Mobile Deep Learning with TensorFlow Lite, ML Kit and Flutter

Self-Attention Generative Adversarial Networks

We discussed an application of Generative Adversarial Networks (GANs) in Chapter 9, Building an Image Super-Resolution Application, where we generated high-resolution images from low-resolution images. GANs do a fairly good job of learning to mimic art and patterns. However, they fail to perform well where longer sequences need to be remembered and in instances where there are multiple parts of the sequence that are important toward making a generated output. Hence, we look toward Self-Attention GANs (SAGANs), introduced by Ian Goodfellow and his team, which are GAN systems that allow attention-driven, long-range dependency modeling for image-generation tasks. This system has had better performance ...