Research map / Detection and segmentation / 3D detection and driving
Towards Efficient 3D Object Detection with Knowledge Distillation
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 pivotal position logit KD (ablation) | CP-Voxel-XXS · Waymo | LEVEL 2 mAPH: 56.26 -> 57.68 | Helped |
| add teacher guided initialization (TGI) (ablation) | CP-Voxel-XXS · Waymo | LEVEL 2 mAPH: 56.26 -> 57.61 | Helped |
| add feature KD on top of PP, label KD and TGI (ablation) | CP-Voxel-XXS · Waymo | LEVEL 2 mAPH: 59.28 -> 58.67 | Hurt |
| add label KD to PP + TGI (ablation) | CP-Voxel-XXS · Waymo | LEVEL 2 mAPH: 58.83 -> 59.28 | Helped |
| replace GID-L with Gaussian PP logit KD (ablation) | CP-Voxel-S · Waymo | LEVEL 2 mAPH: 62.89 -> 64.16 | Helped |
| use OFA remap instead of FNA in TGI (ablation) | CP-Voxel-XS · Waymo | LEVEL 2 mAPH: 62.43 -> 60.06 | Hurt |
| inherit weights from large teacher CP-Voxel rather than CP-Voxel-XS (ablation) | CP-Voxel-XXS · Waymo | LEVEL 2 mAPH: 57.61 -> 54.92 | Hurt |
| add teacher predictions to classification label assignment only (ablation) | CP-Pillar-v0.48 · Waymo | LEVEL 2 mAPH: 56.24 -> 56.47 | Helped |
| add teacher predictions for classification and regression label assignment (ablation) | CP-Pillar-v0.48 · Waymo | LEVEL 2 mAPH: 56.24 -> 57.54 | Helped |
| combine GID-L, label KD and FG naively (ablation) | CP-Voxel-XXS · Waymo | LEVEL 2 mAPH: 57.92 -> 57.62 | Mixed |
| improved KD pipeline vs best existing KD | CP-Pillar-v0.64 · Waymo | LEVEL 2 mAPH: 53.83 -> 55.82 | Helped |
| width compression instead of depth compression (ablation) | CP-Voxel · Waymo | CPR: 0.57 -> 0.67 | Helped |
Closest papers by meaning
- PKD: General Distillation Framework for Object Detectors via Pearson Correlation Coefficient (NeurIPS 2022 · similarity 0.91)
- Learning Efficient Object Detection Models with Knowledge Distillation (NeurIPS 2017 · similarity 0.91)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.90)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.89)
- CrossKD: Cross-Head Knowledge Distillation for Object Detection (CVPR 2024 · similarity 0.87)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.87)
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.