GAN‐based Data Augmentation

• Bowles et al. describe GANs as a way to ‘unlock’ additional information from a dataset
• Another useful strategy for generative modeling worth mentioning is variational auto-encoders. The GAN framework can be extended to improve the quality of samples produced with variational auto-encoders
• Using CycleGANs to translate images from the other 7 classes into the minority classes was very effective in improving the performance of the CNN model on emotion recognition.
• As exciting as the potential of GANs is, it is very difficult to get high-resolution outputs from the current cutting-edge architectures. Increasing the output size of the images produced by the generator will likely cause training instability and non-convergence