def normalize_annotations(self, muscima_pp_directory: str,
output_directory: str) -> None:
destination_annotation_file = os.path.join(output_directory,
"annotations.csv")
muscima_pp_image_generator = MuscimaPlusPlusImageGenerator()
xml_file_paths = muscima_pp_image_generator.get_all_xml_file_paths(
muscima_pp_directory)
all_crop_objects = muscima_pp_image_generator.load_crop_objects_from_xml_files(
xml_file_paths)
data = []
for crop_object in tqdm(all_crop_objects, "Converting annotations"):
writer = re.search("W-..", crop_object.doc).group().lower()
page = int(re.search("N-..", crop_object.doc).group()[2:])
filename = "images/{0}_p{1:03d}.png".format(writer, page)
class_name = crop_object.clsname
top = crop_object.top
left = crop_object.left
bottom = crop_object.bottom
right = crop_object.right
data.append((filename, top, left, bottom, right, class_name))
all_annotations = pd.DataFrame(data=data,
columns=[
"path_to_image", "top", "left",
"bottom", "right", "class_name"
])
all_annotations.to_csv(destination_annotation_file,
index=False,
float_format="%.0f")
def create_statistics_for_full_images(glob_pattern_for_retrieving_muscima_images: str,
muscima_pp_raw_data_directory: str,
exported_absolute_dimensions_file_path: str,
exported_relative_dimensions_file_path: str):
image_paths = glob(glob_pattern_for_retrieving_muscima_images)
if os.path.exists(exported_absolute_dimensions_file_path):
os.remove(exported_absolute_dimensions_file_path)
if os.path.exists(exported_relative_dimensions_file_path):
os.remove(exported_relative_dimensions_file_path)
image_generator = MuscimaPlusPlusImageGenerator()
all_xml_files = image_generator.get_all_xml_file_paths(muscima_pp_raw_data_directory)
absolute_dimensions = []
relative_dimensions = []
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(xml_file)
doc = crop_objects[0].doc
result = re.match(r"CVC-MUSCIMA_W-(?P<writer>\d+)_N-(?P<page>\d+)_D-ideal", doc)
writer = result.group("writer")
page = "0" + result.group("page")
for image_path in image_paths:
image_path_result = re.match(r".*w-(?P<writer>\d+).*p(?P<page>\d+).png", image_path)
image_path_writer = image_path_result.group("writer")
image_path_page = image_path_result.group("page")
if image_path_writer == writer and image_path_page == page:
path = image_path
break
image = Image.open(path, "r") # type: Image.Image
image_width = image.width
image_height = image.height
for crop_object in crop_objects:
class_name = crop_object.clsname
top, left, bottom, right = crop_object.bounding_box
width = right - left
height = bottom - top
x_center = width / 2.0 + left
y_center = height / 2.0 + top
absolute_dimensions.append([class_name, left, right, top, bottom, x_center, y_center, width, height])
relative_dimensions.append([class_name, left / image_width, right / image_width,
top / image_height, bottom / image_height,
x_center / image_width, y_center / image_height,
width / image_width, height / image_height])
absolute_statistics = pandas.DataFrame(absolute_dimensions, columns=["class","xmin","xmax","ymin","ymax","x_c","y_c","width","height"])
absolute_statistics.to_csv(exported_absolute_dimensions_file_path, float_format="%.5f", index=False)
relative_statistics = pandas.DataFrame(relative_dimensions, columns=["class","xmin","xmax","ymin","ymax","x_c","y_c","width","height"])
relative_statistics.to_csv(exported_relative_dimensions_file_path, float_format="%.5f", index=False)
def load_all_muscima_annotations(muscima_pp_dataset_directory) -> Dict[str, List[CropObject]]:
"""
:param muscima_pp_dataset_directory:
:return: Returns a dictionary of annotations with the filename as key
"""
image_generator = MuscimaPlusPlusImageGenerator()
raw_data_directory = os.path.join(muscima_pp_dataset_directory, "v1.0", "data", "cropobjects_withstaff")
all_xml_files = [y for x in os.walk(raw_data_directory) for y in glob(os.path.join(x[0], '*.xml'))]
crop_object_annotations = {}
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(xml_file)
doc = crop_objects[0].doc
