The last few years have seen considerable advances in artificial intelligence (AI) using technology inspired by animal brains. Neural networks and deep reinforcement learning allowed Google Deep Mind’s AlphaGo AI to beat champion Go player Lee Se-dol. The same methods are being used to teach AI to explore in Minecraft and win deathmatches in Doom.
Neural networks can even influence the art world. Google’s Deep Mind famously produces surreal art that took the internet by storm. We recently wrote about the Neural Doodle script that can recreate your own doodles in the styles of famous artists. The latest use of neural networks in imaging comes from researchers in Japan working on AI that can automatically colourise black and white photographs.
The AI, developed by researchers at Waseda University in Tokyo, can intelligently add colour to old black and white photographs by correctly identifying the contents of the image. While other AI need to be trained with similar photographs, this new AI is fully automatic and simply making uses of a large database of objects. The AI breaks down the target image into different parts it can label so it recognises the difference between things like buildings, trees, and people and colours them accordingly.
The results look incredible and some of the images used by the researchers are over 100 years old. Colourising older photos really brings them to life. The researchers have uploaded a gallery of before-and-after images so you can take a browse if you want to see just how powerful this new method is.
Main image © Satoshi Iizuka, Edgar Simo-Serra, and Hiroshi Ishikawa
Via The Verge