Generative Adversarial Networks (GANs) are deep learning architectures that generate high quality synthetic images of people, animals or objects. These networks have enabled developers to provide text-based prompts for generating realistic images, modifying existing images and completing missing information in training datasets.
Head over to the Microsoft Open Source Blog to learn about DragGAN, a type of GAN designed to update generated images with finer control over details and expressions. DragGAN enables minor tweaks to StyleGAN generated images without recreating the images again, resulting in efficient high quality image generation. The blog post describes the authors’ implementation of the DragGAN algorithm using ONNX Runtime. It also provides a C# example for integrating the DragGAN model into a native Windows application.
Source: Windows Blog