قالب وردپرس درنا توس
Home / Technology / Nvidia develops AI that reconstructs damaged images

Nvidia develops AI that reconstructs damaged images



  Nvidia's New Autofill AI with Content-Conscious Function

Nvidia researchers have developed a new AI technology that can reconstruct holes or intentionally removed content from images.

The technique of "picture coloring" is similar to the "content" Autofill feature that Adobe added to Photoshop CS5 in the late 2000s, a technology that "magic" a deleted area of ​​an image based on the background of the photo. reconstructed.

Content-aware fills were one of the main features This has allowed Photoshop to retain the upper hand over all other image editors over the years.

Nvidia develops an algorithm inspired by the Photoshop feature

Now, Nvidia says its image injection technique could be "implemented in photo editing software" actions that can be performed using the Photoshop content auto-fill feature.

"Earlier deep learning approaches have focused on rectangular regions that are centered around the center of the image and often rely on post-processing," said the Nvidia research team. "Our model can handle holes of any shape, size, or distance from the edges of the image."

But the Nvidia researchers are not the first to develop such a technology. At the end of 201

7, a team of Yandex and Oxford researchers developed a similar algorithm called Deep Image Prior. Nvidia claims its algorithm is better.

Algorithm Can Handle Holes and Lines of Different Shapes and Sizes

Nvidia researchers say that this is because they trained their algorithm on 55,116 masks of random holes and lines of arbitrary shapes and sizes. These masks were used to delete content from original images. The algorithm was then shown the clipping image and its original shape to teach it how the missing pixels can be reconstructed.

The researchers verified the accuracy of their algorithm with another 25,000 masks, without showing the original image to the algorithm and reconstructing the missing (19659003) The research team published an impressive video of their new algorithm in action, in which a human operator actively extracts parts Deleted photos and the algorithm filled in missing pixels based on nearby content and mixed the new content seamlessly

More information on this technology can be found in a research paper entitled "Image Inpainting for Irregular Holes with Partial Convolution "

This is not the first AI technology Nvidia researchers have been working on. In November of last year, researchers created a tool that generates random human faces.


Source link