Current methods for hair rendering like the NVIDIA GameWorks HairWorks or lead to significant performance degradation.
Researchers from the University of southern California, Pinscreen and Microsoft have developed a technique for hair rendering, based on the methods of deep learning.
Researchers in AI believe that a convolutional neural network will help to implement this complex task. To train the neural network, the researchers gave her a dataset of 40 thousand different hairstyles and 160 thousand two-dimensional images from different angles. In the end, AI was able to create realistic three-dimensional visualization of different hair lengths, styles and colors, using only one 2D image. Neural network can also simulate the video and display the motion of individual hair strands that interact with each other.
Considering that in the future NVIDIA will support speed up machine learning tasks (AI Tensor Cores), the application of such methods for higher quality rendering of hair and other hard algorithmsuite tasks in games would be a logical development.
Current methods for hair rendering like the NVIDIA GameWorks HairWorks or TressFX AMD Hair lead to significant performance degradation.
© 2018, paradox. All rights reserved.
