README.md

---

license: mit
pipeline_tag: image-to-image
---


# The Devil is in the Details: StyleFeatureEditor for Detail-Rich StyleGAN Inversion and High Quality Image Editing (CVPR 2024)

<a href="https://arxiv.org/abs/2406.10601"><img src="https://img.shields.io/badge/arXiv-2404.01094-b31b1b.svg" height=22.5></a>
<a href="https://colab.research.google.com/#fileId=https://github.com/AIRI-Institute/StyleFeatureEditor/blob/main/notebook/StyleFeatureEditor_inference.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" height=22.5></a>
<a href="https://github.com/AIRI-Institute/StyleFeatureEditor"><img src="https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white" height=22.5></a>



> The task of manipulating real image attributes through StyleGAN inversion has been extensively researched. This process involves searching latent variables from a well-trained StyleGAN generator that can synthesize a real image, modifying these latent variables, and then synthesizing an image with the desired edits. A balance must be struck between the quality of the reconstruction and the ability to edit. Earlier studies utilized the low-dimensional W-space for latent search, which facilitated effective editing but struggled with reconstructing intricate details. More recent research has turned to the high-dimensional feature space F, which successfully inverses the input image but loses much of the detail during editing. In this paper, we introduce StyleFeatureEditor -- a novel method that enables editing in both w-latents and F-latents. This technique not only allows for the reconstruction of finer image details but also ensures their preservation during editing. We also present a new training pipeline specifically designed to train our model to accurately edit F-latents. Our method is compared with state-of-the-art encoding approaches, demonstrating that our model excels in terms of reconstruction quality and is capable of editing even challenging out-of-domain examples.
> 



<p align="center">
  <img src="assets/preview_horizontal.webp" width="100%" align="center"/> 
  <br>
SFE is able to edit a real face image with the desired editing. It first reconstructs (inverts) the original image and then edits it according to the chosen direction. On the left is an examples of how our method works for several directions with different editing power p. On the right we display a comparison with previous approaches. LPIPS (lower is better) indicates inversion quality, while FID (lower is better) indicates editing ability. The size of markers indicates the inference time of the method, with larger markers indicating a higher time.
</p>

<p align="center">
  <img src="assets/images_preview.webp" width="100%"/>
  <br>
Examples of how our method works on several real images.
</p>

This repository contains the pretrained weights for our method, the inference and training code can be found on GitHub: [https://github.com/AIRI-Institute/StyleFeatureEditor](https://github.com/AIRI-Institute/StyleFeatureEditor)