Research map / Label-efficient and robust learning / Knowledge distillation
IDa-Det: An Information Discrepancy-Aware Distillation for 1-Bit Detectors
Sheng Xu, Yanjing Li, Bohan Zeng, Teli Ma, Baochang Zhang, Xianbin Cao, Peng Gao, Jinhu Lü
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 IDa proposal selection (with entropy loss) over Hint (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 74.5 -> 76.9 | Improved |
| add IDa proposal selection over FGFI (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 75.0 -> 76.9 | Improved |
| add IDa distillation vs no distillation (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 74.0 -> 76.9 | Improved |
| use entropy loss instead of L2 with IDa (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 76.5 -> 76.9 | Improved |
| use entropy loss instead of L2 with Hint (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 74.1 -> 74.5 | Improved |
| use entropy loss instead of L2 with FGFI (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 74.7 -> 75.0 | Improved |
| use entropy loss instead of inner-product with IDa (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 74.8 -> 76.9 | Improved |
| use entropy loss instead of cosine similarity with IDa (ablation) | 1-bit BiRes18 Faster-RCNN · PASCAL VOC test2007 (component ablation) | mAP (%): 76.4 -> 76.9 | Improved |
| add IDa-Det distillation to 1-bit Faster-RCNN ResNet-18 | 1-bit Faster-RCNN ResNet-18 · COCO minival | mAP@[.5,.95] (%): 26.8 -> 29.3 | Improved |
| add IDa-Det distillation to 1-bit SSD VGG-16 | 1-bit SSD VGG-16 · PASCAL VOC test2007 | mAP (%): 69.5 -> 72.5 | Improved |
| IDa-Det instead of LWS-Det on 1-bit Faster-RCNN ResNet-18 | 1-bit Faster-RCNN ResNet-18 · PASCAL VOC test2007 | mAP (%): 73.2 -> 76.9 | Improved |
| 1-bit IDa-Det student versus real-valued Faster-RCNN ResNet-18 | 1-bit Faster-RCNN ResNet-18 · PASCAL VOC test2007 | mAP (%): 76.4 -> 76.9 | Improved |
Closest papers by meaning
- Learning Efficient Object Detection Models with Knowledge Distillation (NeurIPS 2017 · similarity 0.89)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.89)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.87)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.87)
- Learning 1-Bit Tiny Object Detector with Discriminative Feature Refinement (ICML 2024 · similarity 0.87)
- CrossKD: Cross-Head Knowledge Distillation for Object Detection (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.
