Nvidia’s New Deep Learning Method Could Make Photoshop Obsolete

Many people like using Photoshop to edit their pictures considering how good the program is.  However, would fans of the program jump ship if there was another way to edit and reconstruct images with scary accuracy?  Well, according to Nvidia’s artificial intelligence research team, they have developed a deep learning method that could effectively make Photoshop become obsolete.

Source: blogs.nvidia.com

What is Deep Learning?

Writer Wajeeh Maaz has reported that Nvidia has come up with a way not only to compete with the popular photo editing program but developed a way to take an image, edit and reconstruct it with such accuracy that could make Photoshop become a thing of the past.  It was Nvidia’s research team that focuses on artificial intelligence that came up with a way to take a corrupted image and then edit or reconstruct it.  This new application utilizes what is known as deep learning to accomplish this goal; however, what exactly is deep learning?

Well, deep learning is a function of artificial intelligence which imitates the human brain’s workings in creating patterns and processing data that is used for making a decision.  The software can take parts of an image that has been modified or deleted to be replaced or in-painted with digital reconstructions that are based from the uncorrupted image that remains; the final products are scarily accurate.

Source: christopher5106.github.io

Nvidia Releases Published Whitepaper to Describe Their Product

Nvidia released a video that showed a variety of objects, from gigantic rocks to door panes and pillars, being erased entirely from images seamlessly that had what was left of the image to be intact.  Nvidia stated in a published whitepaper that this blending and natural look with the image that remains was an important element of what they desired to achieved.

The whitepaper said: “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 in-painting 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.”

However, there is more to recognize than developing a way to reconstruct corrupted images into basically accurate ones.  According to the project researchers, they consider themselves to be the first that trained successfully a neural network; this is a form of computing architecture somewhat modeled to represent the brain of a human so that it can process holes that are shaped irregular in images. 

Source: slideshare.net

Therefore, this would remove for graphic designers the need of working for hours to achieve natural, perfect results utilizing masking layers in tools like the Photoshop program.  This project is a proof-of-a-concept, just like was seen with Nvidia’s AI when it changed dogs into cats. 

Time will tell as far as how popular this concept will become and if one day Photoshop users would leave this photo editing tool behind for a product that Nvidia puts out which uses this deep learning method.  Those wishing to see Nvidia’s video that demonstrates this can see it here.