People鈥檚 hair in animated movies and video games could start to look far more realistic, thanks to artificial intelligence.
For decades, hair in video games and animated movies has looked unnatural because of the complexity of modelling its movement. 鈥淎lmost all works that exist today consider hair as a mesh,鈥� says Vanessa Sklyarova at the in Moscow, Russia. The graphical texture is then laid on top of this mesh, she says.
Advertisement
Sklyarova and her colleagues have come up with a way to make hair look natural by using a type of artificial intelligence called a neural network.
Their system takes smartphone photos of someone鈥檚 hair from multiple angles while they are stood still. The images are then processed in two stages. The first stage scans everything to identify where the head, shoulders and hair are, providing a rough outline.
The second stage uses a neural network trained on 343 images of hair styles to try to more accurately render the individual strands. While the resultant digital representation of the head (see video, above) shows the hair as multicoloured, this is to demonstrate how the neural network perceives the individual strands of hair. When the imagery is finalised, or rendered, the hair is its true colour. The technique also worked when using a smartphone video circling someone鈥檚 head.
鈥淚 just think that this is a great tool that could be used for character creation,鈥� says Sklyarova. 鈥淚t could simplify work a lot.鈥�
Shaping the hair digitally isn鈥檛 quick, though. It takes three days on a standard computer graphics card that costs around 拢1500 ($1900): one day for the first stage, and two more for the second.
鈥淭he results look really nice, and they鈥檙e solving a real problem, as anyone who has played a game 鈥� or watched a computer-animated movie 鈥� with atrociously animated hair can attest to,鈥� says at New York University. 鈥淗air is intrinsically hard to capture and render. It鈥檚 just a very different shape to the rest of the body because of the many small strands.鈥�
arXiv
