FGCrossNet

FGCrossNet是由北京大学计算机科学与技术研究所构建的细粒度跨媒体检索数据集，包含200种鸟类的细粒度子类别，涵盖图像、文本、视频和音频四种媒体类型。数据集通过整合CUB-200-2011和YouTube Birds等现有数据集，并从专业网站收集文本和音频数据构建而成。创建过程中，数据经过精心筛选和清洗，确保数据质量。该数据集旨在支持细粒度跨媒体检索研究，解决现有数据集在细粒度检索需求上的不足，特别是在区分外观相似的鸟类物种方面的挑战。

FGCrossNet is a fine-grained cross-modal retrieval dataset constructed by the Institute of Computer Science and Technology, Peking University. It encompasses fine-grained subcategories of 200 bird species, spanning four media modalities: images, text, videos, and audio. The dataset is developed by integrating existing datasets such as CUB-200-2011 and YouTube Birds, as well as collecting text and audio data from professional websites. Rigorous screening and cleaning procedures were implemented throughout the data creation process to ensure high data quality. This dataset aims to support research on fine-grained cross-modal retrieval, addressing the limitations of existing datasets in meeting fine-grained retrieval requirements, especially the challenge of distinguishing visually similar bird species.