This article presents a new image inversion architecture that extracts high-rate latent features for image editing with StyleGAN models. The authors discuss the challenges of image editing with pretrained GAN models due to the trade-off between image reconstruction fidelity and edit quality. The proposed architecture includes a flow estimation module to adapt these features to edits, providing both high-fidelity to the input image and high-quality edits. The method shows significant improvements over state-of-the-art inversion methods.
Publication date: 19 Dec 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2312.11422
