Research map / Image generation / Autoregressive and latent generation
Finding the Global Semantic Representation in GAN through Fréchet Mean
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 |
|---|---|---|---|
| Basis refinement by Fréchet mean on the GANSpace subspace (ablation) | GANSpace · StyleGAN2-FFHQ W-space | DCI: 0.312 -> 0.336 | Helped |
| Basis refinement by Fréchet mean on the GANSpace subspace (ablation) | GANSpace · StyleGAN2-FFHQ W-space | FID (perturbation 2): 8.90 -> 8.63 | Helped |
| Basis refinement by Fréchet mean on the SeFa subspace (ablation) | SeFa · StyleGAN2-FFHQ W-space | FID (perturbation 2): 11.79 -> 7.89 | Helped |
| Basis refinement by Fréchet mean on the SeFa subspace (ablation) | SeFa · StyleGAN2-FFHQ W-space | DCI: 0.307 -> 0.337 | Helped |
| Fréchet basis (Grassmann subspace Fréchet mean plus SO(n) basis Fréchet mean) instead of GANSpace | global semantic basis · StyleGAN2-FFHQ W-space | DCI and FID: DCI 0.350 vs 0.312; FID 7.49 vs 8.90 | Helped |
| Fréchet basis instead of Local Basis for local DCI | global semantic basis · Local DCI on W-space of StyleGAN2, StyleGAN2-e, StyleGAN1 | local DCI: StyleGAN2 0.740 vs 0.665; StyleGAN2-e 0.726 vs 0.704; StyleGAN1 0.733 vs 0.677 | Helped |
| Fréchet basis instead of GANSpace in intermediate layers | global semantic basis · StyleGAN2-FFHQ DCI across 6 intermediate layers | DCI: higher DCI in all layers except 5th layer of StyleGAN2-e | Helped |
| Extrinsic mean instead of Fréchet mean on Grassmannian for global subspace (ablation) | global semantic subspace · StyleGAN2-FFHQ, I=3.0 | DCI and FID | Hurt |
| Fréchet basis instead of GANSpace for robustness under latent traversal | global semantic basis · StyleGAN1, StyleGAN2-e, StyleGAN2 FFHQ FID | FID under traversal | Helped |
Closest papers by meaning
- Do Not Escape From the Manifold: Discovering the Local Coordinates on the Latent Space of GANs (ICLR 2022 · similarity 0.91)
- Region-Based Semantic Factorization in GANs (ICML 2022 · similarity 0.88)
- Closed-Form Factorization of Latent Semantics in GANs (CVPR 2021 · similarity 0.88)
- FEditNet: Few-Shot Editing of Latent Semantics in GAN Spaces (AAAI 2023 · similarity 0.87)
- Low-Rank Subspaces in GANs (NeurIPS 2021 · similarity 0.87)
- 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.