Code for intermediate feature extraction?

Hi,

Thanks for sharing this great work!
I'd like to extract features before the prediction layer for custom videos. Is there any simple way to achieve the goal?

Thanks,
Melo