Google has recently introduced two AI-powered tools – image super-resolution (SR3) and cascaded diffusion models (CDM) -that use diffusion model to transform low-resolution images into high-resolution images. (Resolution is Pixels Per Inch (PPI), the high resolution deals with more PPI whereas low resolution deals with few PPI)
In a blog post, Google Researche’s Brain Team introduced these two diffusion models to generate high-resolution images. Let’s take a look at both them.
Super-Resolution Model (SR3)
As per Brain Team, this model is a super-resolution diffusion model that turns low-resolution images to high-resolution images from pure noise. It takes an input of low-resolution images and then goes through an image corruption process which makes it trained in adding noise until only pure noise remains in the image. The process is then reversed by removing the noise to get the final image.
The model can convert a 64 × 64 image to a 1024 × 1024 image. For super-resolution of the face and natural images, SR3 gives strong results and this is due to the training given to SR3. As per the blog post, the super-resolution or SR3 can have multiple applications like enhancing the existing medical imaging systems and restoring old family portraits.
Here’s an exapmle of SR3 model being used to conert a low-resolution image of 64×64 to 1024×1024 resolution image.
Cascaded Diffusion Model (CDM)
Researchers describe this model as a class-conditional diffusion model, trained on ImageNet data, which produces the natural image of high resolution by making chain of several generative models over several spatial resolutions.
The process of converting images from low to high resolution is completed in two steps:
- The diffusion model is used to produce data at low resolution.
- The sequence of SR3 super-resolution diffusion models is made.
Here’s an expmple of the CDM Model being used to convert a low-resolution image of 64×64 to 264×264 resolution and then further to 1024×1024.
Additionally, researchers also introduced a new data augmentation technique named “Conditioning Augmentation.” Its work is to improve the output given by CDM by making use of Gaussian noise and Gaussian blur.
With all this, we can say that Google is putting its efforts to improve the image quality by introducing AI tools lke SR3 and CDM.
“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,” the researchers wrote in the blog post.