String Quartet OMR

• Input: PDF + JSON file specifying where the time signature shifts were
• Output: MusicXML file

Prequisites

• Python 3.9
• Optional: NVidia GPU with CUDA 12.4

Set Up

• Clone the repository
• Install dependencies using pip install -r requirements.txt If having issues, try using pip install -r requirements-no-versions.txt

Test run

• Demo score:
  • run the pdf2musicXML.py, which will run on beethoven1's string quartet (found in string_dataset/pdf_data/beethoven1.pdf)
  • output can be found in string_dataset/output once the model finishes execution.
• Test your own score:
  • Add your pdf in pdf_data/[piece_name][piece_name].pdf (for instance, if your pdf is called mozart1, save it in pdf_data/mozart1/mozart1.pdf)
  • Add the specification of where the time signature shifts were to a new .json file, template can be found in jsonTemplate.json, save it in pdf_data/[piece_name][piece_name].json (for instance, if your pdf is called mozart1, save it in pdf_data/mozart1/mozart1.json)
  • Add the piece name in string_dataset/piecesToRun, for instance ['mozart1']
  • JSON file explained

      {
  "numPage": 20,  # how many pages are in the pdf
  "numPerPage": 1, # default 1 
  "rotate": false, # default false
  "tsChange": [
      # MUST have the fist one
      {
      "page": 1,
      "loc":[0,0],
      "time_signature": [4,4], # the inital time signature
      "mov": 1 # not used for now
      },
      # on page 7, the first line fisrt bar, 
      # time signature changed to 2/4
      {
      "page": 7,
      "loc":[0,0],
      "time_signature": [2,4],
      "mov":2
      },
      # on page 9, the 4th line first bar, 
      # time signature changed to 3/4
      {
      "page": 9,
      "loc":[3,0],
      "time_signature": [3,4],
      "mov":3
      },
      # ... etc.
      {
      "page": 11,
      "loc":[2,0],
      "time_signature": [12,8],
      "mov":3
      }
  ],
  "numTrack": 4, # how many instruments are there
  "track_shift": [0,0,0,0], # not used for now (for transposing instruments, not used in string quartet)
  "clef_options": [[1],[1],[0],[-1,-2]]
  # what are the possible clefs in that track
  # 1: treble, 0: alto (viola), -1: bass, -2: tenor (cello)
  # [[first track options],[2nd track options],[3rd ...], [4th...]]
  }

This project utilized the segmentation models from oemer