Image Generation from Brainwaves using Dual Generative Adversarial Training

Representing content of brainwaves captured through a non-invasive EEG device is of practical importance to various diagnostic applications. Although existing generative adversarial networks can obtain decent results in certain fields of neuroscience, performance declines significantly when representing content of brainwaves due to the insufficiency of real brainwaves. This lets us introduce a dual generative adversarial training paradigm to learn a smooth transition between brainwave and target image distributions. Our experiments demonstrate that our training approach outperforms the state-of-the-art training strategy on three benchmark datasets.