Research map / Image generation / GANs
In-Domain GAN Inversion for Real Image Editing
Jiapeng Zhu, Yujun Shen, Deli Zhao, Bolei Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train encoder through the generator on real images (domain-guided) instead of latent-space regression | encoder-based inversion · Image reconstruction (Face) | FID: 88.48 -> 52.85 | Improved |
| use domain-guided encoder as initialization and regularizer for optimization | MSE-based optimization · Image reconstruction (Face) | FID: 58.04 -> 42.64 | Improved |
| use domain-guided encoder as initialization and regularizer for optimization | MSE-based optimization · Image reconstruction (Tower) | FID: 69.16 -> 44.77 | Improved |
| use domain-regularized optimization instead of plain MSE optimization | StyleGAN face inversion · Image reconstruction (Face) | MSE: 0.026 -> 0.030 | Hurt |
| start optimization from the domain-guided encoder output | optimization-based inversion · Inference speed | time per image: 290s -> 8s | Improved |
| use in-domain inversion instead of MSE-based optimization | StyleGAN · Image interpolation (Face) | FID: 112.09 -> 91.18 | Improved |
| use in-domain inversion instead of MSE-based optimization | StyleGAN · Image interpolation (Tower) | FID: 121.38 -> 57.22 | Improved |
| use in-domain inversion instead of MSE-based optimization | StyleGAN · Image manipulation (Face) | FID: 83.69 -> 76.43 | Improved |
| use in-domain inversion instead of MSE-based optimization | StyleGAN · Image manipulation (Tower) | FID: 113 -> 57.92 | Improved |
| increase domain regularization weight lambda_dom from 0 to 40 (ablation) | domain-regularized optimization · Face reconstruction and manipulation | image quality | Mixed |
Closest papers by meaning
- In-Domain GAN Inversion for Faithful Reconstruction and Editability (TPAMI 2024 · similarity 0.98)
- Editing Out-of-Domain GAN Inversion via Differential Activations (ECCV 2022 · similarity 0.93)
- High-Fidelity GAN Inversion with Padding Space (ECCV 2022 · similarity 0.93)
- Diverse Inpainting and Editing with GAN Inversion (ICCV 2023 · similarity 0.93)
- GAN Inversion for Out-of-Range Images With Geometric Transformations (ICCV 2021 · similarity 0.93)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