crop_object_annotations[doc] = crop_objects
return crop_object_annotations
def prepare_annotations(muscima_image_directory: str, output_path: str,
muscima_pp_raw_dataset_directory: str,
exported_annotations_file_path: str,
annotations_path: str):
image_paths = glob(muscima_image_directory)
os.makedirs(output_path, exist_ok=True)
image_generator = MuscimaPlusPlusImageGenerator()
raw_data_directory = os.path.join(muscima_pp_raw_dataset_directory, "v1.0",
"data", "cropobjects_manual")
all_xml_files = [
y for x in os.walk(raw_data_directory)
for y in glob(os.path.join(x[0], '*.xml'))
]
if os.path.exists(exported_annotations_file_path):
os.remove(exported_annotations_file_path)
shutil.rmtree(annotations_path, ignore_errors=True)
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(
xml_file)
doc = crop_objects[0].doc
result = re.match(
r"CVC-MUSCIMA_W-(?P<writer>\d+)_N-(?P<page>\d+)_D-ideal", doc)
writer = result.group("writer")
page = result.group("page")
image_path = None
for path in image_paths:
result = re.match(r".*(?P<writer>w-\d+).*(?P<page>p\d+).png", path)
if ("w-" + writer) == result.group("writer") and (
'p' + page.zfill(3)) == result.group("page"):
image_path = path
break
image = Image.open(image_path, "r") # type: Image.Image
image_width = image.width
image_height = image.height
output_file_path = os.path.join(
output_path, "w-{0}_p{1}.jpg".format(writer, page.zfill(3)))
image.save(output_file_path, "JPEG", quality=95)
create_annotations_in_pascal_voc_format(
annotations_path, os.path.basename(output_file_path), crop_objects,
image_width, image_height, 3)
def test_download_extract_and_render_all_symbols(self):
# Arrange
datasetDownloader = MuscimaPlusPlusDatasetDownloader()
# Act
datasetDownloader.download_and_extract_dataset("temp/muscima_pp_raw")
image_generator = MuscimaPlusPlusImageGenerator()
image_generator.extract_and_render_all_symbol_masks(
"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'))
]
expected_number_of_symbols = 91254
actual_number_of_symbols = len(all_image_files)
# Assert
self.assertEqual(expected_number_of_symbols, actual_number_of_symbols)
# Cleanup
os.remove(datasetDownloader.get_dataset_filename())
shutil.rmtree("temp")
def prepare_annotations(muscima_pp_dataset_directory: str,
exported_annotations_file_path: str,
annotations_path: str):
muscima_image_directory = os.path.join(muscima_pp_dataset_directory,
"v1.0", "data", "images", "*.png")
image_paths = glob(muscima_image_directory)
image_generator = MuscimaPlusPlusImageGenerator()
xml_annotations_directory = os.path.join(muscima_pp_dataset_directory,
"v1.0", "data",
"cropobjects_manual")
all_xml_files = [
y for x in os.walk(xml_annotations_directory)
for y in glob(os.path.join(x[0], '*.xml'))
]
if os.path.exists(exported_annotations_file_path):
os.remove(exported_annotations_file_path)
shutil.rmtree(annotations_path, ignore_errors=True)
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(
xml_file)
doc = crop_objects[0].doc
image_path = None
for path in image_paths:
if doc in path:
image_path = path
break
image = Image.open(image_path, "r") # type: Image.Image
image_width = image.width
image_height = image.height
create_annotations_in_pascal_voc_format_from_crop_objects(
annotations_path, os.path.basename(image_path), crop_objects,
image_width, image_height, 3)
def get_data(muscima_image_directory,
muscima_pp_raw_data_directory: str,
visualize=False) -> Tuple[list[dict], dict, dict]:
all_imgs = {}
classes_count = {}
class_mapping = {}
image_generator = MuscimaPlusPlusImageGenerator()
all_xml_files = image_generator.get_all_xml_file_paths(
muscima_pp_raw_data_directory)
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(
xml_file)
doc = crop_objects[0].doc
result = re.match(
r"CVC-MUSCIMA_W-(?P<writer>\d+)_N-(?P<page>\d+)_D-ideal", doc)
writer = result.group("writer")
page = result.group("page")
# image_path = os.path.join(image_directory, "w-{0}".format(writer), "p0{0}.png".format(page))
image_path = os.path.join(muscima_image_directory,
