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.
Comments
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.
Seam carving? http://www.youtube.com/watch?v=vIFCV2spKtg
Person-silhouetting seems more difficult to do well than content-replacing. I wonder how this falls apart in complex scenarios? Still, an interesting idea.
Patch match: http://www.youtube.com/watch?v=otAbvtiabeY
Yes, that's the main one I was thinking of, thanks. The seam one also I had in mind.
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.
On an unrelated note, "Belongie" is a fantastic name.
Yes, I know it's French :(
That's cool, but do they have a "pedestrian" one too?