Research map / Detection and segmentation / Weakly supervised detection and segmentation
Revisiting Knowledge Transfer for Training Object Class Detectors
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 |
|---|---|---|---|
| Add class-generic entity scores from the source-trained detector during MIL re-localization | MIL on ILSVRC 2013 target classes · ILSVRC 2013 target train | CorLoc IoU > 0.5: 41.4 -> 70.3 | Helped |
| Add class-generic entity scores from the source-trained detector during MIL re-localization | MIL on ILSVRC 2013 target classes · ILSVRC 2013 target train | CorLoc IoU > 0.7: 20.0 -> 58.8 | Helped |
| Use closest-source scores instead of class-generic scores (ablation) | MIL knowledge transfer · ILSVRC 2013 target train | CorLoc IoU > 0.5: 41.4 -> 68.5; class-generic: 70.3 | Mixed |
| Use the closest common ancestor score instead of class-generic scores (ablation) | MIL knowledge transfer · ILSVRC 2013 target train | CorLoc IoU > 0.5: 41.4 -> 68.4; class-generic: 70.3 | Mixed |
| Use the closest ancestor with at least 5 source classes instead of class-generic scores (ablation) | MIL knowledge transfer · ILSVRC 2013 target train | CorLoc IoU > 0.5: 41.4 -> 68.6; class-generic: 70.3 | Mixed |
| Replace SSD proposals with EdgeBoxes proposals (ablation) | MIL with class-generic transfer · ILSVRC 2013 target train | CorLoc IoU > 0.5: 70.3 -> 50.5 | Hurt |
| Remove objectness scoring during re-localization (ablation) | MIL with EdgeBoxes proposals · ILSVRC 2013 target train | CorLoc IoU > 0.5: 39.4 -> 50.5 | Hurt |
| Select closest source classes by visual similarity rather than WordNet hierarchy (ablation) | Closest-source transfer · ILSVRC 2013 target train | CorLoc IoU > 0.5: 68.5 -> 68.0 | No clear effect |
| Choose the best source class using target bounding-box supervision (ablation) | Closest-source transfer · ILSVRC 2013 target train | CorLoc IoU > 0.5: 68.5 -> 69.6 | Helped |
| Add class-generic transfer instead of selecting even the best individual source class (ablation) | MIL knowledge transfer · ILSVRC 2013 target train | CorLoc IoU > 0.5: 69.6 -> 70.3 | Helped |
| Add class-generic entity scores during MIL re-localization | MIL detector training · ILSVRC 2013 target test | mAP IoU > 0.5: 23.2 -> 36.9 | Helped |
| Add class-generic entity scores during MIL re-localization | MIL detector training · ILSVRC 2013 target test | mAP IoU > 0.7: 7.0 -> 27.2 | Helped |
Closest papers by meaning
- Large Scale Semi-Supervised Object Detection Using Visual and Semantic Knowledge Transfer (CVPR 2016 · similarity 0.90)
- Weak-Shot Object Detection Through Mutual Knowledge Transfer (CVPR 2023 · similarity 0.90)
- Distilling Image Classifiers in Object Detectors (NeurIPS 2021 · similarity 0.90)
- Mixed Supervised Object Detection by Transferring Mask Prior and Semantic Similarity (NeurIPS 2021 · similarity 0.90)
- UniT: Unified Knowledge Transfer for Any-Shot Object Detection and Segmentation (CVPR 2021 · similarity 0.89)
- Cyclic Contrastive Knowledge Transfer for Open-Vocabulary Object Detection (ICLR 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.