jgdildsengafiller.py

from backend import SengaFiller
from numpy import ndarray, cast, int8, ones
from PIL import Image


class JGDILDSengaFiller(SengaFiller):
    def __init__(self) -> None:
        super().__init__()
        from tensorflow import keras

        self.model = keras.models.load_model("./model1.h5")

    def run(self, image_pil: Image.Image) -> ndarray:
        input_width, input_height = image_pil.size
        image_mono_pil = image_pil.point(lambda x: int(x > 200), mode="L")
        image_numpy = ones((input_height, input_width)) * image_mono_pil
        x = image_numpy.reshape((1, input_height, input_width, 1))
        y = self.model(x)
        output_height, output_width = y.shape[1:3]
        output_image = y.numpy().reshape(output_height, output_width)
        return output_image