Research map / Label-efficient and robust learning / Recognition and long-tail
Visual Attention Consistency Under Image Transforms for Multi-Label Image Classification
Hao Guo, Kang Zheng, Xiaochuan Fan, Hongkai Yu, Song Wang
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 |
|---|---|---|---|
| Enforce attention consistency under flipping and scaling | ResNet101 · WIDER Attribute test | mAP: 84.8 -> 87.5 | Improved |
| Enforce attention consistency under flipping and scaling | ResNet101 · MS-COCO val | mAP: 74.9 -> 77.5 | Improved |
| Enforce attention consistency under flipping and scaling | ResNet50 · PA-100K test | F1-score: 85.86 -> 87.59 | Improved |
| Use flipping, rotation, or scaling as data augmentation only | ResNet50 · WIDER Attribute test | mAP: 83.4 -> 84.2 | No clear effect |
| Enforce attention consistency under translation (ablation) | ResNet50 · WIDER Attribute test | mAP: 83.4 -> 83.9 | No clear effect |
| Enforce attention consistency under rotation (ablation) | ResNet50 · WIDER Attribute test | mAP: 83.4 -> 85.0 | Improved |
| Enforce attention consistency under scaling (ablation) | ResNet50 · WIDER Attribute test | mAP: 83.4 -> 85.6 | Improved |
| Enforce attention consistency under horizontal flipping (ablation) | ResNet50 · WIDER Attribute test | mAP: 83.4 -> 86.3 | Improved |
| Enforce attention consistency under flipping and scaling instead of either transform alone (ablation) | ResNet50 · WIDER Attribute test | mAP: 86.3 -> 86.8 | Improved |
| Enforce attention consistency under translation (ablation) | ResNet101 · WIDER Attribute test | mAP: 84.8 -> 84.6 | Hurt |
| Enforce attention consistency under rotation (ablation) | ResNet101 · WIDER Attribute test | mAP: 84.8 -> 86.0 | Improved |
| Enforce attention consistency under flipping and scaling (ablation) | ResNet101 · WIDER Attribute test | mAP: 87.1 -> 87.5 | Improved |
Closest papers by meaning
- Multi-Label Image Recognition by Recurrently Discovering Attentional Regions (ICCV 2017 · similarity 0.87)
- Unsupervised Deep Metric Learning with Transformed Attention Consistency and Contrastive Clustering Loss (ECCV 2020 · similarity 0.87)
- Saliency-based Sequential Image Attention with Multiset Prediction (NeurIPS 2017 · similarity 0.86)
- A Shared Multi-Attention Framework for Multi-Label Zero-Shot Learning (CVPR 2020 · similarity 0.86)
- Sharpen Focus: Learning With Attention Separability and Consistency (ICCV 2019 · similarity 0.86)
- Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning (CVPR 2025 · similarity 0.84)
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.
