Convert ALTO-XML/PAGE-XML polygon coordinates segmentation to YOLOv8 polygons labels (not OBB)

[johnlockejrr]

I want to convert PAGE-XML polygon points segmentation like this: 208,277 208,297 854,291 1844,307 1849,286 1847,261 1742,255 1707,241 1706,241 1704,241 1702,241 1661,260 1660,260 1658,260 1589,241 1587,241 1586,241 1584,241 1551,253 1390,258 1357,258 1229,258 1226,256 1110,236 1108,236 709,241 618,233 616,233 208,250 208,277 into YOLOv8 polygons labels for training like this one: 0 0.5793838862559242 0.03832116788321168 0.5793838862559242 0.06204379562043796 0.9135071090047393 0.06204379562043796 0.9135071090047393 0.03832116788321168 0.5793838862559242 0.03832116788321168 (unrelated).
What I did so far:

def parse_polygon(polygon_str):
    """Parses the polygon points from a string into a list of tuples."""
    points = polygon_str.strip().split(" ")
    return [tuple(map(float, point.split(","))) for point in points]

def convert_points_to_yolo_format(points, image_width, image_height):
    """Converts points to YOLO format as normalized coordinates."""
    yolo_points = []
    for x, y in points:
        # Normalize the points by the image dimensions
        norm_x = x / image_width
        norm_y = y / image_height
        yolo_points.append(norm_x)
        yolo_points.append(norm_y)
    return yolo_points

def format_yolo_label(class_id, yolo_points):
    """Format the YOLO label with class ID and coordinates."""
    return f"{class_id} " + " ".join(map(str, yolo_points))

def main():
    # Example polygon string
    polygon_str = "208,277 208,297 854,291 1844,307 1849,286 1847,261 1742,255 1707,241 1706,241 1704,241 1702,241 1661,260 1660,260 1658,260 1589,241 1587,241 1586,241 1584,241 1551,253 1390,258 1357,258 1229,258 1226,256 1110,236 1108,236 709,241 618,233 616,233 208,250 208,277"

    # Image dimensions (for normalization, replace with your image dimensions)
    image_width = 2537  # Example width
    image_height = 2837  # Example height

    # Class ID (for YOLO, you can replace it based on your class)
    class_id = 0

    # Parse the polygon string into points
    points = parse_polygon(polygon_str)

    # Convert points to YOLO format
    yolo_points = convert_points_to_yolo_format(points, image_width, image_height)

    # Format the YOLO label
    yolo_label = format_yolo_label(class_id, yolo_points)

    print(yolo_label)

if __name__ == "__main__":
    main()

Output:

0 0.08198659834450138 0.09763835037010928 0.08198659834450138 0.1046880507578428 0.3366180528182893 0.10257314064152273 0.7268427276310603 0.10821290095170956 0.7288135593220338 0.10081071554458935 0.7280252266456445 0.09199859005992245 0.686637761135199 0.0898836799436024 0.6728419392983839 0.08494888967218893 0.6724477729601892 0.08494888967218893 0.6716594402837998 0.08494888967218893 0.6708711076074103 0.08494888967218893 0.6547102877414269 0.09164610504053577 0.6543161214032321 0.09164610504053577 0.6535277887268427 0.09164610504053577 0.6263303113914072 0.08494888967218893 0.6255419787150177 0.08494888967218893 0.625147812376823 0.08494888967218893 0.6243594797004336 0.08494888967218893 0.6113519905400079 0.08917870990482904 0.5478912100906582 0.09094113500176243 0.5348837209302325 0.09094113500176243 0.4844304296413086 0.09094113500176243 0.4832479306267245 0.09023616496298907 0.4375246353961372 0.08318646457525555 0.4367363027197477 0.08318646457525555 0.2794639337800552 0.08494888967218893 0.24359479700433584 0.08212900951709552 0.2428064643279464 0.08212900951709552 0.08198659834450138 0.08812125484666902 0.08198659834450138 0.09763835037010928

Seems much! Am I wrong?

[glenn-jocher]

Your approach to converting polygon coordinates to YOLO format looks correct. The output appears lengthy because it includes normalized coordinates for each point in the polygon. Ensure your image dimensions are accurate for normalization. If you need further assistance, please check the Ultralytics documentation for additional guidance.