Used the key ideas from Transcoda: Synthetic datasets and better data augmentations

https://arxiv.org/pdf/2605.10835

Author: liebharc

homr/main.py

@@ -2,6 +2,7 @@
 import glob
 import os
 import sys
+import xml.etree.ElementTree as ET
 from concurrent.futures import Future
 from dataclasses import dataclass
 from enum import Enum
@@ -209,7 +210,7 @@ def process_image(
 
         eprint("Writing XML", result_staffs)
         xml = generate_xml(xml_generator_args, result_staffs, title)
-        xml.write(xml_file)
+        ET.ElementTree(xml).write(xml_file, encoding="unicode", xml_declaration=True)
 
         eprint("Finished parsing " + str(len(result_staffs)) + " staves")
         teaser_file = replace_extension(image_path, "_teaser.png")

homr/music_xml_generator.py

@@ -35,8 +35,10 @@ pyclean = "^3.5.0"
 coverage = "^7.13.4"
 augly = "^1.0.0"
 editdistance = "^0.8.1"
+augraphy = "^8.2.6"
 transformers = {extras = ["torch"], version = "^4.53.2"}
 cairosvg = "^2.8.2"
+verovio = "^6.2.1"
 types-requests = "^2.32.4.20260107"
 segmentation-models-pytorch = "^0.5.0"
 pytorch-lightning = "^2.6.1"
@@ -49,6 +51,7 @@ albumentations = "^2.0.8"
 tensorboard = "^2.20.0"
 onnxscript = "^0.6.2"
 music21 = "^10.3.0"
+pyvips = "^3.1.1"
 musicdiff = "^5.2"
 datasets = "^5.0.0"
 
@@ -105,7 +108,9 @@ ignore = [
     "PLR0912",  # the parsing & conversion logic tends to create large functions, if someone has an idea then open a PR please!
     "PLR0915",  # the parsing & conversion logic tends to create large functions, if someone has an idea then open a PR please!
     "C901",  # the parsing & conversion logic tends to create large functions , if someone has an idea then open a PR please!
-    "PLR2004"  # 
+    "PLR2004",  #
+    "PLR0911",  # the parsing & conversion logic tends to create large functions, if someone has an idea then open a PR please!
+
 ]
 
 [tool.ruff.lint.flake8-tidy-imports]