How to run classifications on video frames in Inception, without starting a new tensorflow session for every frame?

Question

I am using this repo to run a classifier on images with Inception. However, instead of running on images, I want it to loop over a video and run a detection on every frame.