Research map / Label-efficient and robust learning / Knowledge distillation
Prediction-Guided Distillation for Dense Object Detection
Chenhongyi Yang, Mateusz Ochal, Amos Storkey, Elliot J. Crowley
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 |
|---|---|---|---|
| use PGD prediction-guided distillation instead of FGD | student distillation · COCO val ATSS R101->R50 | AP: 42.6 -> 44.2 | Improved |
| use PGD instead of FGD with MobileNetV2 student | ATSS · COCO val | AP: 35.8 -> 38.3 | Improved |
| use PGD instead of FGD on CrowdHuman | DDOD · CrowdHuman val | AP: 89.3 -> 90.0 | Improved |
| use PGD instead of FGD on CrowdHuman | DDOD · CrowdHuman val | MR: 43.1 -> 42.8 | Improved |
| add Gaussian-weighted top-K key predictive regions over uniform GT-box distillation (ablation) | ATSS · COCO val | AP: 43.3 -> 44.2 | Improved |
| add Gaussian centre weighting to Box strategy (ablation) | Box · COCO val ATSS | AP: +0.4 AP | Improved |
| distil fixed centre region instead of whole box (ablation) | Box · COCO val ATSS | AP: 43.3 -> 43.1 | Hurt |
| use top-K selection instead of quality-score weighting over all positions (ablation) | Quality · COCO val ATSS | AP: 43.8 -> 44.2 | Improved |
| use K=30 top-K positions instead of K=1 (ablation) | PGD · COCO val ATSS | AP: 43.2 -> 44.2 | Improved |
| use K=30 instead of K=60 (ablation) | PGD · COCO val ATSS | AP: 43.9 -> 44.2 | Improved |
| add regression KD loss to classification KD (ablation) | PGD · COCO val ATSS | AP: 43.8 -> 44.2 | Improved |
| add classification KD to regression KD (ablation) | PGD · COCO val ATSS | AP: 41.7 -> 44.2 | Improved |
Closest papers by meaning
- Learning Efficient Object Detection Models with Knowledge Distillation (NeurIPS 2017 · similarity 0.93)
- Structured Knowledge Distillation for Accurate and Efficient Object Detection (TPAMI 2023 · similarity 0.93)
- Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors (ICLR 2021 · similarity 0.93)
- When Object Detection Meets Knowledge Distillation: A Survey (TPAMI 2023 · similarity 0.93)
- 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 0.90)
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.
