def download_datasets(self, root_dir):
if not os.path.exists(root_dir):
os.mkdir(root_dir)
muscima_plus_plus_path = os.path.join(root_dir, "MuscimaPlusPlus_V2")
if not os.path.exists(muscima_plus_plus_path):
Downloader().download_and_extract_dataset(OmrDataset.MuscimaPlusPlus_V2, muscima_plus_plus_path)
else:
print("MuscimaPlusPlus_V2 already exists")
cvc_muscima_path = os.path.join(root_dir, "CVC_Muscima_Augmented")
if not os.path.exists(cvc_muscima_path):
Downloader().download_and_extract_dataset(OmrDataset.CvcMuscima_MultiConditionAligned, cvc_muscima_path)
else:
print("CVC_Muscima_Augmented already exists")
muscima_plus_plus_measure_annotations_path = os.path.join(root_dir, "MuscimaPlusPlus_Measure_Annotations")
if not os.path.exists(muscima_plus_plus_measure_annotations_path):
Downloader().download_and_extract_dataset(OmrDataset.MuscimaPlusPlus_MeasureAnnotations, muscima_plus_plus_measure_annotations_path)
else:
print("MuscimaPlusPlus_Measure_Annotations already exists")
measure_bounding_box_annotations_v2_path = os.path.join(root_dir, "Measure_Bounding_Box_Annotations_v2")
if not os.path.exists(measure_bounding_box_annotations_v2_path):
Downloader().download_and_extract_dataset(OmrDataset.MeasureBoundingBoxAnnotations_v2, measure_bounding_box_annotations_v2_path)
else:
print("Measure_Bounding_Box_Annotations_v2 already exists")
def download_dataset_and_verify_correct_extraction(
self: unittest.TestCase, destination_directory: str,
number_of_samples_in_the_dataset: int, target_file_extension: str,
dataset: OmrDataset):
# Arrange and Cleanup
if os.path.exists(dataset.get_dataset_filename()):
os.remove(dataset.get_dataset_filename())
shutil.rmtree(destination_directory, ignore_errors=True)
downloader = Downloader()
# Act
downloader.download_and_extract_dataset(dataset, destination_directory)
# Assert
all_files = glob(destination_directory + "/**/" +
target_file_extension,
recursive=True)
actual_number_of_files = len(all_files)
self.assertEqual(number_of_samples_in_the_dataset,
actual_number_of_files)
self.assertTrue(os.path.exists(dataset.get_dataset_filename()))
# Cleanup
os.remove(dataset.get_dataset_filename())
shutil.rmtree(destination_directory, ignore_errors=True)
if __name__ == '__main__':
unittest.main()
def test_download_extract_and_draw_bitmaps(self):
# Arrange
datasetDownloader = Downloader()
# Act
datasetDownloader.download_and_extract_dataset(OmrDataset.Homus_V2, "temp/homus_raw")
HomusImageGenerator.create_images("temp/homus_raw", "temp/homus_img", [3], 96, 192, 14)
all_image_files = [y for x in os.walk("temp/homus_img") for y in glob(os.path.join(x[0], '*.png'))]
actual_number_of_files = len(all_image_files)
# Assert
self.assertEqual(15200, actual_number_of_files)
# Cleanup
os.remove("HOMUS-2.0.zip")
shutil.rmtree("temp")
def test_download_and_prepare_dataset(self):
# Arrange
dataset_downloader = Downloader()
expected_number_of_images = 503
# Act
dataset_downloader.download_and_extract_dataset(OmrDataset.OpenOmr, "temp/open_omr_raw2")
image_generator = OpenOmrImagePreparer()
image_generator.prepare_dataset("temp/open_omr_raw2", "temp/open_omr_image2")
all_image_files = [y for x in os.walk("temp/open_omr_image2") for y in glob(os.path.join(x[0], '*.png'))]
actual_number_of_images = len(all_image_files)
# Assert
self.assertEqual(expected_number_of_images, actual_number_of_images)
# Cleanup
os.remove(OmrDataset.OpenOmr.get_dataset_filename())
shutil.rmtree("temp")
def test_download_extract_and_render_training_symbols(self, tmp_path):
# Arrange
dataset_downloader = Downloader()
expected_number_of_images = 44809
# Act
dataset_downloader.download_and_extract_dataset(
OmrDataset.MuscimaPlusPlus_V2, str(tmp_path / "muscima_pp_raw"))
image_generator = MuscimaPlusPlusImageGenerator2()
image_generator.extract_symbols_for_training(
str(tmp_path / "muscima_pp_raw"), str(tmp_path / "muscima_img"))
all_image_files = [
y for x in os.walk(tmp_path / "muscima_img")
for y in glob(os.path.join(x[0], '*.png'))
]
actual_number_of_images = len(all_image_files)
# Assert
assert expected_number_of_images == actual_number_of_images
