Google Brain, the deep learning project, can enhance pixelated images just like the movies
If you are a CSI tv-series fanatic then you definitely know about the infamous zoom and enhance process of suspect’s photos. And of course you know that this is not possible in real life. Well until now. According to a report Google Brain has developed a software that actually make images from pixelated and small photos. And ArsTechnica has a set of photos to prove what Google Brain has achieved.
Now what you look at the left column is a set of very pixelated images. You can barely make any characteristics. The images on the center are the ones that created by Google Brain software. And for comparison the real photos are the ones on the right. But what is the secret behind Google Brain’s achievement? According to the report the combination of two neural networks are responsible for the magic.
The first factor is the conditioning network and tries to map the 8×8 source image against other high resolution images. In other other words it brings high resolution images to 8×8 and tries to make a match.
The other part is the prior network. It uses an implementation of PixelCNN that tries to add more realistic hi-res details to the original 8×8 image.So it fills the missing pixels with similar. Using the two networks Google Brain can create an image that looks pretty similar to the real image.
Overall Google Brain’s software seems to be very effective and successful during the testing.
But it’s not the first time that a software is trying to make our digital content better. For example Twitter bought a machine learning startup, by the name Magic Pony, that can sharpen videos in real time. So we won’t be surprised if similar software will appear in various platforms.
If you want to learn more about the software by Google Brain you can check out the research paper and find out all the details.
And now when we hear “zoom and enhance” on a TV show or a movie we will know that it is possible.