"CvcMuscima-Distortions", "ideal",
"w-{0}".format(writer), "image",
"p0{0}.png".format(page))
img = cv2.imread(image_path)
for crop_object in crop_objects:
class_name = crop_object.clsname
(top, left, bottom, right) = crop_object.bounding_box
x1, y1, x2, y2 = left, top, right, bottom
if class_name not in classes_count:
classes_count[class_name] = 1
else:
classes_count[class_name] += 1
if class_name not in class_mapping:
class_mapping[class_name] = len(class_mapping)
if image_path not in all_imgs:
all_imgs[image_path] = {}
(rows, cols) = img.shape[:2]
all_imgs[image_path]['filepath'] = image_path
all_imgs[image_path]['width'] = cols
all_imgs[image_path]['height'] = rows
all_imgs[image_path]['bboxes'] = []
if np.random.randint(0, 6) > 0:
all_imgs[image_path]['imageset'] = 'trainval'
else:
all_imgs[image_path]['imageset'] = 'test'
all_imgs[image_path]['bboxes'].append({
'class': class_name,
'x1': x1,
'x2': x2,
'y1': y1,
'y2': y2
})
if visualize:
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255))
if visualize:
cv2.imshow('img', img)
cv2.waitKey(0)
all_data = []
for key in all_imgs:
all_data.append(all_imgs[key])
return all_data, classes_count, class_mapping
def cut_images(muscima_image_directory: str,
staff_vertical_positions_file: str, output_path: str,
muscima_pp_raw_data_directory: str,
exported_annotations_file_path: str, annotations_path: str):
image_paths = [
y for x in os.walk(muscima_image_directory)
for y in glob(os.path.join(x[0], '*.png'))
]
os.makedirs(output_path, exist_ok=True)
image_generator = MuscimaPlusPlusImageGenerator()
all_xml_files = image_generator.get_all_xml_file_paths(
muscima_pp_raw_data_directory)
if os.path.exists(exported_annotations_file_path):
os.remove(exported_annotations_file_path)
shutil.rmtree(annotations_path, ignore_errors=True)
crop_object_annotations: List[Tuple[str, str, List[CropObject]]] = []
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(
xml_file)
doc = crop_objects[0].doc
result = re.match(
r"CVC-MUSCIMA_W-(?P<writer>\d+)_N-(?P<page>\d+)_D-ideal", doc)
writer = result.group("writer")
page = result.group("page")
crop_object_annotations.append(
('w-' + writer, 'p' + page.zfill(3), crop_objects))
with open(staff_vertical_positions_file) as f:
content = f.readlines()
# you may also want to remove whitespace characters like `\n` at the end of each line
annotations = [x.strip().split(';') for x in content]
images_to_cut: List[Tuple[str, str, str, str]] = []
for image_path in image_paths:
result = re.match(r".*(?P<writer>w-\d+).*(?P<page>p\d+).png",
image_path)
writer = result.group("writer")
page = result.group("page")
coordinates = None
for annotation in annotations:
if annotation[0] == writer and annotation[1] == page:
coordinates = annotation[2]
break
if coordinates is not None:
images_to_cut.append((image_path, writer, page, coordinates))
for image_to_cut in tqdm(images_to_cut, desc="Cutting images"):
path, writer, page, coordinates = image_to_cut
staff_line_pairs = coordinates.split(',')
image = Image.open(path, "r") # type: Image.Image
width = image.width
objects_appearing_in_image: List[CropObject] = None
for crop_object_annotation in crop_object_annotations:
if writer == crop_object_annotation[
0] and page == crop_object_annotation[1]:
objects_appearing_in_image = crop_object_annotation[2]
break
if objects_appearing_in_image is None:
# Image has annotated staff-lines, but does not have corresponding crop-object annotations, so skip it
continue
i = 1
for pair in staff_line_pairs:
y_top, y_bottom = map(int, pair.split(':'))
previous_width = 0
overlap = 100
for crop_width in range(500, 3501, 500):
if crop_width > width:
crop_width = width
image_crop_bounding_box = (previous_width, y_top, crop_width,
y_bottom)
image_crop_bounding_box_top_left_bottom_right = (
y_top, previous_width, y_bottom, crop_width)
previous_width = crop_width - overlap
file_name = "{0}_{1}_{2}.jpg".format(writer, page, i)
i += 1
objects_appearing_in_cropped_image = \
compute_objects_appearing_in_cropped_image(file_name,
