What is Digg?Sponsored by newegg
Tips to be first in line without camping out in the cold! view!
newegg.com - The Pre-Black Friday Sale. No Lines, No Crowds; Click and Save in the comfort of your own home.
92 Comments
- captainpete, on 10/10/2007, -1/+74I definitely misread their domain name.
- waz67, on 10/10/2007, -1/+70That is freakingly cool, if it works as it says it does. Can't wait for this to be usable by the general public.
- nyx210, on 10/10/2007, -2/+43Will this work with porn?
- scabbers, on 10/10/2007, -1/+42I just hope they have lots of boobs in that database.
- cyrusthevirus, on 10/10/2007, -5/+33This technology was inevitable. With a database of millions of pictures available, you can easily create a "franken-photo." Gone are (will be) the days of Photoshop. This is drawing from real imagery, and thus potentially more realistic. It just remains to be seen how noticeable the editing is.
- kenvsryu, on 10/10/2007, -2/+24Edit out your life and replace it with something cooler.
- tech42er, on 10/10/2007, -1/+23Technically, it can do that now. If the images are from two different places, it produces an image no one has ever seen before.
- StickWST, on 10/10/2007, -3/+25Then funny part is he's thinking of semen.
- idc5, on 10/10/2007, -2/+23So if I were in the dial up days, would it be able to fill the rest of the incomplete porn photos which got stuck before the good parts?
- Darkhacker, on 10/10/2007, -3/+18Wait, does this work with any image? Or does this just try to find the same picture from its database and fill in the selected part? I honestly don't see how in the hell you can take your own unique photo, crop out a part, and have this program fill it in so that it looks real and seamless. It would be interesting to see the effect of having done this on a picture of a human face. I seriously think that I am misunderstanding the article. No way this could be real in the way that I am thinking.
- fbriggs, on 10/10/2007, -1/+15Here's a direct link to the researchers site:
http://graphics.cs.cmu.edu/projects/scene-completion/ - Coded1, on 10/10/2007, -1/+15I'll be really impressed when the program starts making images of things no one has ever seen before ;)
- myfanwy, on 10/10/2007, -3/+17it adds in info which COULD be there, not necessarily WAS there. big difference, and the examples show that.
it makes the image look nice, but not real.
digg for the tech, bury for the misleading and not necessarily useful application - 0firefly0, on 10/10/2007, -3/+17Hah. Spoke too soon, *****.
- tech42er, on 10/10/2007, -1/+14Exactly. That's why it's so cool and why we're digging it.
- sakuraz, on 10/10/2007, -0/+131.Steal someone else's brilliant idea.
2.Apply it on porn
3.???
4.Profit! - drouk1556, on 10/10/2007, -1/+14This is what Photoshop's vanishing point should have been.
- Sandurz, on 10/10/2007, -0/+12It doesn't take what's missing and guarantee it will fill in exactly what's missing. It just uses something that fits seamlessly into the image.
Like in the example, a house was cut out and it filled ships and water in.
If it fits, it can be used. And if somehow exactly what's missing is in the database, then that'll be used. - captainpete, on 10/10/2007, -2/+13You're ***** badass. This is a better link though: http://graphics.cs.cmu.edu/projects/scene-completion/
- Dorkbot101, on 07/15/2009, -0/+9Here's another project with a similar idea... using an algorithm to detect patterns in one image and apply it to another seamlessly...
http://cpl.cc.gatech.edu/projects/CBIS/ - gkskillz, on 10/10/2007, -0/+8It isn't as seemless at it those two examples make it seem. Look at the many examples given in the original research paper (http://graphics.cs.cmu.edu/projects/scene-completion/scene-completion.pdf warning 11mb pdf) and you will see that it more or less works well but does not produce perfect seamless images except in a few rare instances.
- GetShorty, on 10/10/2007, -1/+8Hot female celebrities beware.
- capiCrimm, on 10/10/2007, -0/+7You mean like...
http://en.wikipedia.org/wiki/AARON
... computers can also get patents and create original music. But don't worry, protein machines are still superior -- for the moment. - undetected, on 10/10/2007, -1/+7I'm blinded by the clutter on that page.
- geminitojanus, on 10/10/2007, -0/+6You're going to be waiting for a little while. This algorithm used a few million high resolution images for semantic scene matching (for a grand total of about 400GB in source images) and a very complex algorithm for edge matching ("Local Context Matching" which requires two colorspace conversions per pixel, LAB->RGB and RGB->LAB, and an 80 pixel radius around the edge fragments). You could pre-generate a lot of the information and use a very huge hash table to help speed up the process, but it's still not a desktop application, and won't be for at least another 5 years (or another year or so if the graphics card companies change their business models and give us better access to the internals of their chips).
