Not perfect, but not too bad either, given that it took 30 seconds to do it. I'm sure the size and shape of the selection changes the result, and then the filter has a "radius" parameter which by default is set to 100 and can be changed, and I guess it will change the result.
Well, it's still quite obvious that something is missing there, as the central area of the picture is visibly less blurred than its surrounding. I guess you could fix that easily by just blurring it with ordinary gimp blur. (But great start indeed, comrade. :o))
Another technique is to normal-crop out the sky and the distant grass that is so blurry, "content-aware" crop out the duck, then paste back in the stuff you originally cropped out. The white bits at the bottom are sky. The blurry bits are where grass far in the background was stitched in.
Not nearly as good as what the photoshop demo did. In particular, the photoshop demo had examples of content aware fill with shadows moving through the object to be removed. In this example you can see the line/shadow of the overhang being eliminated with the purse. Maybe the photoshop fill would have done the same and they just cherry-picked some examples, but right now I'm thinking it's just better.
It's hard to tell, because unless you had access to a different video than I did you can't really see what Photoshop did, even after bumping it up to 480p. In this case I think we have the raw files, so you have the chance to examine the results up close and personal. It's not a direct comparison.
In the third picture, if you look in the bottom left and right corners, you can see a sudden colour change along the lines where the original picture's content ended. On the right, for example, the original cloud is sort of yellowish, then it suddenly becomes a more neutral grey along a diagonal line.
In the sky it's harder to tell, but it's still there too.
Probably not. There are a number of image inpainting, texture synthesis, and reconstruction algorithms that can be applied, with different behaviors regarding noise, color correction, detail scale, handling of regularity/irregularity, temporal coherence, run time, etc.
GIMP's Resynthesizer is based on a 2005 PhD thesis, while the Photoshop thing is probably based on a 2009 SIGGRAPH paper.
It looks like it doesn't really have a UI -- is there a version factored out of GIMP that just takes the source image and a mask image as arguments?
For the art producers the Photoshop version would me massively superior just for integrating with the UI, but for our purposes a version of this as a standalone utility or baked into PIL or ImageMagick would be fucking awesome.
Gimp has a batch mode that lets you run commands without loading the UI. That would probably achieve the same thing, although it might be a little much if you're sure that all you want is resynthesizer.
As I said in the CS story, this is cool but nothing new. I don't see how it's different from Alien Skin's Image Doctor (http://www.alienskin.com/imagedoctor/index.aspx ), which has been available as a plugin for years. I've been using it from PaintShopPro for quite some time.