Research map / Detection and segmentation / 3D detection and driving
MetaBEV: Solving Sensor Failures for 3D Detection and Map Segmentation
Chongjian Ge, Junsong Chen, Enze Xie, Zhongdao Wang, Lanqing Hong, Huchuan Lu, Zhenguo Li, Ping Luo
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 switched modality training (ablation) | MetaBEV · nuScenes val, missing LiDAR (detection) | NDS: 9.8 -> 42.6 | Improved |
| add switched modality training (ablation) | MetaBEV · nuScenes val, full modality (detection) | NDS: 71.0 -> 71.5 | Improved |
| add switched modality training (ablation) | MetaBEV · nuScenes val, full modality (segmentation) | mIoU: 70.4 -> 68.5 | Hurt |
| add two self-attention layers after cross-attention layers (ablation) | MetaBEV decoder · nuScenes val (detection) | NDS: 68.3 -> 69.3 (vs 6 cross); 67.0 -> 69.3 (vs 4 cross) | Improved |
| stack more cross-attention layers (6 or 8) (ablation) | MetaBEV decoder · nuScenes val (detection) | NDS: 68.3 (6 layers) vs 68.2 (8 layers) | No clear effect |
| increase number of reference points beyond 8 (ablation) | MetaBEV decoder · nuScenes val (detection) | mAP/NDS: 62.7/68.2 (8) vs 62.3/68.0 (12) | No clear effect |
| add routing-based MoE (R-MoE) to FFN (ablation) | MetaBEV multi-task · nuScenes val multi-task | mIoU: 64.7 -> 66.9 | Improved |
| add task-specific experts without router (H-MoE) (ablation) | MetaBEV multi-task · nuScenes val multi-task | mIoU: 64.7 -> 66.3 | Improved |
| use 2/8 experts instead of 1/2 expert (ablation) | MetaBEV R-MoE · nuScenes val multi-task | mIoU: 63.3 -> 66.9 | Improved |
| extend switched modality training from 8 to 20 epochs (ablation) | MetaBEV segmentation · nuScenes val, missing camera | mIoU: 53.7 -> 58.3 | Improved |
| replace CNN concat fusion with meta-BEV cross-modal attention decoder | BEVFusion vs MetaBEV · nuScenes val, missing camera (segmentation) | mIoU: 4.1 -> 53.7 | Improved |
| replace CNN concat fusion with meta-BEV decoder under beam reduction (zero-shot) | BEVFusion vs MetaBEV · nuScenes val, 4 beams zero-shot | NDS: 58.3 -> 57.7 | Hurt |
Later papers that compared with this method
1 later accepted paper reports MetaBEV as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- GaussianFusion: Unified 3D Gaussian Representation for Multi-Modal Fusion Perception (ICLR 2026 · better on 2 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
- PointBeV: A Sparse Approach for BeV Predictions (CVPR 2024 · similarity 0.87)
- GraphBEV: Towards Robust BEV Feature Alignment for Multi-Modal 3D Object Detection (ECCV 2024 · similarity 0.87)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 2025 · similarity 0.87)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.86)
- BEVFormer: Learning Bird’s-Eye-View Representation From LiDAR-Camera via Spatiotemporal Transformers (TPAMI 2024 · similarity 0.86)
- M-BEV: Masked BEV Perception for Robust Autonomous Driving (AAAI 2024 · similarity 0.86)
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.
