EUFCC-CIR
收藏资源简介:
EUFCC-CIR数据集是为画廊、图书馆、档案馆和博物馆(GLAM)收藏中的组合图像检索(CIR)任务设计的,基于EUFCC-340K数据集构建,包含超过180,000个注释的CIR三元组。每个三元组由一个多模态查询(输入图像加上描述所需属性操作的简短文本)和一组相关的目标图像组成。数据集通过自动化过程和过滤启发式方法创建,确保了查询-响应对的丰富性和多样性。该数据集旨在填补数字人文领域中CIR特定资源的空白,通过定性和定量分析展示了其在数字人文研究中的价值,旨在增强用户与文化遗产收藏互动的方式,提供更深入的见解和增强的可访问性。
The EUFCC-CIR dataset is designed for the combined image retrieval (CIR) task within collections of galleries, libraries, archives, and museums (GLAM). Built upon the EUFCC-340K dataset, it contains over 180,000 annotated CIR triples. Each triple comprises a multimodal query (an input image paired with a short text describing the required attribute manipulations) and a set of relevant target images. The dataset is constructed via automated processes and filtering heuristics, ensuring the richness and diversity of query-response pairs. It aims to fill the gap in CIR-specific resources for the Digital Humanities field, demonstrates its value in Digital Humanities research through qualitative and quantitative analyses, and is intended to enhance the ways users interact with cultural heritage collections, providing deeper insights and enhanced accessibility.

- 1EUFCC-CIR: a Composed Image Retrieval Dataset for GLAM Collections计算机视觉中心,巴塞罗那自治大学 · 2024年



