Export a polygon mask to PNG

Create a class-label mask from the polygon example. This Python exercise supports polygons only. For dataset conversion to training formats, see the Pro toolkit documentation.

Prepare the annotation#

Put dogs.json and dogs.jpg in the same folder. Install Pillow:

python -m pip install pillow

Export the mask#

Save this as export_annotated.py beside the sample files:

import json
from pathlib import Path

from PIL import Image, ImageDraw

annotation = json.loads(Path("dogs.json").read_text())
shapes = annotation["shapes"]
if any(shape["shape_type"] != "polygon" for shape in shapes):
    raise ValueError("This example exports polygons only.")

labels = sorted({shape["label"] for shape in shapes})
if len(labels) > 255:
    raise ValueError("This example supports at most 255 labels.")

mask = Image.new("L", (annotation["imageWidth"], annotation["imageHeight"]), 0)
draw = ImageDraw.Draw(mask)
for shape in shapes:
    label_id = labels.index(shape["label"]) + 1
    draw.polygon([tuple(point) for point in shape["points"]], fill=label_id)

output = Path("dogs_exported")
output.mkdir(exist_ok=True)
mask.save(output / "label.png")
(output / "label_names.json").write_text(json.dumps(["_background_", *labels]))
print("Created", output / "label.png")

Run it:

python export_annotated.py

You should get a dogs_exported folder containing label.png and label_names.json. Background pixels have value 0 and dog pixels have value 1. The mask will look almost black in an image viewer because 1 is close to 0; it is data, not a colored preview. Later polygons overwrite earlier ones where they overlap.

Running the script again overwrites those two files. Use a separate folder if you need to keep an earlier export.

Check the result#

Continue with Read an exported mask to inspect its pixel values and see a black-and-white preview. You can also download the example export.