Quantcast
inpainting

Nvidia’s AI Tech Can Reconstruct Photos With Scary Accuracy

Nvidia's research team has developed a deep learning method to edit and reconstruct images that makes Photoshop seem obsolete.

Wajeeh Maaz

Want to remove your ex from your prom photos? Nvidia’s AI research team has got your back—the company has showcased a new application of deep learning that can edit or reconstruct corrupted images. The software allows for parts of images that have been deleted or modified to be ‘inpainted,’ or replaced with digital reconstructions based off of the remaining uncorrupted image. The results are scarily accurate.

The video released by Nvidia shows various objects, from pillars and door panes to gigantic rocks, being entirely erased from images seamlessly with the remainder of the image intact. In the published whitepaper, Nvidia stated that this natural look and blending with the remaining image was a key element of what they wanted to achieve.

“Previous deep learning approaches have focused on rectangular regions located around the center of the image, and often rely on expensive post-processing. The goal of this work is to propose a model for image inpainting that operates robustly on irregular hole patterns, and produces semantically meaningful predictions that incorporate smoothly with the rest of the image without the need for any additional post-processing or blending operation.”

According to the researchers behind the project, they’re the first to successfully train a neural network, a type of computing architecture loosely modeled after the human brain, to process irregular shaped holes in images. It removes the need for hours of work from graphic designers to achieve perfect, natural results with masking layers in tools like Photoshop. Much like Nvidia's AI that turned dogs into cats this project is just a proof-of-concept.