Research map / Label-efficient and robust learning / Recognition and long-tail
An Unsupervised Sampling Approach for Image-Sentence Matching Using Document-level Structural Information
Zejun Li, Zhongyu Wei, Zhihao Fan, Haijun Shan, Xuanjing Huang
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 intra-document negative sampling and dropout sub-document positive sampling, with the fine-grained Transformer model | MulLink · MSCOCO | p@1: 95.0 -> 97.6 | Improved |
| Add intra-document negative sampling and dropout sub-document positive sampling, with the fine-grained Transformer model | MulLink · Story-DII | AUC: 82.9 -> 85.5 | Improved |
| Add intra-document negative sampling and dropout sub-document positive sampling, with the fine-grained Transformer model | MulLink · Story-SIS | AUC: 68.8 -> 70.2 | Improved |
| Add concepts to the Transformer (ablation) | Ours · SIS | p@1: 75.8 -> 77.2 | Improved |
| Add Transformer to the backbone (ablation) | Ours · SIS | AUC: 84.0 -> 85.5 | Improved |
| Add Transformer and concepts rather than neither (ablation) | Ours · SIS | p@1: 74.6 -> 77.2 | Improved |
| Add dropout sub-document objective (D) (ablation) | Ours with C+I objectives · SIS | AUC: 85.2 -> 85.5 | Improved |
| Add intra-document objective (I) (ablation) | Ours with C+D objectives · SIS | AUC: 85.4 -> 85.5 | Improved |
| Add cross-document objective (C) (ablation) | Ours with I+D objectives · SIS | AUC: 84.1 -> 85.5 | Improved |
| Add intra-document and dropout sub-document objectives (I+D) (ablation) | Ours with C objective · SIS | AUC: 85.0 -> 85.5 | Improved |
| Add only concept features without Transformer modeling (ablation) | Ours · SIS | AUC: 85.3 vs 85.5 | No clear effect |
| Replace Transformer with softmax pooling for objects and concepts (ablation) | Ours · SIS | AUC: 85.5 -> 85.1 | Hurt |
Closest papers by meaning
- Expanding the Scope of Negatives: Boosting Image-Text Matching with Negatives Distribution Guided Learning (AAAI 2025 · similarity 0.87)
- Learning Semantic Relationship Among Instances for Image-Text Matching (CVPR 2023 · similarity 0.87)
- Image and Sentence Matching via Semantic Concepts and Order Learning (TPAMI 2018 · similarity 0.87)
- Unpaired Image-Text Matching via Multimodal Aligned Conceptual Knowledge (TPAMI 2024 · similarity 0.86)
- Negative-Aware Attention Framework for Image-Text Matching (CVPR 2022 · similarity 0.86)
- Expressing Objects Just Like Words: Recurrent Visual Embedding for Image-Text Matching (AAAI 2020 · 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.
