|Guest Speaker - Takato Yasuno, Yachiyo Engineering:
Generative Deep Learning for Monitoring Concrete Damage via CycleGAN
|13:30 – 13:50|
|Guest Speaker Q&A||13:50 – 14:00|
|MathWorks (incl. Q&A): Design, customize, and train advanced neural networks||14:00 – 15:00|
In the first part, Dr. Yasuno will show how a CycleGAN is used for the automated monitoring of concrete damage in dams. Supervised learning requires a great deal of dataset and annotation works, so it takes long time to accumulate data source. In contrast, unsupervised deep learning approach for anomaly detection have progressed. Using the difference between the real damaged image and the generated fake output without damage, it is possible to detect the interest feature automatically. This session presents an anomaly detection application using unpaired image-to-image translation mapping from a damaged image to a fake output without damage using a generator network. Actually, we apply our method to field studies for monitoring dam surface using drone flight images.
In the second part, we will take a deeper dive into designing, customizing, and training advanced neural networks. We will demonstrate MATLAB's extended deep learning framework, which enables you to implement advanced network architectures such as generative adversarial networks (GANs), variational autoencoders (VAEs), or Siamese networks.