|
|
Leaf floating simulation based on GAN
FENG Qian-tai, YANG Meng, FU Hui
2020, 41(1):
27-34.
DOI: 10.11996/JG.j.2095-302X.2020010027
Inspired by the progress of deep learning algorithm, we propose a method based on the
generative adversarial network (GAN) to simulate the effect of leaf falling, in order to achieve
realistic leaf floating in the 3D rendering, virtual reality and other environments. Firstly, the training
data set is produced by experiment or in the field of 6-element time series of leaf falling under
different environmental conditions (wind level, disturbance level, etc.), and then the double-channel
auxiliary classifier generative adversarial networks model (DACGAN) is trained using the data set.
Secondly, the trained model can output 3D leaf data according to our demands (wind level,
disturbance level, etc.). Finally, the data can be used in a variety of graphic rendering environments.
Compared with the traditional mathematics-based method controlling the leaf floating and the
artificial animation method setting keyframes, our method, as the experimental results indicate, can
provide the possibility to capture the realism of the falling leaves. In addition, with the development
of computer vision and deep learning, the 3D data in reality is more easily acquired, the generated
model is more optimized, and the cost of DACGAN will be lower.
Related Articles |
Metrics
|
