Research map / Vision-language and multimodal / Human and embodied modeling
Stacked Cross Attention for Image-Text Matching
Kuang-Huei Lee, Xi Chen, Gang Hua, Houdong Hu, Xiaodong He
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 stacked cross attention (SCAN t-i AVG vs Sum-Max t-i) (ablation) | Sum-Max t-i · Flickr30K | Sentence R@1: 59.6 -> 61.8 | Improved |
| add stacked cross attention (SCAN i-t AVG vs Sum-Max i-t) (ablation) | Sum-Max i-t · Flickr30K | Sentence R@1: 56.7 -> 67.9 | Improved |
| add stacked cross attention (SCAN i-t AVG vs Sum-Max i-t) (ablation) | Sum-Max i-t · Flickr30K | Image R@1: 36.8 -> 43.9 | Improved |
| infer latent region-word alignment (Sum-Max t-i vs VSE++ single embedding) (ablation) | VSE++ · Flickr30K | Sentence R@1: 31.9 -> 59.6 | Improved |
| add hard negatives in triplet loss (ablation) | SCAN i-t AVG · Flickr30K | Sentence R@1: 45.8 -> 67.9 | Improved |
| add hard negatives in triplet loss (ablation) | SCAN i-t AVG · Flickr30K | Image R@1: 33.9 -> 43.9 | Improved |
| normalize image embedding (ablation) | SCAN i-t AVG · Flickr30K | Sentence R@1: 67.8 -> 67.9 | No clear effect |
| use AVG pooling instead of SUM (ablation) | SCAN i-t · Flickr30K | Sentence R@1: 63.9 -> 67.9 | Improved |
| use AVG pooling instead of MAX (ablation) | SCAN i-t · Flickr30K | Sentence R@1: 59.7 -> 67.9 | Improved |
| use SUM pooling instead of AVG (ablation) | SCAN i-t · Flickr30K | Image R@1: 43.9 -> 45.0 | Mixed |
| use bi-directional GRU instead of one-directional (ablation) | SCAN i-t AVG · Flickr30K | Sentence R@1: 63.6 -> 67.9 | Improved |
| ensemble t-i AVG and i-t LSE models | SCAN single models · Flickr30K | Image R@1: 45.8 -> 48.6 | Improved |
Closest papers by meaning
- Fine-Grained Image-Text Matching by Cross-Modal Hard Aligning Network (CVPR 2023 · similarity 0.93)
- Text Embedding is Not All You Need: Attention Control for Text-to-Image Semantic Alignment with Text Self-Attention Maps (CVPR 2025 · similarity 0.90)
- Visual-Text Cross Alignment: Refining the Similarity Score in Vision-Language Models (ICML 2024 · similarity 0.89)
- GOAL: Global-local Object Alignment Learning (CVPR 2025 · similarity 0.89)
- Identification of Necessary Semantic Undertakers in the Causal View for Image-Text Matching (AAAI 2024 · similarity 0.89)
- Composing Object Relations and Attributes for Image-Text Matching (CVPR 2024 · similarity 0.88)
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.
