In "Wizard 3: hunting", "Island Fright 4", "call of Duty: the Ghost" and other triple-A games, you will often see the use of NVIDIA Hairworks / AMD TressFX
The most difficult to use digital virtual characters is to model the hair, because unlike simple changes in the face, the hair can have a very large shape, and there is a highly complex structure, and the volume and volume of the volume and volume of the collision are highly variable.
So the researchers tried to use
After using the convolution depth neural network, the speed of generating 300 thousand hair is 1000 times faster than the traditional method, and the hair model generated by this method has better local details, which seems to be more natural, and is better than the NVIDIA Hairworks and AMD TressFX techniques.
However, this technology may not be applied to the current graphics card because the test platform of the researchers is multiple NVIDIA.Titan XP graphics card, though much faster, is still too tight for a card. If the next generation of NVIDIA graphics cards introduce Tensor tensor units or performance increases, perhaps NVIDAI's new generation of AI based Hairworks can also be used.