Google’s New AI Model Turn Low-Resolution Images into High-Quality Pictures. As researchers push their limits to develop advanced artificial intelligence (AI) technologies, we have seen several AI tools and systems to improve imaging technology.
For example, we have seen AI tools to remove backgrounds from images and de-blur photos instantly. In addition, Google has developed two AI-based tools based on diffusion models to convert low-res images into high-quality photos.
Google’s Research team has introduced two new approaches which use machine learning to enhance images. Super-resolution can be used to restore old family portraits and to improve medical imaging systems.
Google’s New AI Model
Sometimes 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. Google notes that super-resolution models transform a low-resolution image into a detailed high-resolution image.
Super-resolution can be used to restore old family portraits and to improve medical imaging systems. Google notes that the Diffusion models were originally proposed in 2015 but have seen a revival now due to their training stability and their promising sample quality results on image and audio generation.
Now, Google’s Research team has introduced two new approaches which use machine learning to enhance images. Google has introduced two models, including SR3 — Image Super-Resolution and CDM — Class-Conditional ImageNet Generation, which according to Google, “push the boundaries of the image synthesis quality for diffusion models.”
Google notes that the SR3 is a super-resolution diffusion model that inputs a low-resolution image and builds a corresponding 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,” Google noted in a blog post.
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)
Researchers describe this model as a class-conditional diffusion model, trained on ImageNet data, which produces the natural image of high resolution by making a 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 example of the CDM Model converting a low-resolution image of 64×64 to 264×264 resolution and then 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 using Gaussian noise and Gaussian blur.
With all this, we can say that Google is improving image quality by introducing AI tools like SR3 and CDM.
As you can see, the results are impressive, and the final photos, despite having some errors (such as gaps in the frames of glasses), would likely pass as actual original photographs for most viewers at first glance.
“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 researchers write. “We are excited to test further the limits of diffusion models for a wide variety of generative modeling problems.”