BosphorusSign22k

BosphorusSign22k是一个大规模公开的手语识别数据集，由博阿齐奇大学计算机工程系创建，专注于土耳其手语识别。该数据集包含744个独特的土耳其手语词汇，由六名本地手语者重复表演，总时长约19小时，视频分辨率为1080p，帧率为30fps。数据集的创建过程包括视觉审查和清理，以及重新定义标签方案，以更适合识别和生产研究。BosphorusSign22k旨在为计算机视觉、视频识别和深度学习研究社区提供新的基准，特别是在减少对手语翻译员依赖的日常生活中的应用。

BosphorusSign22k is a large-scale public sign language recognition dataset developed by the Department of Computer Engineering at Boğaziçi University, focusing on Turkish Sign Language recognition. It contains 744 unique Turkish Sign Language glosses, repeatedly performed by six native signers, with a total recording duration of approximately 19 hours. All videos are captured at 1080p resolution with a frame rate of 30 fps. The dataset creation process includes visual review and data cleaning, as well as a redefined labeling scheme tailored to better support recognition and production-related research. BosphorusSign22k aims to provide a new benchmark for the computer vision, video recognition, and deep learning research communities, particularly for real-world applications that reduce dependence on human sign language interpreters.