speed up inference

[wooni-github]

Thank you for your outstanding work.

I am currently implementing your solution using the command "python run.py --type Motion".

I was wondering if there are any methods to accelerate the inference process.
I have a substantial amount of available GPU memory and am working with a Full HD image and a large motion model.