Research map / Label-efficient and robust learning / Knowledge distillation
Distilling Object Detectors With Fine-Grained Feature Imitation
Check 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 fine-grained feature imitation | 0.5× toy detector · KITTI validation | mAP: 57.98 -> 60.46 | Helped |
| Add fine-grained feature imitation | 0.25× toy detector · KITTI validation | mAP: 45.96 -> 52.63 | Helped |
| Use full feature imitation instead of fine-grained regions (ablation) | 0.25× toy detector · KITTI validation | mAP: -8.9 | Hurt |
| Use scaled ground-truth boxes as imitation regions (ablation) | 0.25× toy detector · KITTI validation | mAP: +2.0 | Helped |
| Add vanilla knowledge-distillation loss (ablation) | 0.25× toy detector · KITTI validation | mAP: +0.9 | Helped |
| Combine vanilla distillation loss with fine-grained imitation loss (ablation) | 0.25× toy detector · KITTI validation | mAP: 0.25×-I +6.7; 0.25×-ID +5.2 | Hurt |
| Add fine-grained feature imitation | Halved Faster R-CNN res101h · Pascal VOC07 | mAP: 67.4 -> 71.2 (+3.8) | Helped |
| Add fine-grained feature imitation | VGG11 Faster R-CNN · Pascal VOC07 | mAP: 59.6 -> 67.6 (+8.0) | Helped |
| Add fine-grained feature imitation | ResNet-50 Faster R-CNN · Pascal VOC07 | mAP: 69.1 -> 72.0 (+2.9) | Helped |
| Add fine-grained feature imitation | Halved Faster R-CNN res101h · COCO | AP: 28.8 -> 31.6 (+2.8) | Helped |
| Add fine-grained feature imitation across FPN layers | Halved ResNet-50 FPN Faster R-CNN · COCO | AP: 31.2 -> 34.8 (+3.6) | Helped |
| Vary the imitation threshold ψ, using ψ=0 rather than a selective mask (ablation) | 0.5× and 0.25× toy detectors · KITTI validation | mAP | Hurt |
Closest papers by meaning
- Knowledge Distillation for Object Detection via Rank Mimicking and Prediction-Guided Feature Imitation (AAAI 2022 · similarity 0.93)
- Localization Distillation for Dense Object Detection (CVPR 2022 · similarity 0.92)
- CrossKD: Cross-Head Knowledge Distillation for Object Detection (CVPR 2024 · similarity 0.89)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.89)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · 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.