The source – ai.googleblog.com
Google has unveiled technology that can increase image resolution by up to sixteen times without loss of quality. The company found that the results of the new approach surpass all existing technologies.
Google Brain Team trained a machine learning model to turn low-resolution photos into detailed images with virtually no loss of quality. The new solution combined two approaches: SR3 (Super-Resolution via Repeated Refinement) and CDM (class-conditional diffusion model).
The SR3 technology creates a high resolution image from noise, based on a low resolution original image. The model is trained to gradually distort the image, and then the algorithm reverses the process by gradually removing the noise and adapting the denoising models, guided by the original low-resolution picture. SR3 performs best when scaling portraits and nature shots.
Further, CDM technology is used in the scaling process. It augments the generated high-resolution image with new data and offers a cascading approach. It first generates low resolution data and then gradually increases the resolution of the generated image to its highest value.
Of course, this can have a significant impact on the quality of the content later on.