Research map / Image generation / Editing and style transfer
Image2StyleGAN: How to Embed Images Into the StyleGAN Latent Space?
Rameen Abdal, Yipeng Qin, Peter Wonka
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 |
|---|---|---|---|
| initialize with mean latent instead of random (ablation) | embedding · Face images | loss L (x10^5): 0.351 -> 0.309 | Improved |
| initialize with mean latent instead of random (ablation) | embedding · Face images | distance to mean latent: 35.60 -> 30.67 | Improved |
| initialize with mean latent instead of random (ablation) | embedding · Dog images | loss L (x10^5): 0.845 -> 0.922 | Hurt |
| initialize with mean latent instead of random (ablation) | embedding · Cat images | loss L (x10^5): 0.740 -> 0.752 | No clear effect |
| initialize with mean latent instead of random (ablation) | embedding · Painting images | loss L (x10^5): 3.451 -> 3.530 | Hurt |
| initialize with mean latent instead of random (ablation) | embedding · Car images | loss L (x10^5): 1.269 -> 1.390 | Hurt |
| embed into extended latent space W+ instead of W (ablation) | StyleGAN embedding · Face and other images | qualitative embedding quality | Improved |
| add multi-layer VGG perceptual loss instead of single layer (ablation) | embedding loss · Five image classes | qualitative embedding quality | Improved |
| add perceptual loss to pixel-wise MSE loss (ablation) | embedding loss · Five image classes | qualitative embedding quality | Improved |
| apply perceptual loss on images resized to 256x256 (ablation) | embedding loss · 1024x1024 images | qualitative embedding quality | Improved |
| resample noise at each update step instead of keeping it constant (ablation) | embedding · Face and other images | qualitative embedding quality | Hurt |
| use pretrained FFHQ weights instead of random weights (ablation) | embedding · Face and other images | qualitative embedding quality | Improved |
Closest papers by meaning
- Delving StyleGAN Inversion for Image Editing: A Foundation Latent Space Viewpoint (CVPR 2023 · similarity 0.91)
- Image2StyleGAN++: How to Edit the Embedded Images? (CVPR 2020 · similarity 0.91)
- StyleSpace Analysis: Disentangled Controls for StyleGAN Image Generation (CVPR 2021 · similarity 0.89)
- Warping the Residuals for Image Editing with StyleGAN (IJCV 2024 · similarity 0.89)
- Encoding in Style: A StyleGAN Encoder for Image-to-Image Translation (CVPR 2021 · similarity 0.89)
- The Devil is in the Details: StyleFeatureEditor for Detail-Rich StyleGAN Inversion and High Quality Image Editing (CVPR 2024 · 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.
