VisionScores
收藏资源简介:
VisionScores是一个系统分割的图像评分数据集,旨在为机器学习和深度学习任务提供结构丰富、高信息密度的图像。该数据集专注于双手钢琴作品,考虑了图形相似性和创作模式,因为这些创造性过程高度依赖于乐器。数据集提供了两种场景:第一种由14k个样本组成,考虑来自不同作者但相同创作类型的作品,即Sonatinas;第二种由10.8K个样本组成,呈现相反的情况,来自同一作者的多种创作类型,所选的作曲家是Franz Liszt。所有24.8k个样本都被格式化为128 × 512像素的灰度jpg图像。VisionScores不仅提供了格式化的样本,还提供了系统的顺序和作品的元数据。此外,还包含了未分割的全页评分和预格式化的图像,供进一步分析。VisionScores可在https://github.com/ alroamz/VisionScores免费获取。
VisionScores is a systematically segmented image scoring dataset developed to provide structurally rich, high-information-density images for machine learning and deep learning tasks. Focused on two-handed piano works, this dataset takes into account graphical similarity and creative patterns, as such creative processes are highly dependent on the specific musical instrument. The dataset offers two distinct scenarios: the first comprises 14k samples, consisting of works from different authors but belonging to the same creation type, i.e., Sonatinas; the second comprises 10.8k samples, presenting the reverse scenario: works from a single author but spanning multiple creation types, with the selected composer being Franz Liszt. All 24.8k samples are formatted as grayscale JPEG images with a resolution of 128 × 512 pixels. Beyond formatted samples, VisionScores also provides systematic sequence information and metadata for each work. Additionally, unsegmented full-page scores and pre-formatted images are included to support further analytical research. VisionScores is freely accessible at https://github.com/alroamz/VisionScores.




