Research map / Detection and segmentation / 3D detection and driving
SA-BEV: Generating Semantic-Aware Bird's-Eye-View Feature for Multi-view 3D Object Detection
Jinqing Zhang, Yanan Zhang, Qingjie Liu, Yunhong Wang
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 SA-BEVPool (filter background virtual points) to BEVDepth (ablation) | BEVDepth · nuScenes val (ResNet-50) | mAP / NDS: 0.330 -> 0.340 mAP, 0.436 -> 0.449 NDS | Improved |
| add BEV-Paste augmentation on top of SA-BEVPool (ablation) | BEVDepth · nuScenes val (ResNet-50) | mAP / NDS: 0.340 -> 0.354 mAP, 0.449 -> 0.464 NDS | Improved |
| add MSCT head on top of pool and paste (ablation) | BEVDepth · nuScenes val (ResNet-50) | mAP / NDS: 0.354 -> 0.365 mAP, 0.464 -> 0.483 NDS | Improved |
| apply SA-BEVPool and BEV-Paste to BEVDet | BEVDet · nuScenes val (ResNet-50) | mAP: 0.278 -> 0.304 | Improved |
| apply SA-BEVPool and BEV-Paste to BEVStereo | BEVStereo · nuScenes val (ResNet-50) | mAP: 0.349 -> 0.364 | Improved |
| raise semantic threshold T_S from 0.25 to 0.50 (ablation) | SA-BEVPool · nuScenes val (ResNet-50) | mAP / NDS: 0.340 -> 0.329 mAP, 0.449 -> 0.432 NDS | Hurt |
| paste more frames in BEV-Paste (N_P from 1 to 2) (ablation) | BEV-Paste · nuScenes val (ResNet-50) | NDS: 0.453 -> 0.452 | No clear effect |
| add extra BEV data augmentation to the pasted frame (ablation) | BEV-Paste · nuScenes val (ResNet-50) | NDS: 0.453 -> 0.464 | Improved |
| add multi-task distillation module to the head (ablation) | MSCT head · nuScenes val (ResNet-50) | NDS: 0.464 -> 0.468 | Improved |
| use multi-scale image features in the head (ablation) | MSCT head · nuScenes val (ResNet-50) | NDS: 0.464 -> 0.478 | Improved |
| SA-BEV versus BEVStereo with V2-99 on test set | multi-view 3D detection · nuScenes test | mAAE: 0.138 -> 0.139 | Hurt |
| SA-BEV versus BEVStereo on val with ConvNeXt-B | multi-view 3D detection · nuScenes val | mAP: 0.478 -> 0.479 | No clear effect |
Later papers that compared with this method
2 later accepted papers report SA-BEV as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- CLIPDet3D: Vision-Language Collaborative Distillation for 3D Object Detection (AAAI 2026 · better on 1 of 1 shared result)
- FSD-BEV: Foreground Self-Distillation for Multi-view 3D Object Detection (ECCV 2024 · better on 4 of 4 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PETRv2 — PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images (ICCV 2023) · better on 0 of 1 shared result
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
- Instance-Aware Multi-Camera 3D Object Detection with Structural Priors Mining and Self-Boosting Learning (AAAI 2024 · similarity 0.91)
- Improving Bird's Eye View Semantic Segmentation by Task Decomposition (CVPR 2024 · similarity 0.90)
- MamBEV: Enabling State Space Models to Learn Birds-Eye-View Representations (ICLR 2025 · similarity 0.89)
- A Black-Box Evaluation Framework for Semantic Robustness in Bird’s Eye View Detection (AAAI 2025 · similarity 0.89)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.88)
- SOGDet: Semantic-Occupancy Guided Multi-View 3D Object Detection (AAAI 2024 · 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.
