Using opencv image (frame) as input for Inception prediction

Question

I am using this script to run a prediction on an image: https://github.com/googlecodelabs/tensorflow-for-poets-2/blob/master/scripts/label_image.py/

Specifically, the