NVIDIA trained an AI to remove noise and watermarks from photos

If you thought Lightroom is doing a pretty good job of removing noise from your images, well then you are in for a treat. Researchers from NVIDIA, Aalto University and MIT developed a system to remove noise grain and even watermarks from images with AI. But while the concept of using AI to perform noise reduction isn’t new, this system has a rather unique starting point.

Unlike most existing systems which require to be trained by being shown example pairs of noisy and clean images, the AI here doesn’t need that. All this neural network needs are two noisy images and it will be able to remove artifacts, noise, grain and even automatically enhance your images.

The research paper goes on to claim that this AI performs on par with methods which require clean examples to work. The neural network was trained on 50 000 images in the ImageNet validation set, and runs on NVIDIA Tesla P100 GPUs with the cuDNN-accelerated TensorFlow deep learning framework.

Interestingly, the researchers presents use cases for the neural network which include enhancing MRI images, astronomical imaging and also physically-based rendering. While I believe enhancing MRI images with AI might result in false positives, denoising physically-based rendering is already happening with NVIDIA’s Turing-based GPUs. Would this technology further improve on what NVIDIA is already capable of?

Pokdepinion: Kinda interested if Adobe or any photo-editing software will harness this technology. I mean, we do have the AI chops in our PCs now with NVIDIA’s Turing GPUs bringing Tensor cores.