Optimizing Latent Space Directions For GAN-based Local Image Editing
This work addresses a specific challenge in image editing for users needing precise control, though it appears incremental as it builds on existing GAN and segmentation methods.
The paper tackled the problem of ambiguity in GAN-based localized image editing by introducing a novel objective function to evaluate edit locality, resulting in a framework that is fast, highly disentangled, and applicable across datasets and GAN architectures.
Generative Adversarial Network (GAN) based localized image editing can suffer from ambiguity 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 Space Direction (LELSD), is applicable to any dataset and GAN architecture. Our method is also computationally fast and exhibits a high extent of disentanglement, which allows users to interactively perform a sequence of edits on an image. Our experiments on both GAN-generated and real images qualitatively demonstrate the high quality and advantages of our method.