image_crop_bounding_box_top_left_bottom_right,
objects_appearing_in_image)
cropped_image = image.crop(image_crop_bounding_box).convert(
'RGB')
create_annotations_in_plain_format(
exported_annotations_file_path,
objects_appearing_in_cropped_image)
create_annotations_in_pascal_voc_format(
annotations_path, file_name,
objects_appearing_in_cropped_image, cropped_image.width,
cropped_image.height, 3)
# draw_bounding_boxes(cropped_image, objects_appearing_in_cropped_image)
output_file = os.path.join(output_path, file_name)
cropped_image.save(output_file, "JPEG", quality=95)
#%% Open OMR
from omrdatasettools.downloaders.OpenOmrDatasetDownloader import OpenOmrDatasetDownloader
dataset_downloader = OpenOmrDatasetDownloader("./data/openomr")
dataset_downloader.download_and_extract_dataset()
#%% Capitan
'''
from omrdatasettools.downloaders.CapitanDatasetDownloader import CapitanDatasetDownloader
dataset_downloader = CapitanDatasetDownloader("./data/capitan")
dataset_downloader.download_and_extract_dataset()
'''
#%% MUSCIMA
from omrdatasettools.downloaders.MuscimaPlusPlusDatasetDownloader import MuscimaPlusPlusDatasetDownloader
dataset_downloader = MuscimaPlusPlusDatasetDownloader("./data/muscima")
dataset_downloader.download_and_extract_dataset()
from omrdatasettools.image_generators.MuscimaPlusPlusImageGenerator import MuscimaPlusPlusImageGenerator
imgen = MuscimaPlusPlusImageGenerator()
imgen.extract_and_render_all_symbol_masks(
raw_data_directory='./data/muscima',
destination_directory='./data/muscima/images')
#%% Robelo
from omrdatasettools.downloaders.RebeloMusicSymbolDataset1Downloader import RebeloMusicSymbolDataset1Downloader
dataset_downloader = RebeloMusicSymbolDataset1Downloader("./data/robelo")
dataset_downloader.download_and_extract_dataset()
def prepare_annotations(muscima_image_directory: str, output_path: str,
muscima_pp_raw_dataset_directory: str,
exported_annotations_file_path: str,
exported_class_mapping_path: str):
image_paths = glob(muscima_image_directory)
os.makedirs(output_path, exist_ok=True)
image_generator = MuscimaPlusPlusImageGenerator()
raw_data_directory = os.path.join(muscima_pp_raw_dataset_directory, "v1.0",
"data", "cropobjects_manual")
all_xml_files = [
y for x in os.walk(raw_data_directory)
for y in glob(os.path.join(x[0], '*.xml'))
]
if os.path.exists(exported_annotations_file_path):
os.remove(exported_annotations_file_path)
if os.path.exists(exported_class_mapping_path):
os.remove(exported_class_mapping_path)
classes = []
with open(exported_annotations_file_path, "w") as annotations_csv:
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(
xml_file)
doc = crop_objects[0].doc
result = re.match(
r"CVC-MUSCIMA_W-(?P<writer>\d+)_N-(?P<page>\d+)_D-ideal", doc)
writer = result.group("writer")
page = result.group("page")
image_path = None
for path in image_paths:
result = re.match(r".*(?P<writer>w-\d+).*(?P<page>p\d+).png",
path)
if ("w-" + writer) == result.group("writer") and (
'p' + page.zfill(3)) == result.group("page"):
image_path = path
break
image = Image.open(image_path, "r") # type: Image.Image
output_file_path = os.path.join(
output_path, "w-{0}_p{1}.png".format(writer, page.zfill(3)))
image.save(output_file_path)
for crop_object in crop_objects:
class_name = crop_object.clsname
classes.append(class_name)
ymin, xmin, ymax, xmax = crop_object.bounding_box
annotations_csv.write(
"muscima_pp_images/{0},{1},{2},{3},{4},{5}\n".format(
os.path.basename(output_file_path), xmin, ymin, xmax,
ymax, class_name))
unique_classes = list(set(classes))
unique_classes.sort()
with open(exported_class_mapping_path, "w") as mapping_csv:
for index, class_name in enumerate(unique_classes):
mapping_csv.write("{0},{1}\n".format(class_name, index))
def cut_images(muscima_image_directory: str, staff_vertical_positions_file: str, output_path: str,
muscima_pp_raw_data_directory: str, ):
image_paths = [y for x in os.walk(muscima_image_directory) for y in glob(os.path.join(x[0], '*.png'))]
