In a post from the company’s AI blog, researchers present findings about a machine-learning tool that can upscale photo quality and the results look ripped from an early 2000s sci-fi thriller.
One of the models in question is known as SR3 or super image resolution, which takes a low-res photo as an input and manages to construct a high-resolution image from pure noise.
“The model is trained on an image corruption process in which noise is progressively added to a high-resolution image until only pure noise remains. It then learns to reverse this process, beginning from pure noise and progressively removing noise to reach a target distribution through the guidance of the input low-resolution image.”
In addition to the super-resolution model, Google has been experimenting with a second model in CDM or class-conditional image generation.
An example of the CDM model blowing up and increasing the quality of a 32 x 32 image.
PetaPixel“We built CDM as a cascade of multiple diffusion models. This cascade approach involves chaining together multiple generative models over several spatial resolutions: one diffusion model that generates data at a low resolution, followed by a sequence of SR3 super-resolution diffusion models that gradually increase the resolution of the generated image to the highest resolution.”
These AI-powered models will not only be able to restore those old photos you have somewhere on an external hard drive, but also help with medical imaging.