import xml.etree.ElementTree as ET class CVATPreprocessor(): """Helper class to preprocess annotations in `CVAT for images 1.1` XML-encoded format""" @staticmethod def get_all_image_names(annotation_path): """Returns a list of all image names present in the annotation file""" annotations = ET.parse(annotation_path).getroot() images = annotations.findall("image") return [image.attrib["name"] for image in images] @staticmethod def get_all_image_polygons(image_name, annotation_path): """ Returns a dictionary of all polygons for the given image name. The key is the label and the value is a list of polygons (= each a list of points) associated with that label. """ annotations = ET.parse(annotation_path).getroot() image = annotations.find(f"image[@name='{image_name}']") raw_polygons = image.findall("polygon") # Extract the label and the raw points for each polygon, # parse the points to (x, y) and store each label-polygon pair in a list processed_polygons = {} for raw_polygon in raw_polygons: label, points = raw_polygon.attrib["label"], raw_polygon.attrib["points"].split(";") # Parse the points to (x, y) int pairs points = [(int(float(point.split(",")[0])), int(float(point.split(",")[1]))) for point in points] processed_polygons[label] = processed_polygons.get(label, []) + [points] return processed_polygons if __name__ == "__main__": # Example usage PATH_TO_ANNOTATIONS = "offline learning/semantic segmentation/data/annotations/" PATH_TO_IMAGES = "offline learning/semantic segmentation/data/frames/" CVAT_XML_FILENAME = "segmentation_annotation.xml" imgs = CVATPreprocessor.get_all_image_names(PATH_TO_ANNOTATIONS + CVAT_XML_FILENAME) polygons = CVATPreprocessor.get_all_image_polygons(imgs[0], PATH_TO_ANNOTATIONS + CVAT_XML_FILENAME) print(f"Loaded {len(imgs)} images from {PATH_TO_ANNOTATIONS + CVAT_XML_FILENAME}") print(f"Image '{imgs[0]} has {len(polygons)} polygon categories")