StyleDisentangle: Disentangled Image Editing Based on StyleGAN2

Thanks to the development of Generative Adversarial Networks (GANs), StyleGAN2 can generate highly realistic images by inputting a latent code and then editing them in the latent space. Disentangled image editing is crucial, where the goal is to change the desired attributes of an image while keeping the other attributes intact. As a solution, we introduce the StyleDisentangle framework for image editing. The fundamental concept of StyleDisentangle is to define attributes through two distinct sets of information: semantic segmentation coordinates - identifying the region in the image related to the attribute, and latent code coordinates - identifying the dimensions related to attributes in latent code. By utilizing these two distinct sets of coordinates, we can precisely determine the position of each attribute within the attribute editing space, resulting in disentangled image editing. We conducted extensive experiments to demonstrate the effectiveness of our method on multiple datasets and additionally compared our results with state-of-the-art methods.