Image Completion AKA Content Aware Fill

One challenge with 360 photos and videos is that everything is in frame! There is no place to hide, so the tripod is in every shot, and the photographer is in a lot of the shots as well. For my 360 image pipeline it would be nice to automate removing the tripod at the "nadir" and also removing people. Adobe's content aware fill does this fairly well, the steps are to convert an equirectangular to cube faces, grab the downward cube face, export to Adobe, content fill, import, and finally combine the bottom cube face into an equirectangular. That's a lot of work that should be scripted, or ideally a Fusion node created.

A paper was presented at SIGGRAPH 2017 by Satoshi Iizuka on his work with image completion. His code is available on Github, and definitely worth taking a look.