Research map / Point clouds / Segmentation and detection
Frustum PointNets for 3D Object Detection From RGB-D Data
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 Frustum PointNets (v2) instead of MV3D | KITTI 3D detection · KITTI test, cars moderate | 3D AP: 62.35 -> 70.39 | Helped |
| use 3D instance segmentation rather than no mask (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 33.5 -> 74.3 | Helped |
| use 3D instance segmentation instead of 2D masks (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 61.6 -> 74.3 | Helped |
| combine 2D and 3D masks rather than use 3D segmentation alone (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 74.3 -> 70.0 | Hurt |
| add frustum rotation (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 12.5 -> 48.1 | Helped |
| add mask centralization (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 12.5 -> 64.6 | Helped |
| add T-Net alignment to the object center (ablation) | Frustum PointNet with rotation and mask centralization · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 71.5 -> 74.3 | Helped |
| replace regression-only box loss with cls-reg loss (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 62.9 -> 71.8 | Helped |
| normalize cls-reg residual targets (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 71.8 -> 72.2 | Helped |
| add corner loss to normalized cls-reg (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 72.2 -> 74.3 | Helped |
| replace PointNet v1 with PointNet++ v2 for segmentation and box estimation (ablation) | Frustum PointNet · KITTI val, cars | 3D box estimation accuracy (IoU=0.7): 74.3 -> 77.1 | Helped |
| combine bird's-eye-view proposals with RGB frustum proposals (ablation) | Frustum PointNet · KITTI val, cars hard | 3D AP: 63.65 -> 67.47 | Helped |
Closest papers by meaning
- 3D Object Detection Using Scale Invariant and Feature Reweighting Networks (AAAI 2019 · similarity 0.90)
- PointFusion: Deep Sensor Fusion for 3D Bounding Box Estimation (CVPR 2018 · similarity 0.89)
- PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud (CVPR 2019 · similarity 0.89)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.88)
- PerspectiveNet: 3D Object Detection from a Single RGB Image via Perspective Points (NeurIPS 2019 · similarity 0.88)
- Rooms from Motion: Un-posed Indoor 3D Object Detection as Localization and Mapping (NeurIPS 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.