Research map / Detection and segmentation / Open-set and few-shot detection
Learning to Infer Unseen Single-/ Multi-Attribute-Object Compositions With Graph Networks
Hui Chen, Jingjing Jiang, Nanning Zheng
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 |
|---|---|---|---|
| use graph-based attribute-object model instead of CGE | unseen composition recognition · MIT-States (test) | Test AUC@1: 5.1 -> 5.5 | Improved |
| use proposed model instead of CGE on UT-Zappos50K | composition recognition · UT-Zappos50K (test) | Test AUC@1: 26.4 -> 33.5 | Improved |
| use proposed model instead of CGE on MAD | multi-attribute recognition · MAD (test) | Test AUC@1: 18.6 -> 21.4 | Improved |
| add decoding loss L_de (ablation) | full model · UT-Zappos50K (test) | Test AUC@1: 27.5 -> 33.5 | Improved |
| add balance loss L_bal (ablation) | full model · UT-Zappos50K (test) | Test AUC@1: 19.5 -> 33.5 | Improved |
| add attribute-object composer module (ablation) | full model · MAD (test) | Test AUC@1: 15.4 -> 21.4 | Improved |
| use contrastive fusion loss instead of L1 loss (ablation) | fusion loss · MIT-States (test) | Test AUC@1: 1.6 -> 5.5 | Improved |
| use proposed balance loss instead of alternative L_bal' (ablation) | balance loss · MIT-States (test) | Test AUC@1: 2.8 -> 5.5 | Improved |
| add graph message passing with trainable sparse random graph (ablation) | linguistic encoder · UT-Zappos50K (test) | Test AUC@1: 24.4 -> 33.5 | Improved |
| increase graph depth from 2 to 4 layers (ablation) | graph network · UT-Zappos50K (test) | Test AUC@1: 33.5 -> 28.6 | Hurt |
| use deeper visual backbone ResNet-101 instead of ResNet-18 (ablation) | visual encoder · MIT-States (test) | Test AUC@1: 5.5 -> 6.8 | Improved |
| use one-hot labels instead of GloVe (ablation) | linguistic input · MIT-States (test) | Test AUC@1: 5.5 -> 5.0 | Hurt |
Closest papers by meaning
- Adversarial Fine-Grained Composition Learning for Unseen Attribute-Object Recognition (ICCV 2019 · similarity 0.89)
- Automatic Discovery, Association Estimation and Learning of Semantic Attributes for a Thousand Categories (CVPR 2017 · similarity 0.88)
- Recognizing Unseen Attribute-Object Pair with Generative Model (AAAI 2019 · similarity 0.88)
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.85)
- Compositional Caching for Training-free Open-vocabulary Attribute Detection (CVPR 2025 · similarity 0.85)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · 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.
