Research map / Detection and segmentation / 3D detection and driving
Distilling Focal Knowledge From Imperfect Expert for 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 FD3D focal distillation | BEVFormer-Tiny · nuScenes val | NDS: 39.02 -> 43.09 | Helped |
| Add FD3D focal distillation | BEVFormer-Small · nuScenes val | NDS: 46.26 -> 48.73 | Helped |
| Add FD3D focal distillation | DETR3D-R50 · nuScenes val | NDS: 35.78 -> 38.95 | Helped |
| Add selective masked generative distillation to CWD (ablation) | BEVFormer Base-Tiny · nuScenes val | NDS: 41.80 -> 42.87 | Helped |
| Add query-based focal distillation to CWD (ablation) | BEVFormer Base-Tiny · nuScenes val | NDS: 41.80 -> 42.75 | Helped |
| Combine selective masked generation and query-based focal distillation (ablation) | BEVFormer Base-Tiny · nuScenes val | NDS: 41.80 -> 43.09 | Helped |
| Replace focal mask with global random mask (ablation) | BEVFormer Base-Tiny · nuScenes val | NDS: 41.80 -> 41.14 | Hurt |
| Use ground-truth foreground regions for the mask (ablation) | BEVFormer Base-Tiny · nuScenes val | NDS: 41.80 -> 42.41 | Helped |
| Mask predicted focal regions instead of using a global random mask (ablation) | BEVFormer Base-Tiny · nuScenes val | NDS: 41.14 -> 42.87 | Helped |
| Combine global distribution and local pattern distillation (ablation) | BEVFormer Base-Tiny · nuScenes val | NDS: 41.80 -> 42.68 | Helped |
| Apply global distillation on BEV features only (ablation) | BEVFormer Base-Small · nuScenes val | NDS: 46.26 -> 45.24 | Hurt |
| Apply global distillation on both PV and BEV features (ablation) | BEVFormer Base-Small · nuScenes val | NDS: 46.26 -> 47.60 | Helped |
Closest papers by meaning
- Towards Efficient 3D Object Detection with Knowledge Distillation (NeurIPS 2022 · similarity 0.89)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.89)
- Weak-to-Strong 3D Object Detection with X-Ray Distillation (CVPR 2024 · similarity 0.86)
- FD3D: Exploiting Foreground Depth Map for Feature-Supervised Monocular 3D Object Detection (AAAI 2024 · similarity 0.86)
- SimDistill: Simulated Multi-Modal Distillation for BEV 3D Object Detection (AAAI 2024 · similarity 0.86)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · 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.