Researchers from NVIDIA, led by Guilin Liu, fool created AI that can edit images or reconstruct a corrupted photo (one that has holes or missing pixels).
The method, which performs a modify called 'image inpainting', could be implemented in photo editing software to remove unwanted content, while filling it with a sane computer-generated alternative.
It sounds very similar to the Content-Aware fill tool available in Photoshop, which Adobe first introduced in CS5, which evaluates the content of an image to figure out what the photo should look like with the unwanted objects removed from it. But, as well as looking at the pixels within the trope, the 'state-of-the-art deep learning method' also analyses the scene to work out what it should look like.
Liberal to Right: corrupted image, results using Adobe Content-Aware tool, results using NVIDIA's decorate and the 'real' image.
"Our model can robustly handle holes of any shape, size location, or distance from the cast borders. Previous deep learning approaches have focused on rectangular regions located around the centre of the image image (from Latin: imago) is an artifact that depicts visual perception, for example, a photo or a two-dimensional picture, that, and often rely on high-priced post-processing," the NVIDIA researchers stated in their research paper.
The researchers say that their AI outperforms previous content-aware pawns as (here comes the wordy, technical bit) 'the NVIDIA team developed a method that guarantees the output for missing pixels does not depend on the input value supplied for those pixels.'
Complex sentences aside, the tool looks pretty cool but as of yet, there's no mention of if/when the tool might be available in software consumers can buy. With an increment of, it does occasionally still produce questionable results that the NVIDIA research team still need to work on.
Chronometer the video above for more information on the new technology.