Research map / Detection and segmentation / Open-set and few-shot detection
Disentangle and Remerge: Interventional Knowledge Distillation for Few-Shot Object Detection from a Conditional Causal Perspective
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 |
|---|---|---|---|
| Apply D&R distillation during fine-tuning | DeFRCN · VOC Novel Set 2, 1-shot (30-run average) | AP50: 27.4 -> 30.7 | Helped |
| Apply D&R distillation during fine-tuning | DeFRCN · COCO novel classes, 1-shot (10-run average) | mAP: 4.8 -> 6.1 | Helped |
| Apply D&R distillation during fine-tuning | DeFRCN · COCO novel classes, 10-shot (10-run average) | mAP: 16.7 -> 16.4 | Hurt |
| Add FBD+ to FBD- + FCD (ablation) | D&R · COCO novel classes, 1-shot | mAP: 7.91 -> 8.29 | Helped |
| Add FBD- to FBD+ + FCD (ablation) | D&R · COCO novel classes, 1-shot | mAP: 8.25 -> 8.29 | Helped |
| Add FCD to FBD+ + FBD- (ablation) | D&R · COCO novel classes, 1-shot | mAP: 7.71 -> 8.29 | Helped |
| Add TDD to FBD+ + FBD- + FCD (ablation) | D&R · COCO novel classes, 1-shot | mAP: 8.29 -> 8.05 | Hurt |
| Replace D&R objective with vanilla KD (ablation) | DeFRCN with distillation · COCO novel classes, 1-shot | mAP: 8.29 -> 7.69 | Hurt |
| Add distillation rather than use no KD (ablation) | DeFRCN · COCO novel classes, 1-shot | mAP: 6.77 -> 7.69 | Helped |
| Set distillation temperature to 5 instead of 1 (ablation) | D&R on COCO · COCO novel classes, 1-shot | mAP: 7.00 -> 7.31 | Helped |
| Set KD loss weight to 5 instead of 20 (ablation) | D&R on COCO · COCO novel classes, 1-shot | mAP: 7.11 -> 7.69 | Helped |
| Increase shot count from 1 to 10 while using D&R | DeFRCN with D&R on COCO · COCO novel classes, 1-shot versus 10-shot | mAP gain over DeFRCN: 6.1 vs. 4.8; 16.4 vs. 16.7 | Mixed |
Closest papers by meaning
- Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors (ICLR 2021 · similarity 0.89)
- Instance-Conditional Knowledge Distillation for Object Detection (NeurIPS 2021 · similarity 0.88)
- Distilling Image Classifiers in Object Detectors (NeurIPS 2021 · similarity 0.88)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.86)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.86)
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.