Research map / Point clouds / Segmentation and detection
FCAF3D: Fully Convolutional Anchor-Free 3D Object Detection
Danila Rukhovich, Anna Vorontsova, Anton Konushin
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 |
|---|---|---|---|
| use Mobius OBB parametrization instead of naive (ablation) | VoteNet (IoU loss) · SUN RGB-D | mAP@0.5: 38.4 -> 40.4 | Improved |
| use Mobius OBB parametrization instead of sin-cos (ablation) | VoteNet (IoU loss) · SUN RGB-D | mAP@0.5: 37.1 -> 40.4 | Improved |
| use Mobius OBB parametrization instead of naive (ablation) | ImVoteNet (IoU loss) · SUN RGB-D | mAP@0.5: 39.1 -> 40.8 | Improved |
| use Mobius OBB parametrization instead of naive (ablation) | ImVoxelNet · SUN RGB-D | mAP@0.5: 13.8 -> 14.6 | Improved |
| use Mobius OBB parametrization instead of naive (ablation) | FCAF3D · SUN RGB-D | mAP@0.5: 46.8 -> 48.9 | Improved |
| add anchors to GSDN (versus anchor-free GSDN) (ablation) | GSDN · ScanNet V2 | mAP@0.5: 22.7 -> 34.8 | Improved |
| replace GSDN with anchor-free FCAF3D at same voxel size and stride | GSDN · ScanNet V2 | mAP@0.5: 34.8 -> 46.2 | Improved |
| reduce voxel size from 0.03 to 0.01 m (ablation) | FCAF3D · ScanNet V2 | mAP@0.5: 42.6 -> 57.3 | Improved |
| increase input points from 20k to 100k (ablation) | FCAF3D · ScanNet V2 | mAP@0.5: 52.8 -> 57.3 | Improved |
| add centerness (ablation) | FCAF3D · ScanNet V2 | mAP@0.5: 56.1 -> 57.3 | Improved |
| add centerness (ablation) | FCAF3D · S3DIS | mAP@0.25: 67.9 -> 66.7 | Hurt |
| use 18 center-sampling locations instead of 9 (ablation) | FCAF3D · ScanNet V2 | mAP@0.5: 55.7 -> 57.3 | Improved |
Later papers that compared with this method
6 later accepted papers report FCAF3D as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Real-time 3D Object Detection with Inference-Aligned Learning (AAAI 2026 · better on 2 of 2 shared results)
- One for All: Multi-Domain Joint Training for Point Cloud Based 3D Object Detection (NeurIPS 2024 · better on 2 of 2 shared results)
- Uni3DETR: Unified 3D Detection Transformer (NeurIPS 2023 · better on 6 of 6 shared results)
- ConDaFormer: Disassembled Transformer with Local Structure Enhancement for 3D Point Cloud Understanding (NeurIPS 2023 · better on 2 of 2 shared results)
- GaPro: Box-Supervised 3D Point Cloud Instance Segmentation Using Gaussian Processes as Pseudo Labelers (ICCV 2023 · better on 1 of 1 shared result)
- Autoregressive Uncertainty Modeling for 3D Bounding Box Prediction (ECCV 2022 · 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
- Uni3D: A Unified Baseline for Multi-Dataset 3D Object Detection (CVPR 2023 · similarity 0.89)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.88)
- Rooms from Motion: Un-posed Indoor 3D Object Detection as Localization and Mapping (NeurIPS 2025 · similarity 0.88)
- SPGroup3D: Superpoint Grouping Network for Indoor 3D Object Detection (AAAI 2024 · similarity 0.88)
- UniDet3D: Multi-dataset Indoor 3D Object Detection (AAAI 2025 · similarity 0.88)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · 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.
