I was watching this https://www.youtube.com/watch?v=n1ViNeWhC24 and one of the main examples is with feature detection in an image. So there is a way to derive a sparse encoding for a particular image using these common features that were detected with unsupervised learning. Could that type of sparse encoding be used as a basis for encoding many frames, i.e. a whole movie? Maybe with the time dimension being represented somehow in the encoding/feature detection?
Sorry if this question doesn't sound right to you, I'm obviously not an expert in ML or video compression.
[link][2 comments]