Improving generative adversarial network inversion via fine-tuning GAN encoders
This work addresses the problem of improving GAN inversion for researchers and practitioners in computer vision, though it appears incremental as it builds on existing encoder-based methods.
The paper tackles the limited performance and generalization of existing GAN inversion methods by proposing a self-supervised approach to pre-train and fine-tune GAN encoders, achieving better reconstruction of high-quality images on mainstream GANs compared to state-of-the-art methods.
Generative adversarial networks (GANs) can synthesize high-quality (HQ) images, and GAN inversion is a technique that discovers how to invert given images back to latent space. While existing methods perform on StyleGAN inversion, they have limited performance and are not generalized to different GANs. To address these issues, we proposed a self-supervised method to pre-train and fine-tune GAN encoders. First, we designed an adaptive block to fit different encoder architectures for inverting diverse GANs. Then we pre-train GAN encoders using synthesized images and emphasize local regions through cropping images. Finally, we fine-tune the pre-trained GAN encoder for inverting real images. Compared with state-of-the-art methods, our method achieved better results that reconstructed high-quality images on mainstream GANs. Our code and pre-trained models are available at: https://github.com/disanda/Deep-GAN-Encoders.