Latent Space Direction for Generative Adversarial Networks
Optimizing Latent Space Directions For GAN-based Local Image Editing
Localized image editing can sufferambiguity between semantic attributes.
We thus present a novel objective function to evaluate the locality of an image edit.
By introducing the supervision from a pre-trained segmentation network and optimizing the objective function, our framework, called locally effective latent spacedirection (lelsd), is applicable to any dataset and generative adversarial network architecture.
Our method is also computationally fast and exhibits a high extent of disentanglement, which allows users to interactively perform a sequence of image edits on an image.
Our experiments on both generated and real images qualitatively demonstrate the high quality and advantages of our method.