Research map / Vision-language and multimodal / Cross-modal learning
ViSTA: Vision and Scene Text Aggregation for Cross-Modal Retrieval
Mengjun Cheng, Yipeng Sun, Longchao Wang, Xiongwei Zhu, Kun Yao, Jie Chen, Guoli Song, Junyu Han, Jingtuo Liu, Errui Ding, Jingdong Wang
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 |
|---|---|---|---|
| Aggregate scene text with image features using ViSTA (ablation) | ViSTA-S · CTC-1K | Image-to-text R@1: 47.0 -> 52.5 | Improved |
| Aggregate scene text with image features using ViSTA (ablation) | ViSTA-S · CTC-1K | Text-to-image R@1: 34.6 -> 36.7 | Improved |
| Add image-text contrastive loss alongside fusion-text loss (ablation) | ViSTA-S · CTC-1K | Image-to-text R@1: 46.6 -> 52.5 | Improved |
| Add image-text contrastive loss alongside fusion-text loss (ablation) | ViSTA-S · CTC-1K | Text-to-image R@1: 30.3 -> 36.7 | Improved |
| Use fusion-token aggregation rather than global, cross, or late fusion (ablation) | ViSTA-S · CTC-1K | Image-to-text R@1: 48.4, 50.5, 49.2 -> 52.5 | Improved |
| Use fusion-token aggregation rather than global, cross, or late fusion (ablation) | ViSTA-S · CTC-1K | Text-to-image R@1: 34.7, 31.1, 34.9 -> 36.7 | Improved |
| Use 4 fusion layers rather than 1 or 2 (ablation) | ViSTA-S · CTC-1K | Image-to-text R@1: 48.2, 52.2 -> 52.5 | Improved |
| Use 4 fusion layers rather than 1 or 2 (ablation) | ViSTA-S · CTC-1K | Text-to-image R@1: 35.6, 35.4 -> 36.7 | Mixed |
| Use ViSTA instead of STARNet | Scene-text-aware retrieval · CTC-1K | Image-to-text R@1: 44.1 -> 52.5 | Improved |
| Use ViSTA instead of STARNet | Scene-text-aware retrieval · CTC-1K | Text-to-image R@1: 31.5 -> 36.7 | Improved |
| Use ViSTA-L instead of ViLT-B | Scene-text-free retrieval · MS-COCO (5K), fine-tuned | Image-to-text R@1: 61.5 -> 68.9 | Improved |
| Use ViSTA-L instead of ViLT-B | Scene-text-free retrieval · MS-COCO (5K), fine-tuned | Text-to-image R@1: 42.7 -> 52.6 | Improved |
Closest papers by meaning
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Heterogeneous Prompt-Guided Entity Inferring and Distilling for Scene-Text Aware Cross-Modal Retrieval (AAAI 2025 · similarity 0.89)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.87)
- Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion (ICLR 2025 · similarity 0.87)
- Dynamic Adapter with Semantics Disentangling for Cross-lingual Cross-modal Retrieval (AAAI 2025 · similarity 0.87)
- CrossOver: 3D Scene Cross-Modal Alignment (CVPR 2025 · 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.
