In recent times, most companies have moved their focus towards Artificial Intelligence (AI) and machine learning, in an attempt to develop human-level decision making in them as well as improve their problem-solving capabilities.
Samsung and Google developed digital assistants to improve their smartphones, Ali Baba used AI to help solve daunting shopping queues, while Facebook made their own version to do content review, and so on.
A Better Alternative
Similarly, Nvidia – a famous GPU company – has developed an AI which can clean up/reconstruct images by “painting in” missing pixels with what should have been there.
It also removes unneeded content and replaces it with something it determines to be a better alternative. An Nvidia researcher said,
Our model can robustly handle holes of any shape, size location, or distance from the image borders. Previous deep learning approaches have focused on rectangular regions located around the center of the image, and often rely on expensive post-processing.
NVIDIA’s “Partial Convulation”
If you have used Adobe Photoshop or other photo editing software before, you might know that they let you manually reconstruct photos by using reference pixels near the damaged area to fix them.
This AI, instead of referencing the same picture, uses cognition to determine what should have been there instead. The company trained it using tens of thousands of different scenarios to teach it how to restore images without any reference.
Here’s how it works:
The results, in some cases, are not exactly flawless as it’s still under development and it cannot really resolve larger areas that require more than just “figuring out a better alternative”.
Still, it will accelerate photo restoration that, if done manually, can take hours and hours of work. Moreover, the researchers say that it can also sharpen blurry or low-res photos more realistically by removing and replacing worn off pixels.
You can read the complete research for yourself here.