why the prediction mask from image segmentation network is fliped?

Question

i trained a unet network in order to predict the plate locations from a car image, here is the code that displays my images

path = "/home/mimus/apifave/i