3D APH: 71.04 -> 72.10 | Improved |
| use patch search instead of point cloud region pooling for grouping (ablation) | Graph R-CNN · WOD val | runtime (ms): 69.7 -> 1.9 | Improved |
| increase graph iterations from T=1 to T=3 (ablation) | Graph R-CNN · WOD val vehicle LEVEL 2 | 3D APH: 68.76 -> 69.48 | Improved |
| replace channel-wise Transformer aggregation with max pooling (ablation) | Graph R-CNN · WOD val vehicle LEVEL 2 | 3D APH: 71.67 -> 72.10 | Improved |
| use DFVS instead of FPS for sampling (ablation) | 3DSSD · KITTI val car | 3D mAP / runtime: 81.73 -> 81.75; 29.6 -> 20.3 | No clear effect |
| add image features (2D detector output) to local graphs (ablation) | Graph R-CNN · KITTI val car | 3D AP: 85.38 -> 86.87 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Voxel R-CNN — Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detection (AAAI 2021) · better on 3 of 3 shared results
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
- Pyramid R-CNN: Towards Better Performance and Adaptability for 3D Object Detection (ICCV 2021 · similarity 0.92)
- Fast Point R-CNN (ICCV 2019 · similarity 0.91)
- Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detection (AAAI 2021 · similarity 0.91)
- VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection (CVPR 2018 · similarity 0.91)
- HVPR: Hybrid Voxel-Point Representation for Single-Stage 3D Object Detection (CVPR 2021 · similarity 0.91)
- ViKIENet: Towards Efficient 3D Object Detection with Virtual Key Instance Enhanced Network (CVPR 2025 · similarity 0.89)
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.
