Rudder

Rudder是一个多语言视频和文本检索数据集，由印度理工学院孟买分校创建。该数据集包含3272个视频，涵盖Marathi、Hindi、Tamil、Kannada、Malayalam和Telugu六种语言。数据集中的视频主要是科学玩具制作教程，每个视频配有多种语言的音频叙述和文本描述。Rudder数据集旨在解决多模态学习中的数据稀缺问题，通过引入部分排序损失来增强监督学习，提高视频与文本之间嵌入的密集覆盖。该数据集适用于跨语言视频-文本检索、多语言视频推荐和视频引导的机器翻译等应用。

Rudder is a multilingual video-text retrieval dataset created by the Indian Institute of Technology Bombay. This dataset contains 3272 videos covering six languages: Marathi, Hindi, Tamil, Kannada, Malayalam, and Telugu. The videos in the dataset are primarily scientific toy-making tutorials, with each video accompanied by audio narrations and textual descriptions in multiple languages. The Rudder dataset aims to address the data scarcity issue in multimodal learning, enhance supervised learning by introducing partial ranking loss, and improve the dense coverage of embeddings between videos and texts. This dataset is applicable to applications such as cross-lingual video-text retrieval, multilingual video recommendation, and video-guided machine translation.