def test_download_and_prepare_dataset(self, tmp_path):
# Arrange
dataset_downloader = Downloader()
expected_number_of_images = 503
# Act
dataset_downloader.download_and_extract_dataset(
OmrDataset.OpenOmr, str(tmp_path / "open_omr_raw2"))
image_generator = OpenOmrImagePreparer()
image_generator.prepare_dataset(str(tmp_path / "open_omr_raw2"),
str(tmp_path / "open_omr_image2"))
all_image_files = [
y for x in os.walk(tmp_path / "open_omr_image2")
for y in glob(os.path.join(x[0], '*.png'))
]
actual_number_of_images = len(all_image_files)
# Assert
assert expected_number_of_images == actual_number_of_images
def test_download_extract_and_crop_bitmaps(self):
# Arrange
dataset_downloader = Downloader()
# Act
dataset_downloader.download_and_extract_dataset(
OmrDataset.Audiveris, "temp/audiveris_omr_raw")
image_generator = AudiverisOmrImageGenerator()
image_generator.extract_symbols("temp/audiveris_omr_raw",
"temp/audiveris_omr_img")
all_image_files = [
y for x in os.walk("temp/audiveris_omr_img")
for y in glob(os.path.join(x[0], '*.png'))
]
actual_number_of_files = len(all_image_files)
# Assert
self.assertEqual(1056, actual_number_of_files)
# Cleanup
os.remove("AudiverisOmrDataset.zip")
shutil.rmtree("temp")
def test_download_extract_and_render_all_symbols(self):
# Arrange
datasetDownloader = Downloader()
# Act
datasetDownloader.download_and_extract_dataset(
OmrDataset.MuscimaPlusPlus_V2, "temp/muscima_pp")
image_generator = MuscimaPlusPlusSymbolImageGenerator()
image_generator.extract_and_render_all_symbol_masks(
"temp/muscima_pp", "temp/muscima_pp/symbols")
all_image_files = [
y for x in os.walk("temp/muscima_pp/symbols")
for y in glob(os.path.join(x[0], '*.png'))
]
expected_number_of_symbols = 102914
actual_number_of_symbols = len(all_image_files)
# Assert
self.assertEqual(expected_number_of_symbols, actual_number_of_symbols)
# Cleanup
os.remove(OmrDataset.MuscimaPlusPlus_V2.get_dataset_filename())
shutil.rmtree("temp")
def test_download_extract_and_render_training_symbols(self):
# Arrange
dataset_downloader = Downloader()
expected_number_of_images = 44809
# Act
dataset_downloader.download_and_extract_dataset(
OmrDataset.MuscimaPlusPlus_V2, "temp/muscima_pp_raw")
image_generator = MuscimaPlusPlusImageGenerator2()
image_generator.extract_symbols_for_training("temp/muscima_pp_raw",
"temp/muscima_img")
all_image_files = [
y for x in os.walk("temp/muscima_img")
for y in glob(os.path.join(x[0], '*.png'))
]
actual_number_of_images = len(all_image_files)
# Assert
self.assertEqual(expected_number_of_images, actual_number_of_images)
# Cleanup
os.remove(OmrDataset.MuscimaPlusPlus_V2.get_dataset_filename())
shutil.rmtree("temp")
def __download_and_extract_datasets(
self, datasets, width, height, use_fixed_canvas,
staff_line_spacing, staff_line_vertical_offsets,
stroke_thicknesses_for_generated_symbols,
random_position_on_canvas: bool):
dataset_downloader = Downloader()
if 'homus' in datasets:
raw_dataset_directory = os.path.join(self.dataset_directory,
"homus_raw")
dataset_downloader.download_and_extract_dataset(
OmrDataset.Homus_V2, raw_dataset_directory)
generated_image_width = width
generated_image_height = height
if not use_fixed_canvas:
# If we are not using a fixed canvas, remove those arguments to
# allow symbols being drawn at their original shapes
generated_image_width, generated_image_height = None, None
bounding_boxes = HomusImageGenerator.create_images(
raw_dataset_directory, self.image_dataset_directory,
stroke_thicknesses_for_generated_symbols,
generated_image_width, generated_image_height,
staff_line_spacing, staff_line_vertical_offsets,
random_position_on_canvas)
bounding_boxes_cache = os.path.join(self.dataset_directory,
"bounding_boxes.txt")
with open(bounding_boxes_cache, "wb") as cache:
pickle.dump(bounding_boxes, cache)
if 'rebelo1' in datasets:
dataset_downloader.download_and_extract_dataset(
OmrDataset.Rebelo1, self.image_dataset_directory)
if 'rebelo2' in datasets:
dataset_downloader.download_and_extract_dataset(
OmrDataset.Rebelo2, self.image_dataset_directory)
if 'printed' in datasets:
