PatchMatch: Structural Image Editing on Vimeo

See all Show me

2. PatchMatch: Structural Image Editing

by Dan Goldman

6 months ago

1. Interactive Video Object Manipulation

by Dan Goldman

1 year ago

This short video demonstrates advanced image editing techniques made possible by a new algorithm we have developed for quickly finding good matches between all parts (patches) of two different image regions.

The project is a collaboration between Adobe, Princeton University, and the University of Washington. We will present it in the technical papers session at SIGGRAPH '09 in New Orleans this August. You can find more information (including a PDF of the paper and the full video with more background and technical details) at

cs.princeton.edu/gfx/pubs/Barnes_2009_PAR/index.php
adobe.com/technology/graphics/patchmatch.html

Enjoy!

Credits

110 Likes

Nate Notermann
5 days ago
Zenzuke
12 days ago
/jw
14 days ago
rozmy
28 days ago
Dock Hoilman
29 days ago
Dom O'Brien
29 days ago
Vadim K
1 month ago
Nils Kunze
1 month ago
Till
1 month ago
Daniel Saegebarth
1 month ago
See all 110 likes

Tags

Kam Sayah 5 months ago

This is big Dan, I wish you all the best in this project.
chiaminhsu 5 months ago

cool!
Carlo 4 months ago

Wow, impressive.
Justin Salsburey 2 months ago

I can't believe there aren't more comments on this. This is huge! I thought it was a hoax at first. Amazing amazing stuff.
Jt Root 2 months ago

any idea of when it will be implemented? standalone or through Photoshop?
josh@firestorm 2 months ago

hopefully they bring something similar for after effects too!
DieTapete 2 months ago

WOOOOOOWWW! This will be implemented in PS CS5?
Roobik 2 months ago

insanely cool
ViDEOïNA 2 months ago

Great!!
benjy fichet 2 months ago

amazing !! wanna see it in CS5 :D
lostmarblemedia 2 months ago

wow....
edvard brun 1 month ago

This may be "stretching" the limits but could these algorithms be extended to video as well (at least in theory)?
Eli Shechtman 1 month ago

In principle yes and we are thinking about this... It's just that it will probably not be so fast as with images, simply because in video one typically has many more pixels (patches) to synthesize and the search space is much larger.
I have done previously some work on video completion - wisdom.weizmann.ac.il/~vision/VideoCompletion.html
and video summarization - wisdom.weizmann.ac.il/~vision/VisualSummary.html
and I can tell you that these methods were VERY SLOW using previous search algorithms (I am talking about hours for tiny videos).
Using video PatchMatch I imagine it could be reduced to be just SLOW :)
Dock Hoilman 30 days ago

NICE! Thanks for sharing.
Nate Notermann 6 days ago

Oh my word this looks promising. PLease continue. Any idea when this could possibly expected to be bundled into a consumer product?