Nvidia, a well-known software maker, recently unveiled a new way to reconstruct images through artificial intelligence. This innovative computer vision method utilizes a process the company refers to as image inpainting. Through this technique, Nvidia can reconstruct images with missing pixels as well as remove unwanted content from a photo and substitute it with a computer-generated one.
In a recently published whitepaper, Nvidia’s researchers explained the difference between its innovative technique for inpainting images using deep learning and the existing technology. The researchers said that Nvidia’s new model can comfortably handle holes of any size location, shape or even distance, especially from the image borders. Previous deep learning methods have concentrated on rectangular areas located around the image’s central region. They also added that the existing approaches also depend on expensive post-processing. Furthermore, the company’s model comfortably handles holes of growing size.
For instance, when you apply Nvidia’s new technology on a blank space, where a nose is supposed to be, the software includes a nose that is computer-generated instead of filling the existing space with the surrounding.
In a given demonstration video, Nvidia‘s technology appeared not to suffer from similar issues affecting the current methods in the market, as far as filling in blank spaces in images is concerned. What’s more impressive about this new technique by the giant software maker is that it does not have blurred edges or granular degradation, which mainly calls for fiddling not only with different brushes but also with various levels of opacity and smoothness.
As a testament of Nvidia’s commendable work, the company’s researchers involved in developing the AI-driven technique said that Nvidia is the first company to train a neural network to handle irregular shaped blank spaces or holes in images successfully.
Nvidia’s AI technology can accomplish instantly what would take an experienced graphic designer minutes or hours to do. It works by using a deep neural network to develop partial convolutional predictions and masks. Simply put, it forms an invisible layer, which it alters until it is satisfied that the image is complete.
According to Nvidia’s researchers, the recent achievement does not mark the first time they have been amazed by one of the company’s AI projects. Last year, they were more impressed by Nvidia’s ability to develop realistic images of people who do not exist, and more importantly the capability of its AI to alter the time of day or weather in a video.
Although Nvidia is yet to announce the release date for the technology, it looks forward to its integration into today’s photo-editing software.
With reality becoming increasingly subjective, especially with the latest breakthrough, it looks like the only method to pinpoint the difference between real pictures and AI-generated imagery would be having individual pixels or the digital signature checked by a computer.
Founded in the 1990s, Nvidia develops graphics chips that that are used in video games, self-driving cars and in computers. Even so, the company has proved to be a force to reckon with in the artificial intelligence (AI) space. Last year, it made a debut in the Fortune 500 list at position #387.