Research map / Image generation / GANs
Region-Based Semantic Factorization in GANs
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 |
|---|---|---|---|
| Region-based factorization via generalized Rayleigh quotient (ReSeFa) vs. StyleSpace and LowRankGAN | StyleGAN2 face local editing · Close Eyes / Close Mouth / With Lipstick / Big Nose | ID similarity | Helped |
| Region-based factorization via generalized Rayleigh quotient (ReSeFa) vs. baselines | StyleGAN2 face local editing · Close Mouth / Close Eyes / Big Nose FID | FID | Helped |
| Region-based factorization via generalized Rayleigh quotient (ReSeFa) vs. baselines | StyleGAN2 face local editing · With Lipstick | FID | Hurt |
| Closed-form eigen-decomposition discovery vs. optimization-based LowRankGAN and StyleSpace annotation-based discovery | Semantic discovery speed · Face model discovery time | seconds per discovery: 0.5s vs 10.0s (StyleSpace) and 393s (LowRankGAN) | Helped |
| Region-based factorization vs. LowRankGAN on pixel preservation outside region | StyleGAN2 face local editing · With Lipstick / Big Nose MSE outside region | MSE outside region (x1e4) | Mixed |
| Region-based factorization vs. StyleSpace on pixel preservation outside region (eyes/mouth) | StyleGAN2 face local editing · Close Mouth / Close Eyes MSE | MSE outside region (x1e4) | Mixed |
| Ridge term tau*tr(J_b^T J_b) added to J_b^T J_b for non-singularity (tau=1e-3) | Background Jacobian in ReSeFa · method description | invertibility of J_b^T J_b | No clear effect |
| Sherman-Morrison-Woodbury low-rank formulation (diagonal-only operation on D) for solving with high-dimensional W+ latent space | ReSeFa discovery in W+ (9216-dim) · discovery runtime | speed | Helped |
Closest papers by meaning
- Closed-Form Factorization of Latent Semantics in GANs (CVPR 2021 · similarity 0.93)
- PandA: Unsupervised Learning of Parts and Appearances in the Feature Maps of GANs (ICLR 2023 · similarity 0.91)
- LARGE: Latent-Based Regression Through GAN Semantics (CVPR 2022 · similarity 0.89)
- Do Not Escape From the Manifold: Discovering the Local Coordinates on the Latent Space of GANs (ICLR 2022 · similarity 0.89)
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.89)
- Style Quantization for Data-Efficient GAN Training (CVPR 2025 · 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.