Research map / Image generation / GANs
In-Domain GAN Inversion for Faithful Reconstruction and Editability
Jiapeng Zhu, Yujun Shen, Yinghao Xu, Deli Zhao, Qifeng Chen, 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 |
|---|---|---|---|
| add w average to the encoder output (ablation) | Domain-guided encoder · FFHQ inversion (StyleGAN2) | FID: 27.95 -> 14.93 | Improved |
| add w average to the encoder output (ablation) | Domain-guided encoder · FFHQ inversion (StyleGAN2) | MSE: 0.074 -> 0.036 | Improved |
| add w average (ablation) | Domain-guided encoder · FFHQ interpolation (StyleGAN2) | FID: 70.69 -> 52.52 | Improved |
| retrain generator with enlarged W+ space (ablation) | Encoder · FFHQ reconstruction (StyleGAN) | MSE: 0.061 -> 0.053 | Improved |
| retrain generator with enlarged W+ space (ablation) | Encoder · FFHQ interpolation (StyleGAN) | FID: 44.97 -> 46.56 | Hurt |
| retrain generator with enlarged W+ space (ablation) | Encoder · FFHQ (StyleGAN2) | FID: reconstruction 15.30 -> 14.93, interpolation 50.46 -> 52.52 | Mixed |
| use ResNet-50 encoder instead of base encoder (ablation) | Domain-guided encoder · FFHQ inversion | SWD: 13.02 -> 7.31 | Improved |
| use MobileNet encoder instead of base (ablation) | Domain-guided encoder · FFHQ inversion | FID: 16.43 -> 17.11 | Hurt |
| use ResNet-34 instead of base encoder (ablation) | Domain-guided encoder · LSUN Church inversion | FID: 12.06 -> 11.18 | Improved |
| replace noise at later generator blocks (ablation) | Generator · FFHQ | FID: 41.89 -> 58.26 | Hurt |
| add in-domain optimization on top of domain-guided encoder | In-Domain Inversion · FFHQ inversion | SSIM: 0.522 -> 0.568 | Improved |
| domain-guided encoder vs Image2StyleGAN | In-Domain Inversion · FFHQ inversion | FID: 25.35 -> 15.07 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Image2StyleGAN — Image2StyleGAN++: How to Edit the Embedded Images? (CVPR 2020) · better on 7 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- In-Domain GAN Inversion for Real Image Editing (ECCV 2020 · similarity 0.98)
- Editing Out-of-Domain GAN Inversion via Differential Activations (ECCV 2022 · similarity 0.94)
- GAN Inversion: A Survey (TPAMI 2022 · similarity 0.94)
- High-Fidelity GAN Inversion with Padding Space (ECCV 2022 · similarity 0.93)
- Gradual Residuals Alignment: A Dual-Stream Framework for GAN Inversion and Image Attribute Editing (AAAI 2024 · similarity 0.90)
- 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.
