NVIDIA Shows Off Impressive AI Photo Reconstruction Abilities
Ron Perillo / 2 years ago
Content-Aware AI Technology from NVIDIA
Researchers at NVIDIA are introducing a state-of-the-art deep learning method for image reconstruction. It has the ability to restore corrupt images, even those that has holes or are missing pixels. NVIDIA is not the only company researching this kind of technology. In fact, Adobe, makers of Photoshop have their own content-aware fill feature. However, NVIDIA’s reconstruction techniques are remarkably close to accurate. So much so that it is even rivals manual reconstruction with human artists. Check the video out below to see it in action:
The paper published on arXiv is titled “Image Inpainting for Irregular Holes Using Partial Convulutions”. It uses deep learning technology which the researchers trained with 55,116 random masks of holes and streaks on the photos. By analyzing the image, the neural network is able to reconstruct missing pixels and gets better with each try. The researchers then used an additional 25,000 masks to validate the accuracy of their work.
How Well Does This Technology Work?
Here are some sample photos of reconstructed images. Note that this not only applies to human faces, but to other objects as well. That includes creatures, natural environments and even complex architectures. This technology has plenty of obvious applications, although there is no word yet on whether this will appear in a software program any time soon.