✨ We've just launched our NEW website design!

Learn More Here
News

NVIDIA Researchers Show Off Impressive AI Photo-Retouching

NVIDIA Researchers Show Off Impressive AI Photo-Retouching

Fixing Grainy Photos via Deep Learning

Researchers from NVIDIA, Aalto University and MIT are presenting their impressive deep learning research at Sweden this week. Specifically, at the International Conference on Machine Learning. Their work involves using deep learning-based approach to fixing photos by simply looking at examples of corrupted photos only.

NVIDIA Researchers Show Off Impressive AI Photo-Retouching

The recent and typical deep learning work in the field focuses on training a neural network to restore images by showing example pairs of noisy and clean images. The AI then learns how to make up the difference. In comparison, this new method differs because it only requires two input images with the noise or grain. That means even without ever being shown what the supposed noise-free image looks like, the AI can still remove artifacts, grain, noise and enhance photos.

The researchers use NVIDIA Tesla P100 GPUs with the cuDNN-accelerated TensorFlow deep learning framework. The team then fed their system 50,000 images in the Imagenet validation set.

What Useful Applications Does This Technology Have?

The obvious use would be for photo restoration which aids with de-aging or corrupted files. However, this method also has application in the medical field. Especially for enhancing MRI images and other medical imaging methods.

In certain situations where a low-light photography hinders getting  proper images, this method is also useful. “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.

NVIDIA Researchers Show Off Impressive AI Photo-Retouching

It does not just work with uniform noise either. The team even demonstrates removing water-marks which occupy most of the screen. Furthermore, this method allows for network training in minutes instead of days with data generated on-the-fly.

For more information, read the full research paper via ARXIV.org.

Related Articles

Back to top button
Close

Adblock Detected

Please consider supporting us by disabling your ad blocker!   eTeknix prides itself on supplying the most accurate and informative PC and tech related news and reviews and this is made possible by advertisements but be rest assured that we will never serve pop ups, self playing audio ads or any form of ad that tracks your information as your data security is as important to us as it is to you.   If you want to help support us further you can over on our Patreon!   Thank you for visiting eTeknix