Research map / Label-efficient and robust learning / Knowledge distillation
Distilling Object Detectors with Feature Richness
Zhixing Du, Rui Zhang, Ming Chang, Xishan Zhang, Shaoli Liu, Tianshi Chen, Yunji Chen
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 |
|---|---|---|---|
| Apply FRS distillation to RetinaNet ResNet-50 | RetinaNet ResNet-50 with ResNet-101 teacher · COCO | mAP: 37.4 -> 39.7 | Improved |
| Apply FRS distillation to GFL ResNet-50 with the 1x schedule | GFL ResNet-50 with ResNet-101 teacher · COCO | mAP: 40.2 -> 43.6 | Improved |
| Apply FRS distillation to GFL ResNet-50 with the 2x schedule | GFL ResNet-50 with ResNet-101 teacher · COCO | mAP: 42.9 -> 44.7 | Improved |
| Apply FRS distillation to Faster R-CNN ResNet-50 | Faster R-CNN ResNet-50 with ResNet-101 teacher · COCO | mAP: 37.4 -> 39.5 | Improved |
| Apply FRS distillation to FCOS ResNet-50 | FCOS ResNet-50 with ResNet-101 teacher · COCO | mAP: 38.5 -> 40.9 | Improved |
| Add FPN feature distillation (ablation) | RetinaNet ResNet-50 FRS · COCO | mAP: 37.4 -> 39.4 | Improved |
| Add classification-head distillation (ablation) | RetinaNet ResNet-50 FRS · COCO | mAP: 37.4 -> 38.4 | Improved |
| Add both FPN and classification-head distillation (ablation) | RetinaNet ResNet-50 FRS · COCO | mAP: 37.4 -> 39.7 | Improved |
| Add high-richness false-positive background regions to TP distillation (ablation) | RetinaNet ResNet-50 FPN distillation · COCO | mAP: 39.1 -> 39.4 | Improved |
| Add false-negative foreground regions to TP distillation (ablation) | RetinaNet ResNet-50 FPN distillation · COCO | mAP: 39.1 -> 39.0 | Hurt |
| Add true-negative background regions to TP distillation (ablation) | RetinaNet ResNet-50 FPN distillation · COCO | mAP: 39.1 -> 38.7 | Hurt |
Closest papers by meaning
- Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors (ICLR 2021 · similarity 0.91)
- Distilling Object Detectors with Global Knowledge (ECCV 2022 · similarity 0.91)
- Prediction-Guided Distillation for Dense Object Detection (ECCV 2022 · similarity 0.91)
- Focal and Global Knowledge Distillation for Detectors (CVPR 2022 · similarity 0.91)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.90)
- VkD: Improving Knowledge Distillation using Orthogonal Projections (CVPR 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.
