Research map / Graph learning / GNN architectures
Efficient Relative Attribute Learning using Graph Neural Networks
Zihang Meng, Nagesh Adluru, Hyunwoo J. Kim, Glenn Fung, Vikas Singh
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 multi-task learning of all six attributes at once (ablation) | Ours · OSR (mean) | pairwise accuracy: 98.96 -> 99.20 | Improved |
| add multi-task learning of all 11 attributes at once (ablation) | Ours · PubFig (mean) | pairwise accuracy: 98.51 -> 98.55 | Improved |
| use group-of-images graph model with AlexNet instead of Souri et al. pairwise Siamese with VGG | Ours · OSR (mean) | pairwise accuracy: 97.77 -> 98.96 | Improved |
| use group-of-images graph model instead of Souri et al. pairwise Siamese | Ours · PubFig (mean) | pairwise accuracy: 94.52 -> 98.51 | Improved |
| use graph model with AlexNet instead of DeepPermNet with VGG, which needs ranked sequences | Ours · OSR (mean) | pairwise accuracy: 98.48 -> 98.96 | Improved |
| use graph model instead of DeepPermNet with ranked sequences | Ours · PubFig (mean) | pairwise accuracy: 98.14 -> 98.51 | Improved |
| use graph model instead of MG-CNN with group encoding for binary attributes | Ours · Clothing Attributes Dataset (total) | pairwise accuracy: 92.82 -> 92.39 | Hurt |
| use graph model instead of M-CNN without group information | Ours · Clothing Attributes Dataset (total) | pairwise accuracy: 91.70 -> 92.39 | Improved |
| use graph model instead of MG-CNN on the Appearance group | Ours · Clothing Attributes Dataset (Appearance) | pairwise accuracy: 91.93 -> 89.53 | Hurt |
| use graph model instead of MG-CNN on the Patterns group | Ours · Clothing Attributes Dataset (Patterns) | pairwise accuracy: 95.37 -> 96.81 | Improved |
| use graph model instead of Souri et al. when pair labels barely overlap | Ours · UT-Zappos50K-lexicon (mean) | pairwise accuracy: 89.31 -> 90.07 | Improved |
| use multi-task graph model instead of single-attribute model on PubFig eyes (ablation) | Ours (multi-task) · PubFig (eyes) | pairwise accuracy: 98.62 -> 97.32 | Hurt |
Closest papers by meaning
- Learning Attributes Equals Multi-Source Domain Generalization (CVPR 2016 · similarity 0.87)
- Semi-supervised Learning with a Teacher-student Network for Generalized Attribute Prediction (ECCV 2020 · similarity 0.87)
- Thinking Outside the Pool: Active Training Image Creation for Relative Attributes (CVPR 2019 · similarity 0.87)
- Attribute Prototype Network for Any-Shot Learning (IJCV 2022 · similarity 0.85)
- Effectively Leveraging Attributes for Visual Similarity (ICCV 2021 · similarity 0.85)
- Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent (NeurIPS 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.
