Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Imbalanced Attribute Classification using Visual Attention Aggregation
Nikolaos Sarafianos, Xiang Xu, Ioannis A. Kakadiaris
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 class weights to ResNet-101 baseline (ablation) | ResNet-101 · WIDER-Attribute | mAP: 83.7 -> 84.0 | Improved |
| add instance-level focal weighting on top of class weights (Lw) (ablation) | ResNet-101 · WIDER-Attribute | mAP: 84.0 -> 84.4 | Improved |
| add visual attention at a single scale (ablation) | ResNet-101 + Lw · WIDER-Attribute | mAP: 84.4 -> 85.0 | Improved |
| add attention loss La (ablation) | ResNet-101 + Lw + attention · WIDER-Attribute | mAP: 85.0 -> 85.7 | Improved |
| add multi-scale attention (without La) (ablation) | ResNet-101 + Lw + attention · WIDER-Attribute | mAP: 85.0 -> 85.9 | Improved |
| add multi-scale attention on top of La (ablation) | ResNet-101 + Lw + attention + La · WIDER-Attribute | mAP: 85.7 -> 86.4 | Improved |
| swap primary network from ResNet-50 to DenseNet-201 (ablation) | primary net only · WIDER-Attribute | mAP: 82.3 -> 84.5 | Improved |
| add weighted focal loss Lw in place of Lb | SRN* · PETA | F1: 83.25 -> 84.92 | Improved |
| add Lw to DenseNet-121 with class weights (ablation) | DenseNet-121 · PETA | F1: 82.1 -> 82.9 | Improved |
| add attention (ablation) | DenseNet-121 + Lw · PETA | F1: 82.9 -> 83.8 | Improved |
| add multi-scale with feature-level aggregation (ablation) | DenseNet-121 + attention · PETA | F1: 83.8 -> 84.1 | Improved |
| use score-level instead of feature-level multi-scale aggregation (ablation) | DenseNet-121 + attention · PETA | F1: 84.1 -> 84.7 | Improved |
Closest papers by meaning
- Compositional Attribute Imbalance in Vision Datasets (AAAI 2026 · similarity 0.90)
- Deep Imbalanced Learning for Face Recognition and Attribute Prediction (TPAMI 2019 · similarity 0.87)
- Class Rectification Hard Mining for Imbalanced Deep Learning (ICCV 2017 · similarity 0.87)
- Learning Deep Representation for Imbalanced Classification (CVPR 2016 · similarity 0.87)
- Learning From Human Attention for Attribute-Assisted Visual Recognition (TPAMI 2024 · similarity 0.87)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · 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.
