Research map / Image generation / GANs
ReStyle: A Residual-Based StyleGAN Encoder via Iterative Refinement
Yuval Alaluf, Or Patashnik, Daniel Cohen-Or
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 |
|---|---|---|---|
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Faces (1024) | runtime: 0.538 -> 0.451 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Faces (1024) | MSE: 0.14 -> 0.13 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Faces (1024) | LPIPS: 0.03 -> 0.03 | No clear effect |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Cars (512) | LPIPS: 0.08 -> 0.07 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Cars (512) | runtime: 0.411 -> 0.361 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · AFHQ Wild (512) | MSE: 0.23 -> 0.21 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · AFHQ Wild (512) | LPIPS: 0.06 -> 0.06 | No clear effect |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Churches (256) | MSE: 0.28 -> 0.26 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Churches (256) | runtime: 0.355 -> 0.298 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Horses (256) | MSE: 0.32 -> 0.31 | Improved |
| use simplified encoder (all styles from final 16x16 feature map) instead of FPN-based encoder (ablation) | ReStyle over pSp · Horses (256) | LPIPS: 0.09 -> 0.09 | No clear effect |
| train a dedicated residual iterative scheme (input plus current output, predict residual) instead of repeatedly feeding output of a pre-trained pSp/e4e encoder (ablation) | ReStyle over pSp · pSp and e4e | FFHQ faces and Stanford Cars reconstruction over steps | Improved |
Closest papers by meaning
- A Style-Based GAN Encoder for High Fidelity Reconstruction of Images and Videos (ECCV 2022 · similarity 0.92)
- HyperInverter: Improving StyleGAN Inversion via Hypernetwork (CVPR 2022 · similarity 0.91)
- In-Domain GAN Inversion for Faithful Reconstruction and Editability (TPAMI 2024 · similarity 0.90)
- In-Domain GAN Inversion for Real Image Editing (ECCV 2020 · similarity 0.90)
- Delving StyleGAN Inversion for Image Editing: A Foundation Latent Space Viewpoint (CVPR 2023 · similarity 0.90)
- ReGANIE: Rectifying GAN Inversion Errors for Accurate Real Image Editing (AAAI 2023 · similarity 0.89)
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.
