US Patent 11895308 Video encoding and decoding system using contextual video learning

Aspects of the disclosure are directed to a video system. In one aspect, a video encoding/decoding method or apparatus includes segmenting each frame of a sequence of raw image frames to derive objects of interest; ranking the objects of interest to generate a ranks and ranked objects of interest; generating frame keys corresponding to the objects of interest; retrieving a source stored model data; and generating a training metric using the frame keys and the source stored model data. In another aspect, it includes receiving frame keys from a source portion; retrieving a destination stored model data and sending the destination stored model data to a destination contextual video learning system; and retrieving objects of interest from the destination stored model data using the frame keys and the destination contextual video learning system.