The CPMS dataset is a camera-based printed music score dataset under realistic scenarios created by us, which is set up by taking photos with mobile devices under different angles and light conditions. The test set is originate from the public repertoire of the 2020 sight-singing exam at the Wuhan Conservatory of Music in China(http://www.hbea.edu.cn/html/2019-09/12349.html), and we take pictures and label the printed msuic scores.

Our project adopts the following directory structure

CPMS_Dataset/
├── data/
|		├── IMG_1609.jpeg
|		├── IMG_1610.jpeg
|		└── ...
├── label/
|		├── IMG_1609.json
|		├── IMG_1610.json
|		└── ...

The data folder stores the pictures of the photographed music scores, and the label folder stores the annotation data. Each music score image corresponds to a json format label.

{
		"0": [
				{
            "x": 125,
            "y": 167,
            "pitch": "G4",
            "duration": 1
        },
        ...
		],
		"1": [
        {
            "x": 111,
            "y": 291,
            "pitch": "C5",
            "duration": 4
        },
        ...
    ],
    ...
}

Notes:

• The annotation file is in JSON format, each music score image corresponds to a json format file.

• The top level field 0、1indicates the row index of the staff, usually a staff has 10 lines.

• The fieldx、yindicate the coordinates of the upper left corner of the notehead bounding box.

• The field pitchindicates the pitch of the note in scientific pitch notation.

• The fielddurationindicates the index of duration classification of the note,and the corresponding relationship is as follows:

  index	0	1	2	3	4
  duration	whole	half	half_dot	quarter	quarter_dot
  index	5	6	7	8	9
  duration	eighth	eighth_dot	sixteenth	sixteenth_dot	thirty_second

​