An AI model that recognizes and classifies objects in a first pass.
This could offer several advantages.
For upscaling:
* Specialized models (for faces, trees…)
* Use information of a close up shot to correctly fill in missing information from further away
For frame interpolation:
* Real knowledge of the location (and possibly rotation) of objects
* Knowledge of how a specific object moves (wheels of a car should move towards driving direction)
