DIRS Meeting - Non-Physics based modeling and data set augmentation
Non-Physics based modeling and data set augmentation via Generative Adversarial Networks (GANs)Jason SloverPh.D. StudentChester F. Carlson Center for Imaging Science, RITAdvisor: Michael GartleyAbstract:One major limitation in training an object detection algorithm is the amount of curated data required to train it. It has already been shown for Electro-Optical images that augmenting a dataset with physics based synthetic data can improve performance. The opposite has been shown with Synthetic Aperture Radar images and a non-physics based approach has been recommended. I attempt to use CycleGAN and pix2pix generative adversarial networks in conjunction with DIRSIG simulated images to augment synthetic aperture radar images in the MSTAR dataset.Intended Audience:No background knowledge needed. All are welcome.
Event Snapshot
When and Where
Who
Open to the Public