Content-aware fill in VIDEO

[feppe]

You've seen it in GIMP and Photoshop CS5, here demoed in video - and claimed to be in real time as well, here. Video looks legit (check the bathroom scene at 0:46 with reflection of the deleted object), but the brief explanation of how it's done is laughable.

[RFPhotography]

If you have to do it in each frame manually, how's it much different from importing the individual video frames to layers in PS and doing it that way? 

[feppe]

If you have to do it in each frame manually, how's it much different from importing the individual video frames to layers in PS and doing it that way? 

If this happens in real time as claimed by the makers, it's very different from doing it in post frame by frame.

[Eric Myrvaagnes]

It's easy. Say the offending object occurs in 1000 consecutive frames. You just set up 1000 animators, each with a computer and CS5. At a signal they each start doing their own content-aware fill, and a fraction of a second later they're all done (except for reassembling the frames). 

Eric

[RFPhotography]

If this happens in real time as claimed by the makers, it's very different from doing it in post frame by frame.

I'll have to go look at the link again but I didn't take from it that this was a real time thing.  There's some pretty serious artifacting in one section as well (the sidewalk, I think it was).

[hjulenissen]

It's easy. Say the offending object occurs in 1000 consecutive frames. You just set up 1000 animators, each with a computer and CS5. At a signal they each start doing their own content-aware fill, and a fraction of a second later they're all done (except for reassembling the frames). 

Eric

Not if random noise/camera-movement means that the "optimal" seam then is jittering back and forth 10 pixels every frame. That will look really bad.

-h