3 ms·

So what would happen if you trained this model on frames from movies that were annotated? You could leverage CC descriptions and maybe use a classifier to autom

So what would happen if you trained this model on frames from movies that were annotated? You could leverage CC descriptions and maybe use a classifier to automate a lot of the annotation. Would it be able to create novel but contiguous predictions for the next frame?

operator-name 4y ago
The problem would still be temporal stability. You'd have to train on series of frames, but then I'm not sure diffusion models are well suited for that?