Research map / Label-efficient and robust learning / Knowledge distillation
Localization Distillation for Object Detection
Zhaohui Zheng, Rongguang Ye, Qibin Hou, Dongwei Ren, Ping Wang, Wangmeng Zuo, Ming‐Ming Cheng
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 |
|---|---|---|---|
| add localization distillation (Main LD) (ablation) | GFocal R50 student · COCO val2017 (R101 teacher, R50 student) | AP: 40.1 -> 41.1 | Improved |
| add valuable localization region LD (ablation) | Main LD · COCO val2017 (R101 teacher, R50 student) | AP: 41.1 -> 41.8 | Improved |
| add Main KD (classification KD in main region) (ablation) | Main LD + VLR LD · COCO val2017 (R101 teacher, R50 student) | AP: 41.8 -> 42.1 | Improved |
| add VLR KD on top of Main KD (ablation) | LD + Main KD · COCO val2017 (R101 teacher, R50 student) | AP: 42.1 -> 42.0 | No clear effect |
| add VLR KD on top of Main KD (ablation) | LD + Main KD · VOC 07+12 | AP: 53.1 -> 53.7 | Mixed |
| raise LD temperature from 1 to 10 (ablation) | LD · COCO val2017 (R101 teacher, R50 student) | AP: 40.3 -> 41.1 | Improved |
| raise LD temperature from 10 to 20 (ablation) | LD · COCO val2017 (R101 teacher, R50 student) | AP: 41.1 -> 40.5 | Hurt |
| replace pseudo bbox regression with LD (ablation) | GFocal R50 · COCO val2017 (R101 teacher, R50 student) | AP: 40.5 -> 41.1 | Improved |
| lower VLR gamma from 1 to 0.25 (ablation) | LD · COCO val2017 (R101 teacher, R50 student) | AP: 41.1 -> 41.8 | Improved |
| remove Main KD from Ours + Fine-Grained (ablation) | Ours + Fine-Grained · COCO val2017 (R101 teacher, R50 student) | AP: 42.4 -> 42.1 | Hurt |
| combine feature imitation (GI) with LD | Ours · COCO val2017 (R101 teacher, R50 student) | AP: 42.1 -> 42.4 | Improved |
| apply LD to lightweight students | ResNet-18 student · COCO val2017 | AP: 35.8 -> 37.5 | Improved |
Closest papers by meaning
- Localization Distillation for Dense Object Detection (CVPR 2022 · similarity 0.97)
- Knowledge Distillation for Object Detection via Rank Mimicking and Prediction-Guided Feature Imitation (AAAI 2022 · similarity 0.92)
- CrossKD: Cross-Head Knowledge Distillation for Object Detection (CVPR 2024 · similarity 0.89)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.89)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.89)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 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.
