Research map / Image generation / Editing and style transfer
HyperInverter: Improving StyleGAN Inversion via Hypernetwork
Check 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 hypernetwork-based Phase II generator refinement (ablation) | HyperInverter · CelebA-HQ test (human faces) | L2: 0.0539 -> 0.0243 | Helped |
| use features from the initial reconstruction x̂_w in the appearance code (ablation) | HyperInverter Phase II · CelebA-HQ test (human faces) | L2: 0.0313 -> 0.0243 | Helped |
| use layer-wise appearance codes instead of a shared code (ablation) | HyperInverter Phase II · CelebA-HQ test (human faces) | L2: 0.0287 -> 0.0243 | Helped |
| set hypernetwork hidden dimension D to 64 rather than 256 (ablation) | HyperInverter · CelebA-HQ test (human faces) | L2: 0.0243 -> 0.0308 | Hurt |
| set hypernetwork hidden dimension D to 32 rather than 256 (ablation) | HyperInverter · CelebA-HQ test (human faces) | L2: 0.0243 -> 0.0378 | Hurt |
| set hypernetwork hidden dimension D to 128 rather than 256 (ablation) | HyperInverter · CelebA-HQ test (human faces) | L2: 0.0243 -> 0.0253 | Hurt |
| use HyperInverter instead of pSp | StyleGAN2 inversion · CelebA-HQ test (human faces) | L2: 0.0351 -> 0.0243 | Helped |
| use HyperInverter instead of PTI | StyleGAN2 inversion · CelebA-HQ test (human faces) | Time (s): 104 -> 0.0920 | Helped |
| use HyperInverter instead of PTI | StyleGAN2 inversion · LSUN Church test | FID: 76.21 -> 46.96 | Helped |
| use the full two-phase HyperInverter rather than its W encoder alone | HyperInverter editing · CelebA-HQ test (human faces), age edit γ=1 | age change: 6.55 -> 4.78 | Hurt |
| encode content in W space rather than using the full refined method | HyperInverter editing · CelebA-HQ test (human faces), age edit γ=1 | age change: 4.78 -> 6.55 | Helped |
| interpolate generator weights as well as latent codes | HyperInverter image interpolation · real-world image interpolation | visual interpolation quality | Helped |
Closest papers by meaning
- ReGANIE: Rectifying GAN Inversion Errors for Accurate Real Image Editing (AAAI 2023 · similarity 0.93)
- WarpGAN: Warping-Guided 3D GAN Inversion with Style-Based Novel View Inpainting (NeurIPS 2025 · similarity 0.89)
- Spatial-Contextual Discrepancy Information Compensation for GAN Inversion (AAAI 2024 · similarity 0.87)
- In-N-Out: Faithful 3D GAN Inversion with Volumetric Decomposition for Face Editing (CVPR 2024 · similarity 0.87)
- Dual Encoder GAN Inversion for High-Fidelity 3D Head Reconstruction from Single Images (NeurIPS 2024 · similarity 0.87)
- The Devil is in the Details: StyleFeatureEditor for Detail-Rich StyleGAN Inversion and High Quality Image Editing (CVPR 2024 · similarity 0.87)
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.