When we talk about artificial intelligence, some already imagine themselves in a disaster movie in which the world is dominated by robots. But before new technologies catch on and enslave us, researchers continue to improve the technologies in this area. And that’s the case with Google, which has just revealed the fruits of its labor of refining the resolution of poor quality images.
Convert low resolution images to HD
This technique is called SR3 for super-resolution by repeated refinement. The concept: applying a Gaussian blur to a low resolution image to get a photo with pure noise. The next step is artificial intelligence to take care of it. After training, the AI uses noise reduction technology to achieve a higher resolution image while removing previously generated noise.
The result is so impressive that every second person even thought it was an original photo.
Artificial intelligence is thus able to reconstruct an image. It enlarges an image with a resolution of 64×64 four times and can then increase it to 256 pixels plus 1024 pixels through a cascaded process!
Create pictures from simple words
In addition to improving the quality of a photo, Google’s algorithm can also produce false natural images using a process called CDM for the Clas-Conditional ImageNet Generation. With a keyword he can suggest the creation of fake images with impressive realism.