Very practical for certain purposes is the method developed by this project ("Video Extrapolation using Neighboring Frames") - it is aimed especially at immersive FOV videos, i.e. videos from the first-person perspective, which must have a particularly wide format, e.g. for use in VR headsets. In order to be able to use older videos, which were produced in the classic 16:9 format, this process converts such videos into wide format.
Extra-wide extrapolated image from multiple Video frames
This works if a larger part of the space is shown by a horizontal camera movement - the algorithm then reconstructs an extra-wide video image for the current image from several preceding and following frames of a pan.