os.makedirs(output_path, exist_ok=True)
with open(os.path.join(output_path, "Annotations.txt"), "w") as annotations_file:
image_generator = MuscimaPlusPlusImageGenerator()
all_xml_files = image_generator.get_all_xml_file_paths(muscima_pp_raw_data_directory)
crop_object_annotations: List[Tuple[str, str, List[CropObject]]] = []
for xml_file in tqdm(all_xml_files, desc='Parsing annotation files'):
crop_objects = image_generator.load_crop_objects_from_xml_file(xml_file)
doc = crop_objects[0].doc
result = re.match(r"CVC-MUSCIMA_W-(?P<writer>\d+)_N-(?P<page>\d+)_D-ideal", doc)
writer = result.group("writer")
page = result.group("page")
crop_object_annotations.append(('w-' + writer, 'p' + page.zfill(3), crop_objects))
with open(staff_vertical_positions_file) as f:
content = f.readlines()
# you may also want to remove whitespace characters like `\n` at the end of each line
annotations = [x.strip().split(';') for x in content]
images_to_cut: List[Tuple[str, str, str, str]] = []
for image_path in image_paths:
result = re.match(r".*(?P<writer>w-\d+).*(?P<page>p\d+).png", image_path)
writer = result.group("writer")
page = result.group("page")
coordinates = None
for annotation in annotations:
if annotation[0] == writer and annotation[1] == page:
coordinates = annotation[2]
break
if coordinates is not None:
images_to_cut.append((image_path, writer, page, coordinates))
for image_to_cut in tqdm(images_to_cut, desc="Cutting images"):
path, writer, page, coordinates = image_to_cut
staff_line_pairs = coordinates.split(',')
image = Image.open(path, "r")
width = image.width
crop_objects_of_image: List[CropObject] = None
for crop_object_annotation in crop_object_annotations:
if writer == crop_object_annotation[0] and page == crop_object_annotation[1]:
crop_objects_of_image = crop_object_annotation[2]
break
if crop_objects_of_image is None:
# Image has annotated staff-lines, but does not have corresponding crop-object annotations, so skip it
continue
i = 1
for pair in staff_line_pairs:
y_top, y_bottom = map(int, pair.split(':'))
previous_width = 0
overlap = 100
for crop_width in range(500, 3501, 500):
objects_of_cropped_image: List[Tuple[str, Tuple[int, int, int, int]]] = []
if crop_width > width:
crop_width = width
image_crop_bounding_box = (previous_width, y_top, crop_width, y_bottom)
file_name = "{0}_{1}_{2}.png".format(writer, page, i)
for crop_object in crop_objects_of_image:
if bounding_box_in(image_crop_bounding_box, crop_object.bounding_box):
top, left, bottom, right = crop_object.bounding_box
translated_bounding_box = (
top - y_top, left - previous_width, bottom - y_top, right - previous_width)
trans_top, trans_left, trans_bottom, trans_right = translated_bounding_box
objects_of_cropped_image.append((crop_object.clsname, translated_bounding_box))
annotations_file.write("{0},{1},{2},{3},{4},{5}\n".format(file_name,
trans_left,
trans_top,
trans_right,
trans_bottom,
crop_object.clsname))
cropped_image = image.crop(image_crop_bounding_box).convert('RGB')
# draw_bounding_boxes(cropped_image, objects_of_cropped_image)
output_file = os.path.join(output_path, file_name)
cropped_image.save(output_file)
previous_width = crop_width - overlap
i += 1
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--dataset_dir",
default="data/muscima_pp/v1.0/data/cropobjects_withstaff/")
parser.add_argument("--mapping_output_path",
default="mapping_all_classes.txt")
parser.add_argument("--remove_line_shaped_or_construct",
type=bool,
default=False)
args = parser.parse_args()
names = glob.glob(os.path.join(args.dataset_dir, "*.xml"))
data = {}
image_generator = MuscimaPlusPlusImageGenerator()
for name in tqdm(names,
desc="Reading all objects from MUSCIMA++ annotations"):
data[name] = image_generator.load_crop_objects_from_xml_file(name)
datas = []
for value in data.values():
for val in value:
datas.append(val)
c = Counter([x.clsname for x in datas])
ignored_classes = [
] # ["double_sharp", "numeral_2", "numeral_5", "numeral_6", "numeral_7", "numeral_8"]
line_shaped_or_construct = [