ReAlign-Trainset
收藏资源简介:
ReAlign-Trainset 是一个用于视觉文档检索任务的训练数据集,基于论文《ReAlign: Optimizing the Visual Document Retriever with Reasoning-Guided Fine-Grained Alignment》中提出的方法。该方法通过利用视觉语言模型(VLMs)的推理能力,生成细粒度的视觉文档描述作为监督信号,以优化查询与视觉文档之间的语义对齐。数据集旨在支持通过推理引导的细粒度对齐来提升视觉文档检索的性能。
ReAlign-Trainset is a training dataset for visual document retrieval tasks, based on the method proposed in the paper *ReAlign: Optimizing the Visual Document Retriever with Reasoning-Guided Fine-Grained Alignment*. This method leverages the reasoning capabilities of Vision-Language Models (VLMs) to generate fine-grained visual document descriptions as supervision signals, thereby optimizing the semantic alignment between queries and visual documents. This dataset aims to support the improvement of visual document retrieval performance through reasoning-guided fine-grained alignment.
ReAlign-Trainset 数据集概述
数据集基本信息
- 数据集名称:ReAlign-Trainset
- 任务类别:视觉文档检索
数据集来源与背景
- 来源:该数据集为 ReAlign 方法的训练数据。
- 关联论文:ReAlign: Optimizing the Visual Document Retriever with Reasoning-Guided Fine-Grained Alignment
- 论文地址:https://huggingface.co/papers/2604.07419
- 方法简述:ReAlign 是一种通过利用视觉语言模型的推理能力来增强视觉文档检索的方法。它识别页面中与查询相关的区域,并生成查询感知的描述,以更好地对齐查询与视觉文档之间的语义。
相关资源链接
- 代码仓库:https://github.com/NEUIR/ReAlign
- 项目合集:https://huggingface.co/collections/yanghaoir/realign
引用信息
bibtex @article{yang2026realign, title={ReAlign: Optimizing the Visual Document Retriever with Reasoning-Guided Fine-Grained Alignment}, author={Yang, Hao and Ji, Yifan and Xu, Zhipeng and Liu, Zhenghao and Yan, Yukun and Chen, Zulong and Wang, Shuo and Gu, Yu and Yu, Ge}, year={2026}, url={https://arxiv.org/abs/2604.07419} }