This kind of thing is better suited for Hollywood special effects departments; cut out some part of the frame, have the algorithm find a new image to fill in the blanks. - Goner, on 10/10/2007, -0/+6they're researching that.. It'll be like you were never here.
- ismith, on 10/10/2007, -0/+5Agreed. Gone are the days of those pictures you get from myspace friends with bad boobs. Just run em through this program after editing them out and voila, saggy 32B to perky 36C!
- EricJD, on 10/10/2007, -1/+6most of those look really fake
- JohnnyXmas, on 10/10/2007, -0/+4FTA: "by drawing on a huge database of more than a million images from the World Wide Web"
This is a contradiction in terms. A quick Google Images search turns up more than a million pics, and none of them are ever what I want. - seanc6610, on 10/10/2007, -0/+4gizmag? seriously?
- danielsan1701, on 10/10/2007, -0/+4My favorite tech blog is spunkrocketcumblaster.com
- inactive, on 10/10/2007, -0/+4"Our algorithm requires over an hour to process one input image on a single CPU (50 min for scene matching, 20 min for local context matching, and 4 min for compositing)."
- qualish, on 10/10/2007, -1/+5Would it fill in missing parts of people photos with... you know... the nudes?
- skinrock, on 10/10/2007, -1/+5Those are very large, well cut chunks. I mean, how hard is it to replace a mountain scene with..gasp...mountains...and a large chunk of water with...gasp...water with some sail boats. In all actuality, the purpose they describe this would be used for would rarely result in a large portion of the image to be bad. I would like to see how "smart" this is for a very small part of the image.
- ggko, on 10/10/2007, -0/+4great, now every photo of a cat will turn into an LOLCAT image.
- ArthurSucks, on 10/10/2007, -0/+3Just in case:
http://www.gizmag.com.nyud.net:8080/go/7812/ - TypeEE, on 10/10/2007, -0/+3remove a penis? the software fill in a bigger one for you!
- HonoredMule, on 10/10/2007, -0/+3You think something is real just because you think you see it "in person?"
- logicbus, on 10/10/2007, -0/+3My guess is that shadows would fail to match a lot; I suppose the algorithm could be adapted to account for them.
- PARAPA, on 10/10/2007, -0/+3Through the interweb's pipes and tuuuubes
- dawnraid101, on 10/10/2007, -0/+3not really
- logicbus, on 10/10/2007, -0/+2The PDF linked above hints at some useful applications. One I can think of is taking modern pictures of ancient cities, and using the algorithm to remove modern elements such as television antennas and power lines.
- Jamminn, on 10/10/2007, -0/+2Yeah, agreed. Dugg for the Algorithm, not too sure about the application of it.
- hailxkill, on 10/10/2007, -0/+2Nothing will ever replace the human hand and eye when it comes to doctoring a photo, creating a photo illustration, or doing many of the other things you can do with Photoshop. What about people who paint in Photoshop, or create graphics for print media? You people are definitely underestimating humanity and Photoshop, and putting way too much faith in your ***** algorithms you babble about all the time.
- najdorf, on 10/10/2007, -1/+3Baaaah, until I see it working with a click of my mouse I'll have serious doubts on this... ok, you can get similar colors, but "semantically valid" (i.e. recognize whats represented in a pic a match it with something appropriate)? I bet most of those pics were made with a lot of human "help"...
- swrlyhrly, on 10/10/2007, -0/+2where you shaking your head when you wrote that or were you moaning?
- dawnraid101, on 10/10/2007, -0/+2dugg down for "You used it to" usage .... awe crap
- diggtard, on 10/10/2007, -0/+2The picture with the boats fits in really well so that you can't tell it was pasted in there. But to me the imagery inside of the block pasted in, the boats, doesn't seem to be too well thought out. The shadows from the boats appear to be cast in one direction while rest of the shadows in the picture form in the other direction. Not perfected yet but still amazing for such little human manipulation.
- emanuelez, on 10/10/2007, -1/+3I implemented an earlier version of this algorithm as a GIMP plug-in one and a half years ago. Even if it was conceptually very easy it also produced extremely good results. The drawback was speed: even for small images the process took several minutes. I would be curious to implement this algorithm as well. It seems to be way more complex and accurate, but once again... what about speed?
- capiCrimm, on 10/10/2007, -0/+2except that it might be a different image every frame? The algo would need to be redesigned I'm assuming to fit film.
-
Show 51 - 92 of 92 discussions
© Digg Inc. 2009
— Content created and posted by Digg users is