Relighting portraits using a neural network

Researchers from UC San Diego and Google have published a paper demonstrating an approach to relighting portraits using a neural network. The technique takes a single RGB image and is able to relight it using an environment map – no depth data is required. The paper suggests possible applications are for consumer apps that relight photographs taken with phones, but it could be very useful for archviz artists or others who often need to add 2D entourage to 3D renders and match the lighting. Watch a demo on YouTube and read the paper on arxiv.org.