Google’s latest upscaling technology makes CSI TV shows a reality


Google has introduced a new upscaling technology that turns low-resolution images into detailed, high-resolution images.

Researchers Jonathan Ho and Chitwan Saharia shared details of the technology in an article on Google’s AI blog. The new approach to oversampling relies on diffusion models, which Google first proposed in 2015. These models learn by gradually corrupting the image until it becomes “pure noise”. Then, the algorithm learns to gradually denoise it until it produces a clean image again.

For the last approach, the researchers combined Super-resolution via repeated refinements (SR3) and a conditioned synthesis model by class, called Cascading diffusion models (CDM).

Super Resolution Image works more or less like the diffusion model described above. Once the algorithm is trained, take a low resolution image as input and construct a corresponding high resolution image from pure noise.

CDM is a class-based conditional diffusion model trained on ImageNet data to generate high-resolution natural images, and Google built it “as a cascade of multiple diffusion models.”

“This waterfall approach consists of stringing together several generative models on several spatial resolutions: a diffusion model that generates data at low resolution, followed by a sequence of SR3 super-resolution diffusion models which gradually increase the resolution of the image generated up to the highest resolution. . “

Along with the latest oversampling technique, Google has also introduced a new data augmentation technique called increased conditioning, which should further improve the quality results of the CDM samples. You can read the entire blog post at Google AI Blog and learn a little more about the whole process and its results.

A few uses of this technique come to mind. One of them is to oversample your own photos downloaded from Facebook, Instagram or WhatsApp which compress them. This can sometimes be faster than searching for originals. On the other hand, it can be tricky because someone can steal a small resolution of your photo, oversample it, and use it however they want. As with many technologies, this one is neither good nor bad in itself, it all depends on who uses it.

Ultimately, this could come in handy for authorities when trying to identify criminals from low-resolution security camera shots. Maybe these “Improve!” Moments from the CSI television series will finally become a reality.

[via DPReview]

Source link

Leave A Reply

Your email address will not be published.