music-ocr-vectors
收藏资源简介:
Music-OCR-Vectors(在HuggingFace上显示为ScoreStroke)是一个开源免费的数据集,专为光学音乐识别、机器学习和计算机视觉研究设计。其核心目标是桥接手写音乐符号与机器可读格式之间的鸿沟。数据集包含手绘乐谱及其对应的多种数字表示形式:SVG矢量图形(用于精确视觉布局和笔画路径)、ABC记谱法(用于轻量级文本序列表示)和LilyPond文件(用于高质量专业数字排版)。该数据集适用于多种任务,包括训练模型进行手写音乐到可播放/可编辑数字格式的转换(OMR)、图像到文本的序列到序列模型训练(如图像到ABC记谱法)、基于图像和SVG的栅格到矢量转换模型训练,以及手写分析和生成式AI应用。数据集起源于热那亚大学学生Davide Brunori的想法和基础工作,采用MIT许可证,允许无限制地用于学术研究、个人项目和商业应用。
Music-OCR-Vectors (displayed as ScoreStroke on HuggingFace) is an open-source, free dataset designed for optical music recognition (OMR), machine learning, and computer vision research. It aims to bridge the gap between handwritten music notation and machine-readable formats. The core content consists of hand-drawn music scores and their corresponding digital representations: SVG vector graphics for precise visual layout and stroke paths, ABC notation for lightweight text sequence representation, and LilyPond (.ly) files for high-quality professional digital typesetting. The dataset is suitable for various tasks, such as training models for converting handwritten music to playable/editable digital formats (OMR), image-to-text sequence-to-sequence model training (e.g., image to ABC notation), raster-to-vector conversion model training based on images and SVGs, and handwriting analysis and generative AI applications. It originated from the idea and foundational work of Davide Brunori, a student at the University of Genoa, and is licensed under the MIT license, allowing unrestricted use for academic research, personal projects, and commercial applications.




