Hm, it's not a great algorithm compared to some others that have been discussed here. There was a crazy one last year where it could replace entire large regions of the photo automatically and there was no indication anything had been altered.
This looks like a poorly done paste and smear job. For a student's semester project, which I suppose it is, it's great.
Person-silhouetting seems more difficult to do well than content-replacing. I wonder how this falls apart in complex scenarios? Still, an interesting idea.
Yeah agreed. Smooth replacement of content is great, but I think you could easily build seam carving into this. The hard part is doing the person-detection in the first place.
This looks like a poorly done paste and smear job. For a student's semester project, which I suppose it is, it's great.