Nvidia Taught an AI to Remove Artifacts in Photos... Including Watermarks.



Noise in images has been a problem ever since the invention of photography and Nvidia is hoping to do away with that. 

NVIDIA, along with researchers from Aalto University, and MIT have taken a Deep Learning approach to this problem. Deep learning, when broken down into the simplest terms is basically Artificial Intelligence algorithms that are based on the human brain. They are artificial neural networks that, unlike traditional machine learning, can automatically learn representations of data. Meaning, the AI can learn on its own what a chair is in a picture or video or what a ball is without being told what it is, that way when there's only part of a ball or a chair in the image... the machine can fix it. 

The researches had the AI study thousands of before and after photos, so it then can learn for itself what the final image is supposed to look like and create it on its own. The Team at Nvidia developing this technology states:

"There are several real-world situations where obtaining clean training data is difficult: low-light photography (e.g., astronomical imaging), physically-based rendering, and magnetic resonance imaging... Our proof-of-concept demonstrations point the way to significant potential benefits in these applications by removing the need for potentially strenuous collection of clean data. Of course, there is no free lunch - we cannot learn to pick up features that are not there in the input data - but this applies equally to training with clean targets."

This technology is incredibly useful. Check out this MRI which are notorious for having grainy and hard to read images. Granted that MRI technology works as well as it does is a miracle of science, but what if each image could look as clear as the one on the far right? 

nvidia MRI grain removal.png

In developing this technology, they created a way to remove watermarks from photos, as watermarks, in a sense, are artifacts that don't belong in the original image. For the watermark feature to work, the program requires at least two versions of the same photo with two different watermarks, but with deep learning, who knows how long that will need to be criteria to work? 

People are already stealing and profiting from other peoples' work posted on the internet on a daily basis, this would make it even easier. The science behind it is beautiful and its potential implications could be very ugly.  Granted this is not something the public has access to... as of yet. 

Check out this full video by explaining the process.

 



Source: NVIDIA