Segmented region in image created from model disappears when using openCV to convert to binary

Question

I have trained an ML model to segment images into red and blue, see code:

# load an image
def load_image(filename, size=(256,256)):
    # load image with the pref