In the data enhancement of the target detection task dataset, the image is vertically flipped and the coordinates of the xml annotation file are adjusted

Requirements:
        In the data enhancement of the dataset, sometimes the operation of flipping the image vertically is needed. After the image is flipped vertically, the coordinates of the corresponding xml annotation file also need to be adjusted.
 
Solution:
        Use python+opencv+import xml.etree.ElementTree to flip the image vertically and adjust the coordinates of the xml annotation file. code show as below:

import cv2
import os
import glob
import xml.etree.ElementTree as et


def flip_images(source_dir):
    images_list = glob.glob(os.path.join(source_dir, "*.jpg"))
    index = 0
    for image_path in images_list:
        image = cv2.imread(image_path)
        flip_image = cv2.flip(image,0)
        cv2.imwrite(image_path.replace(".jpg", "_flip.jpg"), flip_image)
        tree_ = et.ElementTree()
        tree_.parse(image_path.replace(".jpg", ".xml"))

        root = et.Element("annotation")
        folder = et.SubElement(root, "folder")
        folder.text = "images"
        filename = et.SubElement(root, "filename")
        filename.text = tree_.find(".//filename").text.replace(".jpg", "_flip.jpg")
        path = et.SubElement(root, "path")
        path.text = "/home/mapgoo/test"
        source = et.SubElement(root, "source")
        database = et.SubElement(source, "database")
        database.text = "Unknown"
        size = et.SubElement(root, "size")
        width = et.SubElement(size, "width")
        width.text = tree_.find(".//width").text
        height = et.SubElement(size, "height")
        height.text = tree_.find(".//height").text
        depth = et.SubElement(size, "depth")
        depth.text = "3"
        segmented = et.SubElement(root, "segmented")
        segmented.text = "0"

        for bndbox in tree_.findall(".//object"):
            xmin = bndbox.find(".//xmin")
            ymin = bndbox.find(".//ymin")
            xmax = bndbox.find(".//xmax")
            ymax = bndbox.find(".//ymax")
            xmin_text = xmin.text
            ymin_text = ymin.text
            xmax_text = xmax.text
            ymax_text = ymax.text

            object_ = et.SubElement(root, "object")
            name = et.SubElement(object_, "name")
            name.text = bndbox.find("name").text
            pose = et.SubElement(object_, "pose")
            pose.text = "Unspecified"
            truncated = et.SubElement(object_, "truncated")
            truncated.text = "0"
            difficult = et.SubElement(object_, "difficult")
            difficult.text = "0"
            bndbox = et.SubElement(object_, "bndbox")
            xmin = et.SubElement(bndbox, "xmin")
            xmin.text = xmin_text
            ymin = et.SubElement(bndbox, "ymin")
            ymin.text = str(image.shape[0] - int(ymax_text))
            xmax = et.SubElement(bndbox, "xmax")
            xmax.text = xmax_text
            ymax = et.SubElement(bndbox, "ymax")
            ymax.text = str(image.shape[0] - int(ymin_text))

        tree = et.ElementTree(root)
        tree.write(image_path.replace(".jpg", "_flip.xml"), encoding="utf-8")
        print(image_path, index)
        index += 1


if __name__ == '__main__':
    source_dir = "/home/Desktop/test"
    flip_images(source_dir)

 To use the above code, you need to modify the folder path (source_dir) where the original image and annotation file are located. Pro test is available.