Slashdot Mirror
Google Brain Creates Technology That Can Zoom In, Enhance Pixelated Images (softpedia.com)
Google Brain has created new software that can create detailed images from tiny, pixelated images. If you've ever tried zooming in on an image, you know that it generally becomes more blurry. You'd just get larger pixels and not a clear image. Google's new software effectively extracts details from a few source pixels to enhance pixelated images. Softpedia reports: For instance, Google Brain presented some 8x8 pixel images which it then turned into some pretty clear photos where you can actually tell facial features apart. What is this sorcery, you ask? Well, it's Google combining two neural networks. The first one, the conditioning network, works to map the 8x8 pixel source image against other high-resolution images. Basically, it downsizes other high-res images to the same 8x8 size and tries to make a match on the features. Then, the second network comes into play, called the prior network. This one uses an implementation of PixelCNN to add realistic, high-res details to that 8x8 source image. If the networks know that one particular pixel could be an eye, when you zoom in, you'll see the shape of an eye there. Or an eyebrow, or a mouth, for instance. The technology was put to the test and it was quite successful against humans. Human observers were shown a high-resolution celebrity face vs. the upscaled image resulted from Google Brain. Ten percent of the time, they were fooled. When it comes to the bedroom images used by Google for the testing, 28 percent of humans were fooled by the computed image.
Google can put together images based on smaller images that look like faces.
I don't care how fancy the algorithm is, the original data was lost. This is still just a guess about the original content. It's just a better guess than was possible before.
I just hope law enforcement doesn't think they can use this to solve any crimes.
One of our competitors trademarked the term "hypothesis". From now on, we will call them "boneheaded ideas".
Is this Blade Runner-esque? Decker summoned some wicked camera technology. Don't bother me with those pesky limits to the physical laws.
This is just pure, 100% guesswork done with a computer. You cannot 'enhance' information that simply does not exist. The 10% of "fooled" people just mean those people were not familiar enough with what that celebrity actually looks like to tell the difference.
CSI "enhance it!" remains fiction, sorry.
Feed it minecraft screenshots and japanese porn, and see what the result is.
Inheritance is the sincerest form of nepotism.
No. Google hipsters just want to make handing over your privacy as slick and painless as possible, hoping it'll make you less likely to kneejerk. That's why that creepy-friendly corporate newspeak in big anti-aliased fonts is their favorite way to communicate with users.
So in other words... from a small picture of the earth viewed from orbit, Google can now show me my house AND the address on the UPS package sitting at my doorstep?
Amazing!
If it doesn't do uncrop, it's lame.
I have discovered a truly marvelous proof of killer sig, which this margin is too narrow to contain.
Sorry, but the VT state troopers already have this technology: https://www.youtube.com/watch?...
now we know the perception and purpose,
So Google fed its algorithm millions of high-resolution images. It then feeds a pixellated image of a celebrity, and AMAZING! The algorithm can create an image of that celebrity so detailed that it fools people.
Of course, chances are a good portion of the source material is pictures of that celebrity. This would be far more impressive if it could construct a police artist profile image from a pixellated source. One that could be reliably matched to an individual that's not even in the massive collection of images fed into the source.
this is perfect for your "real" profile pic on dating sites, just upload a google enhanced image of your self created from your 8x8 pixel image. yes this celebrity is really me.
TV Detective: "We have this security video showing the murder."
TV Lab Rat: "It's too grainy to tell how is it?"
TV Detective: "Can't you enhance it?"
TV Lab Rat: "Sure. Who do you want it to look like?"
I'd setting for even low res porn that didn't have all those long, loving close ups of the guy's hairy ass and shaved balls. Is all porn made by amateurs these days, and they have their gay buddy run the camera cuz he's not interested in the hot chick they hired?
This is nothing new. They have been doing this in movies for ages.
A while ago, someone made the nnedi upsampler that uses neural networks to upsample. It's still one of the best image upsamplers available.
Google's approach is quite a bit different. Where nnedi worked to better extract detail out of what was already in the image, Google seems to literally fill in detail that was probably in the source but maybe not. Much, I guess, like how our own memories work. It's an interesting approach and the results look quite fantastic. My only question is how well it will work on a random sampling.
