Turn your low-res photo into high-res with this new AI technology

Google’s SR3 is a super-resolution diffusion model that takes as input a low-resolution image and builds a high-resolution image from noise. Using the CDM method, a low-resolution image of 64x64 can be diffused to 264x264 resolution and then further to 1024x1024.

There are times when we wish a distorted, low-image was not the way it was because the same image would make for a great image had the quality been better and in high resolution. Thanks to Google’s AI-based image upscaling technology, you can now enhance the quality of low-resolution images exponentially. In a post on Google’s AI blog, the researchers introduced 2 diffusion models to generate high fidelity images:
1. Image Super-Resolution (SR3)
2. Cascaded Diffusion Models (CDM)

What is Image Super-Resolution
This model takes as input a low-resolution image, and builds a corresponding high-resolution image from pure noise. The machine uses a process of image corruption where noise is consistently added to a high-resolution image until only pure noise remains. It then reverses the process that removes the noise and reaches a target distribution through the guidance of the input low-resolution image.

What is Cascaded Diffusion Models (CDM)
Once the SR3 model has shown effectiveness, the CDM model is brought into action. Google defines CDM as “ a class-conditional diffusion model trained on ImageNet data to generate high-resolution natural images.” As per the post, Google built CDM as “a cascade of multiple diffusion models” since ImageNet was a difficult, high-entropy dataset. The model is a combination of multiple diffusion models that can generate images of increasing resolution. It starts with a standard diffusion model at the lowest resolution and is followed by a sequence of super-resolution models that can successively upscale the image and add higher resolution details.

Along with SR3, Google also uses a new data augmentation technique, called “conditioning augmentation”, that is said to further improve the sample quality results of CDM. Using the CDM method, a low-resolution image of 64x64 can be diffused to 264x264 resolution and then further to 1024x1024.

With the introduction of these models, Google is looking to improve the natural image synthesis that has wide-ranging applications but poses design challenges. “With SR3 and CDM, we have pushed the performance of diffusion models to the state-of-the-art on super-resolution and class-conditional ImageNet generation benchmarks,” Google said.