Can numpy read images run on GPU without using Image Data Generator: Keras and TF

Question

I want to know if it is possible to read images using cv2 and then store them as numpy array from directory without using Image data generator and still run the model on GPU