I find code here could get different result for same video if I compute multi times.
How could I make the code all deterministic to get fix reuslt for one video.
If it has some randomness, it may not good to be a evaluation metric for video generation model.