Research map / Detection and segmentation / 3D detection and driving
H3DNet: 3D Object Detection Using Hybrid Geometric Primitives
Zaiwei Zhang, Bo Sun, Haitao Yang, Qixing Huang
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 refinement of object center (ablation) | H3DNet · ScanNet V2 val | mAP@0.5: 46.3 -> 48.1 | Improved |
| add refinement of object size (ablation) | H3DNet · ScanNet V2 val | mAP@0.5: 44.2 -> 48.1 | Improved |
| add refinement of semantic label (ablation) | H3DNet · ScanNet V2 val | mAP@0.5: 47.3 -> 48.1 | Improved |
| add refinement of object existence score (ablation) | H3DNet · ScanNet V2 val | mAP@0.5: 45.1 -> 48.1 | Improved |
| add refinement of heading angle (ablation) | H3DNet · SUN RGB-D V1 val | mAP@0.25: 58.6 -> 60.1 | Improved |
| add refinement of heading angle (ablation) | H3DNet · SUN RGB-D V1 val | mAP@0.5: 36.6 -> 39.0 | Improved |
| add refinement of object center (ablation) | H3DNet · ScanNet V2 val | mAP@0.25: 66.9 -> 67.2 | Improved |
| add primitive matching and proposal refinement (vs classifying initial proposals only) (ablation) | H3DNet · ScanNet V2 val | mAP@0.5: 14.6% drop on mAP 0.5 without | Improved |
| increase descriptor computation towers from 1 to 4 (ablation) | H3DNet · ScanNet V2 val | mAP@0.25: 64.4 -> 67.2 | Improved |
| increase descriptor computation towers from 1 to 4 (ablation) | H3DNet · ScanNet V2 val | mAP@0.5: 43.4 -> 48.3 | Improved |
| give VoteNet four descriptor towers (ablation) | VoteNet · ScanNet V2 val | mAP@0.5: 33.5 -> 37.12 | No clear effect |
| give VoteNet four descriptor towers (ablation) | VoteNet · SUN RGB-D V1 val | mAP@0.25: 57.7 -> 57.5 | No clear effect |
Later papers that compared with this method
6 later accepted papers report H3DNet as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Distribution Aware VoteNet for 3D Object Detection (AAAI 2022 · better on 3 of 4 shared results)
- Canonical Voting: Towards Robust Oriented Bounding Box Detection in 3D Scenes (CVPR 2022 · better on 1 of 1 shared result)
- 4DContrast: Contrastive Learning with Dynamic Correspondences for 3D Scene Understanding (ECCV 2022 · better on 1 of 1 shared result)
- RandomRooms: Unsupervised Pre-Training From Synthetic Shapes and Randomized Layouts for 3D Object Detection (ICCV 2021 · better on 6 of 6 shared results)
- An End-to-End Transformer Model for 3D Object Detection (ICCV 2021 · better on 0 of 4 shared results)
- Semantic Consistency Networks for 3D Object Detection (AAAI 2021 · better on 0 of 2 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
- HVPR: Hybrid Voxel-Point Representation for Single-Stage 3D Object Detection (CVPR 2021 · similarity 0.91)
- Omni3D: A Large Benchmark and Model for 3D Object Detection in the Wild (CVPR 2023 · similarity 0.89)
- 3D Bounding Box Estimation Using Deep Learning and Geometry (CVPR 2017 · similarity 0.89)
- Deep Hough Voting for 3D Object Detection in Point Clouds (ICCV 2019 · similarity 0.89)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.87)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.87)
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.
