Research map / 3D vision / Depth and stereo
BEVDepth: Acquisition of Reliable Depth 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 |
|---|---|---|---|
| Replace predicted depth with point-cloud ground-truth depth | Lift-splat detector · nuScenes val | mAP: 0.282 -> 0.470 | Helped |
| Replace predicted depth with point-cloud ground-truth depth | Lift-splat detector · nuScenes val | mATE: 0.768 -> 0.393 | Helped |
| Replace learned depth with frozen random soft depth (ablation) | Lift-splat detector · nuScenes val | mAP: 0.282 -> 0.245 | Hurt |
| Replace learned depth with frozen random one-hot depth (ablation) | Lift-splat detector · nuScenes val | mAP: 0.282 -> 0.176 | Hurt |
| Add explicit depth loss (ablation) | Base Detector · nuScenes val | mAP: 0.282 -> 0.304 | Helped |
| Add camera-aware depth prediction (ablation) | BEVDepth with depth loss · nuScenes val | mATE: 0.747 -> 0.706 | Helped |
| Add Depth Refinement Module (ablation) | BEVDepth with depth loss and camera awareness · nuScenes val | mAP: 0.314 -> 0.322 | Helped |
| Add multi-frame fusion (ablation) | BEVDepth with Depth Refinement Module · nuScenes val | NDS: 0.367 -> 0.442 | Helped |
| Use BCE rather than L1 depth loss (ablation) | BEVDepth · nuScenes val | mAP: 0.322 vs 0.321 | No clear effect |
| Combine BCE and L1 depth losses (ablation) | BEVDepth · nuScenes val | mAP: 0.322 -> 0.323 | No clear effect |
| Use a 1×3 kernel that does not exchange information along depth (ablation) | Depth Refinement Module · nuScenes val | mAP: 0.314 -> 0.315 | No clear effect |
| Enable depth-axis feature interaction with a 3×1 kernel (ablation) | Depth Refinement Module · nuScenes val | mAP: 0.314 -> 0.320 | Helped |
Closest papers by meaning
- BEVNeXt: Reviving Dense BEV Frameworks for 3D Object Detection (CVPR 2024 · similarity 0.89)
- Instance-Aware Multi-Camera 3D Object Detection with Structural Priors Mining and Self-Boosting Learning (AAAI 2024 · similarity 0.89)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.86)
- BEVSpread: Spread Voxel Pooling for Bird's-Eye-View Representation in Vision-based Roadside 3D Object Detection (CVPR 2024 · similarity 0.85)
- UniMODE: Unified Monocular 3D Object Detection (CVPR 2024 · similarity 0.85)
- Towards Generalizable Multi-Camera 3D Object Detection via Perspective Rendering (AAAI 2025 · similarity 0.85)
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.