Computing technology firm NVIDIA has showcased its new AI-driven Content-Aware Fill tool for reconstructing corrupted images, and it would appear to give Adobe a run for its money.
NVIDIA, who’s technology is primarily developed for gaming, released the short video below, which explains the technology and shows off some impressive examples of what it can do.
The video coincides with a paper released by NVIDIA’s researchers entitled Image Inpainting for Irregular Holes Using Partial Convolutions. Snappy.
In short, NVIDIA explains that Photoshop’s Content-Aware Fill tool uses information from surrounding pixels to ascertain how gaps in an image should look. NVIDIA’s tool does this and takes it a step further, employing AI technology that then understands what the subject should look like.
As PetaPixel explains, if you removed the eyes from a portrait and ran Photoshop’s Content-Aware Fill tool, Adobe’s version would fill those gaps with eyebrows and other surrounding features.
NVIDIA’s tool would understand that the image is a face, and that the gaps should contain an eye. It then generates an eye to fill that empty space.
NVIDIA’s team says they trained their AI on 55,116 masks applied to a vast number of images, which their tool used to learn how to reconstruct missing pixels.
While the research team released the video and findings, they gave no indication of when this technology would make its way into actual software.