Research map / 3D vision / 3D generation
Open-Vocabulary BEV Segmentation with 3D-Aware Geometric Constraints
Hojun Choi, Seulbin Hwang, Dae Kim, Kisung Kim, Hyunjung Shim, Jinhan Lee
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 pseudo-BEV labeling (PBL) from projected 3D detections (ablation) | OVBEVSeg · nuScenes multi-class OVBS | novel IoU: 0.0 -> 14.0 | Improved |
| add BAGD distillation using vanilla 3DGS (no BAGS) (ablation) | OVBEVSeg · nuScenes multi-class OVBS | mean IoU: 17.3 -> 7.9 | Hurt |
| use BEV-aware 3DGS (BAGS) supervision with BAGD (ablation) | OVBEVSeg · nuScenes multi-class OVBS | mean IoU: 7.9 -> 19.3 (PBL only: 17.3) | Improved |
| increase distillation ratio from 0 to 0.1 (ablation) | OVBEVSeg · nuScenes multi-class OVBS | novel IoU: 14.0 -> 15.3 | Improved |
| use pseudo-labels from projected 3D boxes instead of 2D-OV-then-BEV lifting | OVBEVSeg · nuScenes novel classes (0% GT) | mIoU: 2.1 -> 15.3 | Improved |
| use OVBEVSeg with 0% novel GT vs semi-supervised baselines with GT | OVBEVSeg · nuScenes novel classes | mIoU: 15.3 vs Semi-BEVseg 14.6 (20% GT) and 17.0 (40% GT) | Mixed |
| replace UNION 3D prior with F&B VLM-based prior (ablation) | OVBEVSeg · nuScenes novel classes | IoU: truck 19.0 -> 14.2; motorcycle 6.6 -> 7.9; cone 11.5 | Mixed |
| apply OVBEVSeg vs closed-set projection methods | OVBEVSeg · nuScenes single-class (vehicle novel) | mIoU: 24.4 (PointBeV) -> 26.6 | Improved |
| BAGD-enhanced model on map classes | OVBEVSeg · nuScenes map segmentation | IoU: walkway 50.2 -> 51.9; crossing 46.3 -> 49.4 | Improved |
| OVBEVSeg vs closed-set methods on base class bicycle and car | OVBEVSeg · nuScenes multi-class OVBS | base-class IoU: car 42.8 -> 41.8 (TaDe); bicycle 14.2 -> 12.0 | Mixed |
Closest papers by meaning
- Improving Bird's Eye View Semantic Segmentation by Task Decomposition (CVPR 2024 · similarity 0.90)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.88)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.88)
- GA-VLN: Geometry-Aware BEV Representation for Efficient Vision-Language Navigation (CVPR 2026 · similarity 0.88)
- 3D-AVS: LiDAR-based 3D Auto-Vocabulary Segmentation (CVPR 2025 · similarity 0.88)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 2025 · 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.
