Research map / Image generation / GANs
Closed-Form Factorization of Latent Semantics in GANs
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 |
|---|---|---|---|
| Factorize pretrained generator weights instead of using GANSpace's sampled-data PCA | StyleGAN · FF-HQ | Re-scoring: 0.33 -> 0.38 | Helped |
| Factorize pretrained generator weights instead of using GANSpace's sampled-data PCA | StyleGAN · FF-HQ | User Study: 41% -> 59% | Helped |
| Factorize pretrained generator weights instead of using GANSpace's sampled-data PCA | StyleGAN · FF-HQ | FID: 7.43 -> 7.36 | Helped |
| Use unsupervised weight factorization rather than supervised attribute directions | PGGAN · CelebA-HQ, pose manipulation | Pose re-scoring: 0.53 -> 0.51 | No clear effect |
| Use unsupervised weight factorization rather than supervised attribute directions | PGGAN · CelebA-HQ, gender manipulation | Gender re-scoring: 0.59 -> 0.55 | No clear effect |
| Use unsupervised weight factorization rather than supervised attribute directions | PGGAN · CelebA-HQ, glasses manipulation | Glasses re-scoring: 0.24 -> 0.09 | Hurt |
| Use a top eigen-direction objective that prioritizes large projected variation | SeFa · GAN latent semantics, eyeglasses attribute | Ability to discover the attribute direction | Hurt |
| Interpret generator layers separately as bottom, middle, or top layers | StyleGAN · Cars | Semantic type controlled | Mixed |
| Use SeFa directions for editing across image categories | BigGAN · ImageNet | Cross-category edit applicability | Helped |
Closest papers by meaning
- Region-Based Semantic Factorization in GANs (ICML 2022 · similarity 0.93)
- LARGE: Latent-Based Regression Through GAN Semantics (CVPR 2022 · similarity 0.91)
- Interpreting the Latent Space of GANs for Semantic Face Editing (CVPR 2020 · similarity 0.91)
- Navigating the GAN Parameter Space for Semantic Image Editing (CVPR 2021 · similarity 0.91)
- NoiseCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions in Diffusion Models (CVPR 2024 · similarity 0.88)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 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.