Research map / Image generation / Editing and style transfer
ReGANIE: Rectifying GAN Inversion Errors for Accurate Real Image Editing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the rectifying network to the W-space editing and inversion pipeline | ReGANIE · CelebA-HQ reconstruction | L2: 0.019 -> 0.016 | Helped |
| Add the rectifying network to the W-space editing and inversion pipeline | ReGANIE · CelebA-HQ reconstruction | LPIPS: 0.08 -> 0.07 | Helped |
| Use the two-phase reconstruction-and-editing framework | Real image editing on CelebA-HQ · CelebA-HQ | Age identity similarity: 0.44 -> 0.62 | Helped |
| Use the two-phase reconstruction-and-editing framework | Real image editing on CelebA-HQ · CelebA-HQ | Yaw identity similarity: 0.52 -> 0.75 | Helped |
| Replace global-only modulation with spatial-only modulation (ablation) | ReGANIE rectifying network · Image reconstruction | reconstruction quality | Helped |
| Use spatial modulation only instead of combined spatial and global modulation (ablation) | ReGANIE rectifying network · Pose editing | editing artifacts | Hurt |
| Use global modulation only instead of combined spatial and global modulation (ablation) | ReGANIE rectifying network · Image reconstruction | detail recovery | Hurt |
| Combine spatial and global modulation (ablation) | ReGANIE rectifying network · Image reconstruction | reconstruction quality | Helped |
| Replace residual difference input ΔX with original image X as the rectifier condition (ablation) | ReGANIE rectifying network · Image reconstruction and pose editing | visual quality | Hurt |
| Replace residual difference input ΔX with inversion X̂ as the rectifier condition (ablation) | ReGANIE rectifying network · Image reconstruction | reconstruction similarity | Hurt |
| Train with the full rectifier input, including the residual difference (ablation) | ReGANIE rectifying network · Image reconstruction | reconstruction stability | Helped |
Closest papers by meaning
- HyperInverter: Improving StyleGAN Inversion via Hypernetwork (CVPR 2022 · similarity 0.93)
- The Devil is in the Details: StyleFeatureEditor for Detail-Rich StyleGAN Inversion and High Quality Image Editing (CVPR 2024 · similarity 0.91)
- Gradual Residuals Alignment: A Dual-Stream Framework for GAN Inversion and Image Attribute Editing (AAAI 2024 · similarity 0.90)
- In-N-Out: Faithful 3D GAN Inversion with Volumetric Decomposition for Face Editing (CVPR 2024 · similarity 0.89)
- Spatial-Contextual Discrepancy Information Compensation for GAN Inversion (AAAI 2024 · similarity 0.89)
- WarpGAN: Warping-Guided 3D GAN Inversion with Style-Based Novel View Inpainting (NeurIPS 2025 · similarity 0.88)
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.