Research map / Vision-language and multimodal / Human and embodied modeling
Negative-Aware Attention Framework for Image-Text Matching
Kun Zhang, Zhendong Mao, Quan Wang, Yongdong Zhang
Check 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 negative-aware attention with positive and negative branches | NAAF · Flickr30K test | Image-to-text R@1: 77.8 -> 81.9 | Improved |
| use negative-aware attention with positive and negative branches | NAAF · MS-COCO 1K test | Image-to-text R@1: 79.6 -> 80.5 | Improved |
| use negative-aware attention with positive and negative branches | NAAF · MS-COCO 5K test | Image-to-text R@1: 53.7 -> 58.9 | Improved |
| add the negative attention branch to capture mismatched-fragment effects (ablation) | NAAF · Flickr30K | Image-to-text R@1: 66.2 -> 81.9 | Improved |
| add the designed positive and negative attention masks (ablation) | NAAF · Flickr30K | Image-to-text R@1: 74.2 -> 81.9 | Improved |
| add the image-to-text matching direction (ablation) | NAAF · Flickr30K | Image-to-text R@1: 81.9 -> 78.7 | Hurt |
| use GloVe word embeddings (ablation) | NAAF · Flickr30K | Image-to-text R@1: 75.9 -> 81.9 | Improved |
| add intra-modal word relationships (ablation) | NAAF · Flickr30K | Image-to-text R@1: 79.3 -> 81.9 | Improved |
| set initial mismatch penalty α to 3.0 instead of 1.0 (ablation) | NAAF · Flickr30K | Image-to-text R@1: 78.0 -> 78.9 | Improved |
| adjust α to α* during training (ablation) | NAAF · Flickr30K | Image-to-text R@1: 78.7 -> 79.6 | Improved |
| increase batch size from 32 to 128 (ablation) | NAAF · Flickr30K | Image-to-text R@1: 76.0 -> 79.6 | Improved |
| update negative samples during training (ablation) | NAAF · Flickr30K | Image-to-text R@1: 76.3 -> 79.6 | Improved |
Closest papers by meaning
- Expanding the Scope of Negatives: Boosting Image-Text Matching with Negatives Distribution Guided Learning (AAAI 2025 · similarity 0.92)
- Stacked Cross Attention for Image-Text Matching (ECCV 2018 · similarity 0.90)
- Discovering Clone Negatives via Adaptive Contrastive Learning for Image-Text Matching (ICLR 2025 · similarity 0.90)
- Identification of Necessary Semantic Undertakers in the Causal View for Image-Text Matching (AAAI 2024 · similarity 0.88)
- Aligning Composed Query with Image via Discriminative Perception from Negative Correspondences (AAAI 2025 · similarity 0.88)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (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.
