zzsi/deep-scores-v2

---
license: cc-by-4.0
task_categories:
- object-detection
language:
- en
tags:
- music
- optical-music-recognition
- omr
- sheet-music
- symbol-detection
size_categories:
- 100K<n<1M
---

# DeepScoresV2 — Complete

A HuggingFace-formatted mirror of the **complete** version of the
[DeepScoresV2](https://zenodo.org/records/4012193) dataset for music object detection.

## Dataset description

DeepScoresV2 is a large-scale dataset of synthetically rendered music score pages
annotated with bounding boxes for musical symbols. The **complete** version contains
**255,385 images** with **151 million annotated instances** across 135 symbol classes.

Each image is a full score page rendered from MuseScore across 5 music fonts
(beethoven, emmentaler, gonville, gutenberg1939, lilyjazz). Annotations use
COCO format: `bbox` is `[x, y, width, height]` in pixel coordinates.

## Format

```python
{
"image_id": int,
"file_name": str,
"font": str, # music font used to render
"image": PIL.Image,
"width": int,
"height": int,
"annotation_set": str, # "deepscores" or "muscima++"
"objects": {
"id": List[int],
"bbox": List[List[float]], # [x, y, w, h], COCO format
"category_id": List[int],
"category": List[str],
"area": List[float],
},
}
```

## Usage

```python
from datasets import load_dataset

ds = load_dataset("zzsi/deep-scores-v2")
example = ds["train"][0]
print(example["objects"]["category"][:5])
example["image"].show()
```

## License

[Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/)

## Attribution

This dataset is a reformatted mirror of DeepScoresV2. Please cite the original work:

```bibtex
@inproceedings{DeepScoresV2,
title = {DeepScoresV2: A Dataset for Music Object Detection with a Challenging Test Set},
author = {Tuggener, Lukas and Satyawan, Yvan Putra and Pacha, Alexander
and Schmidhuber, J{\"u}rgen and Stadelmann, Thilo},
booktitle = {British Machine Vision Conference (BMVC)},
year = {2021}
}
```

Original dataset: <https://zenodo.org/records/4012193>
Original authors: Lukas Tuggener, Yvan Putra Satyawan, Alexander Pacha,
Jürgen Schmidhuber, Thilo Stadelmann (ZHAW / IDSIA)

license: CC BY 4.0
task_categories:
- 目标检测（object-detection）
language:
- 英语
tags:
- 音乐
- 光学音乐识别（optical-music-recognition）
- OMR
- 乐谱（sheet-music）
- 符号检测（symbol-detection）
size_categories:
- 10万<样本数<100万
---

# DeepScoresV2 — 完整版

本数据集为音乐目标检测专用数据集DeepScoresV2完整版的HuggingFace格式镜像副本，原始数据集链接为<https://zenodo.org/records/4012193>。

## 数据集说明

DeepScoresV2是一个大规模合成渲染乐谱页面数据集，针对音乐符号标注了边界框。其完整版包含255,385张图像，涵盖135个符号类别，总计1.51亿个标注实例。

所有图像均为基于MuseScore、使用5种音乐字体（beethoven、emmentaler、gonville、gutenberg1939、lilyjazz）渲染得到的完整乐谱页面。标注采用COCO格式：`bbox`为像素坐标下的`[x, y, 宽度, 高度]`。

## 数据格式

python
{
"image_id": int, // 图像ID
"file_name": str, // 文件名
"font": str, // 渲染所用音乐字体
"image": PIL.Image, // PIL图像对象
"width": int, // 图像宽度
"height": int, // 图像高度
"annotation_set": str, // 标注集，可选"deepscores"或"muscima++"
"objects": {
"id": List[int], // 实例ID列表
"bbox": List[List[float]], // COCO格式像素坐标框[x, y, 宽度, 高度]
"category_id": List[int], // 类别ID列表
"category": List[str], // 类别名称列表
"area": List[float], // 实例面积列表
},
}

## 使用方法

python
from datasets import load_dataset

ds = load_dataset("zzsi/deep-scores-v2")
example = ds["train"][0]
print(example["objects"]["category"][:5])
example["image"].show()

## 许可证

[知识共享署名4.0国际许可协议（CC BY 4.0）](https://creativecommons.org/licenses/by/4.0/)

## 引用声明

本数据集为DeepScoresV2的格式重构镜像副本，请引用原始文献：

bibtex
@inproceedings{DeepScoresV2,
title = {DeepScoresV2: A Dataset for Music Object Detection with a Challenging Test Set},
author = {Tuggener, Lukas and Satyawan, Yvan Putra and Pacha, Alexander
and Schmidhuber, J{"u}rgen and Stadelmann, Thilo},
booktitle = {British Machine Vision Conference (BMVC)},
year = {2021}
}

原始数据集链接：<https://zenodo.org/records/4012193>
原始作者：Lukas Tuggener、Yvan Putra Satyawan、Alexander Pacha、Jürgen Schmidhuber、Thilo Stadelmann（ZHAW / IDSIA）