Research map / Detection and segmentation / 3D detection and driving
BEVDistill: Cross-Modal BEV Distillation for Multi-View 3D Object Detection
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 |
|---|---|---|---|
| Add cross-modal BEV distillation | BEVFormer-R50 · nuScenes val | NDS: 42.3 -> 45.7 (+3.4) | Helped |
| Add cross-modal BEV distillation | BEVFormer-T-R50 · nuScenes val | NDS: 48.8 -> 51.5 (+2.7) | Helped |
| Add BEVDistill | BEVDepth · nuScenes test | NDS: 58.9 -> 59.4 | Helped |
| Add dense feature distillation (ablation) | BEVFormer-R50 · nuScenes val | NDS: 42.3 -> 44.7 | Helped |
| Add sparse instance distillation (ablation) | BEVFormer-R50 · nuScenes val | NDS: 42.3 -> 43.9 | Helped |
| Use dense feature and sparse instance distillation together (ablation) | BEVFormer-R50 · nuScenes val | NDS: 42.3 -> 45.7 (+3.4) | Helped |
| Use BEV foreground-guided feature distillation instead of direct feature imitation (ablation) | Cross-modal feature distillation · nuScenes val | NDS: 37.5 -> 38.9 | Helped |
| Use a soft ground-truth heatmap instead of hard center-point masks (ablation) | Foreground mask generation · nuScenes val | NDS: 37.7 -> 38.9 | Helped |
| Use cross-entropy classification distillation with L1 regression (ablation) | Sparse instance distillation · nuScenes val | NDS: 37.5 -> 37.0 | Hurt |
| Replace cross-entropy with InfoNCE for classification distillation (ablation) | Sparse instance distillation · nuScenes val | NDS: 37.0 -> 39.1 | Helped |
| Use InfoNCE with positive and negative feature pairs (ablation) | Sparse instance distillation critic · nuScenes val | NDS: 37.5 -> 38.6 | Helped |
| Use cross-entropy with positive pairs only (ablation) | Sparse instance distillation critic · nuScenes val | NDS: 37.5 -> 35.8 | Hurt |
Closest papers by meaning
- SimDistill: Simulated Multi-Modal Distillation for BEV 3D Object Detection (AAAI 2024 · similarity 0.92)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.90)
- BEVNeXt: Reviving Dense BEV Frameworks for 3D Object Detection (CVPR 2024 · similarity 0.90)
- Towards Generalizable Multi-Camera 3D Object Detection via Perspective Rendering (AAAI 2025 · similarity 0.89)
- UniMODE: Unified Monocular 3D Object Detection (CVPR 2024 · similarity 0.89)
- GAFusion: Adaptive Fusing LiDAR and Camera with Multiple Guidance for 3D Object Detection (CVPR 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.