A more general demo script

[sanshibayuan]

Given a video path and a saved model (checkpoint), get the frames and extract features if needed, then produce classification predictions.

• Integraded in one script so no extra packege needed.
• Modify the rescale_list function to be two wayed.
• Use opencv instead of ffmpeg for more general use.

I'm new to Github, please let me know if there is anything unproper, Thanks!

[harvitronix]

Thanks a lot for this! Will do some testing and merge.