dataset_downloader.download_and_extract_dataset(
OmrDataset.Printed, self.image_dataset_directory)
if 'fornes' in datasets:
raw_dataset_directory = os.path.join(self.dataset_directory,
"fornes_raw")
dataset_downloader.download_and_extract_dataset(
OmrDataset.Fornes, raw_dataset_directory)
image_preparer = FornesMusicSymbolsImagePreparer()
image_preparer.prepare_dataset(raw_dataset_directory,
self.image_dataset_directory)
if 'audiveris' in datasets:
raw_dataset_directory = os.path.join(self.dataset_directory,
"audiveris_omr_raw")
intermediate_image_directory = os.path.join(
self.dataset_directory, "audiveris_omr_images")
dataset_downloader.download_and_extract_dataset(
OmrDataset.Audiveris, raw_dataset_directory)
image_generator = AudiverisOmrImageGenerator()
image_generator.extract_symbols(raw_dataset_directory,
intermediate_image_directory)
image_preparer = AudiverisOmrImageExtractor()
image_preparer.prepare_dataset(intermediate_image_directory,
self.image_dataset_directory)
if 'muscima_pp' in datasets:
raw_dataset_directory = os.path.join(self.dataset_directory,
"muscima_pp_raw")
dataset_downloader.download_and_extract_dataset(
OmrDataset.MuscimaPlusPlus_V2, raw_dataset_directory)
image_generator = MuscimaPlusPlusImageGenerator2()
image_generator.extract_symbols_for_training(
raw_dataset_directory, self.image_dataset_directory)
if 'openomr' in datasets:
raw_dataset_directory = os.path.join(self.dataset_directory,
"open_omr_raw")
dataset_downloader.download_and_extract_dataset(
OmrDataset.OpenOmr, raw_dataset_directory)
image_preparer = OpenOmrImagePreparer()
image_preparer.prepare_dataset(raw_dataset_directory,
self.image_dataset_directory)
import os
from omrdatasettools.Downloader import Downloader, OmrDataset
from image_color_inverter import ImageColorInverter
if __name__ == "__main__":
muscima_pp_dataset_directory = os.path.join("data", "muscima_pp")
muscima_image_directory = os.path.join(muscima_pp_dataset_directory, "v1.0", "data", "images")
downloader = Downloader()
downloader.download_and_extract_dataset(OmrDataset.MuscimaPlusPlus_V2, muscima_pp_dataset_directory)
inverter = ImageColorInverter()
# We would like to work with black-on-white images instead of white-on-black images
inverter.invert_images(muscima_image_directory, "*.png")
i for i in os.listdir(source_folder)
if i.endswith(".png") and i not in broken_symbols
]
for image in all_png_images:
shutil.copy(os.path.join(source_folder, image),
destination_folder)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--raw_dataset_directory",
type=str,
default="../data/fornes_raw",
help="The directory, where the raw Muscima++ dataset can be found")
parser.add_argument(
"--image_dataset_directory",
type=str,
default="../data/images",
help="The directory, where the images will be copied to")
flags, unparsed = parser.parse_known_args()
dataset_downloader = Downloader()
dataset_downloader.download_and_extract_dataset(
OmrDataset.Fornes, flags.raw_dataset_directory)
dataset_preparer = FornesMusicSymbolsImagePreparer()
dataset_preparer.prepare_dataset(flags.raw_dataset_directory,
flags.image_dataset_directory)
action='store',
default="../data/mungs",
help='The root directory that will contain the MuNG XMLs.')
parser.add_argument(
'-i',
'--image_root',
action='store',
default="../data/images",
help='The root directory that will contain the images of'
' scores that are represented by the MuNGs. The'
' image names must correspond to the MuNG file'
' names, up to the file type suffix.')
args = parser.parse_args()
mung_root_directory = args.mung_root
image_root_directory = args.image_root
os.makedirs(mung_root_directory, exist_ok=True)
os.makedirs(image_root_directory, exist_ok=True)
temporary_directory = "temp"
dataset_downloader = Downloader()
dataset_downloader.download_and_extract_dataset(
OmrDataset.MuscimaPlusPlus_V2, temporary_directory)
copy_tree(os.path.join(temporary_directory, "v2.0", "data", "annotations"),
mung_root_directory)
copy_tree(os.path.join(temporary_directory, "v2.0", "data", "images"),
image_root_directory)
shutil.rmtree(temporary_directory)