Research map / Label-efficient and robust learning / Knowledge distillation
Distilling Image Classifiers in Object Detectors
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 classifier-to-detector distillation (KD_cls and KD_loc) | Faster RCNN-QR50 · COCO2017 val | mAP: 23.3 -> 27.2 | Helped |
| Add classifier-to-detector distillation (KD_cls and KD_loc) | SSD512-VGG16 · COCO2017 val | mAP: 29.4 -> 32.1 | Helped |
| Combine classifier-to-detector distillation with FKD | Faster RCNN-R50 · COCO2017 val | mAP: 41.5 -> 41.9 | Helped |
| Add classification distillation (ablation) | SSD300-VGG16 · COCO2017 val | mAP: 25.6 -> 26.3 | Helped |
| Add localization distillation (ablation) | SSD300-VGG16 · COCO2017 val | mAP: 25.6 -> 27.2 | Helped |
| Use ResNet50 rather than ResNet18 as the classification teacher (ablation) | SSD300-VGG16 with KD_cls · COCO2017 val | mAP: 25.9 -> 26.3 | Helped |
| Use ResNeXt101 rather than ResNet50 as the classification teacher (ablation) | SSD300-VGG16 with KD_cls · COCO2017 val | mAP: 26.3 -> 25.3 | Hurt |
| Increase classification-teacher input size among the tested sizes (ablation) | SSD300-VGG16 with KD_cls · COCO2017 val | mAP: 26.2; 26.3; 26.2 | No clear effect |
| Increase localization sampling size from 14 × 14 to 112 × 112 (ablation) | SSD300-VGG16 with KD_loc · COCO2017 val | mAP: 26.4 -> 27.0 | Helped |
| Use 8 × 8 rather than 4 × 4 adaptive pooling (ablation) | SSD300-VGG16 with KD_loc · COCO2017 val | mAP: 27.0 -> 27.1 | Helped |
| Add first-convolution features to image-region localization distillation (ablation) | SSD300-VGG16 with KD_loc · COCO2017 val | mAP: 27.1 -> 27.2 | Helped |
| Add bottleneck-block features to ℓ₀ and ℓ₁ localization distillation (ablation) | SSD300-VGG16 with KD_loc · COCO2017 val | mAP: 27.2 -> 26.9 | Hurt |
Closest papers by meaning
- Learning Lightweight Object Detectors via Multi-Teacher Progressive Distillation (ICML 2023 · similarity 0.92)
- Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors (ICLR 2021 · similarity 0.92)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.91)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.90)
- Understanding the Role of the Projector in Knowledge Distillation (AAAI 2024 · similarity 0.88)
- CrossKD: Cross-Head Knowledge Distillation for 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.