Move over, Mount Rushmore, Google now has an algorithm that can wallpaper the Ceres asteroid with the face of every American who has ever been photographed—all the way back to a pinhole camera exposing an onion skin soaked in lemon juice and potato starch.
The only reason this works as well as it does (poorly - 10% to 28% according to the post) is that humans are hard-wired to find faces among images. We'll accept anything with two small ovals on the same line with another shape roughly in the middle, below. Nose optional. Try the equivalent with random images or music, and I'll bet the success rate would drop to nil.
Fractal Image Format was compression that used self-similar parts of the same image to decopress to higher resolution https://en.wikipedia.org/wiki/...
I was really interested n fractals and fractal compression for a time. And while you could get some insanely high compression ratios, the technique was lossy and decompression took 50+ hours (and 2 hours to compress). A potential use of the technology was picking out interesting artifacts from low rez space ohotos.
The summary's explanation of what this does isn't correct. It says:
Google's new software effectively extracts details from a few source pixels to enhance pixelated images.
It doesn't extract details from a few source pixels. It invents details to add to those source pixels, based on the knowledge that the pixelated image is of a face, and of what faces look like. It produces something that plausibly fits the input data. How close this is to the original image, pre-pixelation, depends on what images were in its training set.
This is an interesting piece of work, but it doesn't mean that you can recover data that has been discarded.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
On a single image the additional detail is guess work, the information isn't there. But think multiple slightly different but related images of the same subject, like any video filmed in the past 100+ years. Now you can cross-check and improve guesses with time correlation, and really find things that are there in terms of information, but invisible to the eye because each bit is spread over multiple frames.
Does this imply that movies have been lying to us all along? :-)
When I post photos that have personal information, license plates on cars, house numbers, names etc, I will do a two fold thing. Use the lens blur to blur them out, then go back over them with a masking color. Pretty simple in photoshop.
Movies won't be able to do this for two more years!
Rule 35 of the internet: "If it can be hacked, it will be". - Charles Stross
At least make the world a better place...
by making this a MAME scaler.
Cwm, fjord-bank glyphs vext quiz
For some reason, the first thing that comes to mind is a 2D pixelated Luigi now transformed into a six packed ultra good looking guy with the most macho mustache ever.
Maybe I need more coffee.
Yeah, I'm still not interested in close-ups of some guy's hairy ass and shaved balls.
nonono.
ways to track people they made UP on the spot!
you see, they're already tracking everyone so they'll just make these computer generated faces to track them.
(you can't track people effectivey with a 8x8 image. you could however generate fake evidence that shows that it _could_ match this and this dude, or even 99% of population.).
world was created 5 seconds before this post as it is.
Once again, Scifi foretells the future.
Good thing Bladerunner has gotten almost everything else wrong so far. At the very least it had the timing all wrong, since we're pretty unlikely to have replicants, off-world colonies, lots of flying cars, an inability to buy real snakes, or that level of smog or whatever choking LA by 2019.
This is a hacked account, for which the owner can not be held responsible.
> 8x8 pixel images which it then turned into some pretty clear photos where you can actually tell facial features apart. This is bullshit. It is mathematically impossible to compensate such loss of information. They might use other images to reconstruct original form a pattern, but this, obviously, has nothing to do with reality if the other images used as sources of details have not been made at the same moment.
If they have, say, a minute of pixelated video presumably they could estimate the orientation and position of key features of the face and then make progressively improving estimates of a higher resolution image.
I know that was not the focus of this research (to match a pixelated image to one of a number of high resolution alternatives). But much of the crappy blurry images I see are in the form of video and it seems to me that there would be multiple independent images of a face that could be assembled into a single better image using related information in other frames.
Nullius in verba
If there are multiple low level images such as would occur in a surveillance camera. It should be possible to combine them a get a much better guess at a true image.
Thin 80s ties, constant rain in L.A., video payphones, "Cityspeak" lingua franca, advertising blimps...
Some other stuff BR got right: Atari still in business, guns with LEDs, giant animated billboards, and revival of analog synthesizers. ^_^
"It's a good computer... for I to BM on!" - apologies to Triumph, the insult comic dog