The Google Brain project has now created a new software to create detailed images from small pixel images. This new pixel recursive super-resolution technology has a real cool effect, it can be extracted from several source pixel load details. For example, Google Brain renders some 8x8-pixel images and then turns them into crisp images.
This involves the coordination of Google's two neural networks. The first is to adjust the network to map 8x8-pixel source images to other high-resolution images. Basically, it reduces the other high-resolution images to the same 8x8 size and tries to match the functionality. Then, a second neuron called the previous network begins to work, using executable PixelCNN to add realistic high-resolution details to the 8x8 source image. If the network knows that a particular pixel may be an eye, when you enlarge the pixel image, you will see the shape of the eye. Or eyebrows or mouth.
The technology was tested and it's very successful in deceiving the human eye. It shows human observers a high-resolution image of famous faces and images taken by Google Brain, 10% of human observers are fooled, and in the Google Brain generated bedroom images, 28% of human observers Google Brain calculates the generated image spoofing.
Google this technology, although based on speculation out of nothing. However, you can guess according to all kinds and get something that may be close to the exact. For example, the image of a suspect and so on.