This Algorithm Turns Photographs of Haircuts Into 3D Models
A new algorithm created by computer scientists at Zhejiang University could help make hair stylists' jobs a lot easier. As Fast Company reports, the program, called AutoHair [PDF], can construct 3D models of hairstyles from a single photograph.
Ph.D. student Menglei Chai and collaborators at the Chinese University’s State Key Lab of CAD&CG spent close to four years working on the algorithm. Animating hair is notoriously difficult, and the team used examples of haircuts from both video games and real-life to get the look right.
They started by analyzing 100,000 portraits from Flickr. By manually training the computer to recognize chunks of hair they were able to model individual strands within the segments and make better guesses as to what unseen parts of the hairstyles looked like.
This method alone resulted in models that looked flatter than what you'd find in reality, so they cross-referenced their data against hundreds of hair models from The Sims to add some extra body.
The models are meant to be seen more as plausible estimations than perfect recreations of real-life cuts. One obvious application for the technology is at the hair salon. Instead of working from one photograph, a 3D model could lead to more accurate cuts from stylists and less surprises for the client.
Chai also envisions the algorithm being used to model other features like entire faces down the road. There’s no shortage of apps out there that show you what you’d look like with a different hairstyle. Future versions of AutoHair could improve upon this idea by providing 3D previews from all angles.
[h/t Fast Company]
All images courtesy of Menglei Chai via YouTube.
Know of something you think we should cover? Email us at email@example.com.