Inference takes one and a half day to run on a single movie

[fatihdinc]

Hi!

I have been using deep interpolation on our large-scale calcium imaging movies, but it seems that even with a state-of-the-art GPU (3090 Ti), inference step takes around 1.5 days in a 512x512x50,000 movie. I am suspecting that the inference is not using the GPU by default? Which value should I change for it to use GPU and also load with multiprocessing? In contrast, a reasonable training process takes 1-2 hours with good convergence on the first 2000 frames, which seems to be enough for these movies.

Also, is there a demo for inference on 1p movies? I was able to see a training script, but not demo. Thank you!