PBSCSR

PBSCSR数据集由华盛顿大学创建，专注于钢琴乐谱的作曲家风格识别。该数据集包含40,000个用于9类分类任务的62x64乐谱图像，100,000个用于100类分类任务的62x64乐谱图像，以及29,310个用于预训练的无标签可变长度乐谱图像。数据集通过固定长度的乐谱特征表示，从IMSLP上的钢琴乐谱图像中采样得到。PBSCSR旨在通过提供易于访问且具有挑战性的数据集，推动作曲家风格识别的研究，解决音乐信息检索中的关键问题。

The PBSCSR dataset was developed by the University of Washington, with a primary focus on composer style recognition for piano sheet music. It comprises three subsets: 40,000 62×64 sheet music images for a 9-class classification task, 100,000 62×64 sheet music images for a 100-class classification task, and 29,310 unlabeled variable-length sheet music images for pretraining. The dataset is sampled from piano sheet music images sourced from IMSLP, utilizing fixed-length sheet music feature representations. PBSCSR is designed to advance research on composer style recognition by providing an accessible and challenging benchmark dataset, addressing critical challenges in the field of music information retrieval.