What if you get an AI technology which could remove all the noise and artifacts from images, even the ones clicked in low-light? Great right?
Now, what if this tech also removes text and watermarks from the image? Not so great now, isn’t it?
It’s said that every technology can be put to good as well as bad use. But that shouldn’t stop innovation. In its latest image processing research, NVIDIA in a report has said that it has managed to train an AI which is able to remove noise from grainy photos through a deep-learning approach. It has partnered with Aalto University and MIT on this project.
Usually, the AI needs to look at both the noisy and the clean images to understand the difference between the two. But for this research the AI learned to fix the photos by simply looking at examples of corrupted photos only.
The AI was never shown what a noise-free imaged looked like, but it still managed to automatically understand how to enhance the photos.
“It is possible to learn to restore signals without ever observing clean ones, at performance close or equal to training using clean examples,” the researchers stated in their paper.“[The neural network] is on par with state-of-the-art methods that make use of clean examples — using precisely the same training methodology, and often without appreciable drawbacks in training time or performance.”
The team used NVIDIA Tesla P100 GPUs with the cuDNN-accelerated TensorFlow deep learning framework, the team trained their system on 50,000 images in the ImageNet validation set.
The AI is expected to be used in the medical field to enhance MRI images that usually need extensive post-processing to remove noise from the images.
Also, it can be applicable in cleaning up long exposure photos of the night sky by telescopes used in astrophotography.
NVIDIA’s AI can help reduce the post-processing time altogether. The question of whether it would increase the chances of image stealing still looms over our heads as the AI currently needs to look at two different versions of the watermarked photo.