Research map / Detection and segmentation / Weakly supervised detection and segmentation
Exploring Bottom-Up and Top-Down Cues With Attentive Learning for Webly Supervised Object Detection
Zhonghua Wu, Qingyi Tao, Guosheng Lin, Jianfei Cai
Check 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 attentive classification loss to weight region classification by class-specific attention (ablation) | WebSOD · PASCAL VOC 2007, novel/base split 1, novel classes | mAP: 58.3 -> 60.5 | Improved |
| add attentive classification loss to weight region classification by class-specific attention (ablation) | WebSOD · PASCAL VOC 2007, novel/base split 2, novel classes | mAP: 52.2 -> 53.0 | Improved |
| add attentive classification loss to weight region classification by class-specific attention (ablation) | WebSOD · PASCAL VOC 2007, novel/base split 3, novel classes | mAP: 58.0 -> 58.9 | Improved |
| add residual feature refinement for domain adaptation (ablation) | WebSOD + ACL · PASCAL VOC 2007, novel/base split 1, novel classes | mAP: 60.5 -> 61.8 | Improved |
| add residual feature refinement for domain adaptation (ablation) | WebSOD + ACL · PASCAL VOC 2007, novel/base split 2, novel classes | mAP: 53.0 -> 54.0 | Improved |
| add residual feature refinement for domain adaptation (ablation) | WebSOD + ACL · PASCAL VOC 2007, novel/base split 3, novel classes | mAP: 58.9 -> 60.0 | Improved |
| add residual feature refinement for domain adaptation (ablation) | WebSOD + ACL · PASCAL VOC 2007, base classes | mAP: 69.0 -> 70.7; 71.2 -> 72.5; 67.3 -> 70.5 | Improved |
| fine-tune all feature-learning layers while fixing the web detector, instead of using RFR (ablation) | WebSOD + ACL · PASCAL VOC 2007, novel/base split 1, novel classes | mAP: 60.5 -> 58.2 | Hurt |
| fine-tune all feature-learning layers while fixing the web detector, instead of using RFR (ablation) | WebSOD + ACL · PASCAL VOC 2007, novel/base split 1, base classes | mAP: 69.0 -> 74.6 | Improved |
| use bottom-up regions from a base-class detector to train on web images | WebSOD · PASCAL VOC 2007, novel/base split 1, novel classes | mAP: 32.4 -> 58.3 | Improved |
| use WebSOD instead of ZAOD | Novel-class detection · PASCAL VOC 2007, novel/base split 1, novel classes | mAP: 32.4 -> 61.8 | Improved |
| use WebSOD instead of Uijlings et al. | ILSVRC13 val2, 100 novel classes · ILSVRC13 val2, 100 novel classes, IoU > 0.5 | mAP: 36.9 -> 37.1 | Improved |
Closest papers by meaning
- Noise-Aware Fully Webly Supervised Object Detection (CVPR 2020 · similarity 0.89)
- Improving Object Detection with Selective Self-Supervised Self-Training (ECCV 2020 · similarity 0.89)
- Zero-Annotation Object Detection with Web Knowledge Transfer (ECCV 2018 · similarity 0.87)
- W2N: Switching from Weak Supervision to Noisy Supervision for Object Detection (ECCV 2022 · similarity 0.86)
- UWSOD: Toward Fully-Supervised-Level Capacity Weakly Supervised Object Detection (NeurIPS 2020 · similarity 0.86)
- Search and Detect: Training-Free Long Tail Object Detection via Web-Image Retrieval (CVPR 2025 · similarity 0.85)
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.
