Xml file in PascalVoc format from predictor

[unrue]

I'm running inference on some images after a training on dataset using COCO format, and writing the predicted box in output images:

for idx, image in enumerate(images):
   im = cv2.imread(os.path.join(inference_images_dir, image))
   outputs = predictor(im)
   v = Visualizer(im[:, :, ::-1], metadata=train_metadata, scale=1.2)
   out = v.draw_instance_predictions(outputs["instances"].to("cpu"))
   cv2.imwrite(os.path.join(inference_out_dir,image), out.get_image()[:, :, ::-1])

It works well but, is there a way to save also the xml file of the image in PascalVoc format in order to create a new database as input? Thanks.