musegroup/omr_benchmark

Muse OMR Benchmark是一个小型、干净的基准数据集，用于光学音乐识别（OMR，即从图像/PDF中识别音乐符号）。它包含1077对数据：一个符号音乐乐谱（地面实况）和一个经过数据增强的PDF渲染。所有基础作品均为公共领域。数据集旨在为社区提供一个实用、可重复、公开的基准。每个PDF都是从我们自己的公共领域乐谱目录生成的，然后经过增强以模拟真实世界的扫描效果，包括墨水污渍、划痕、皱褶或纹理纸张、旋转/倾斜以及其他视觉噪声。数据集以对的形式分发，典型字段包括：`id`（唯一样本ID）、`pdf_image`（增强的PDF文件）和`score`（用于评估的MuseScore Studio文件格式的符号参考）。数据集内容在CC0-1.0许可下发布（无限制；建议署名）。

Muse OMR Benchmark is a small, clean benchmark dataset for Optical Music Recognition (OMR — recognizing music notation from images/PDFs). It contains 1077 pairs: a symbolic music score (the ground truth) and a corresponding PDF rendering with data augmentation applied. All underlying works are Public Domain. The dataset aims to provide the community with a practical, reproducible, public benchmark. Each PDF is generated from our own catalog of PD scores and then augmented to simulate real-world scans, including ink blobs/stains, scratches/wear, crumpled or textured paper, rotation/skew, and other visual noise. The dataset is distributed as pairs, with typical fields including: `id` (unique sample id), `pdf_image` (augmented PDF file), and `score` (symbolic reference in MuseScore Studio file format for evaluation). The dataset content is released under CC0-1.0 (no restrictions; attribution appreciated).