Research map / Image generation / GANs
[Re] GANSpace: Discovering Interpretable GAN Controls
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 |
|---|---|---|---|
| Use PCA directions instead of random directions in StyleGAN latent space | StyleGAN and StyleGAN2 · GAN latent-space edits | separation of pose and appearance changes | Helped |
| Apply PCA to StyleGAN's W space instead of feature tensors and transfer directions to W (ablation) | StyleGAN · GAN latent-space edits | usefulness of controls | Helped |
| Compute BigGAN PCA at the first linear layer instead of later layers (ablation) | BigGAN512-deep · GAN latent-space edits | usefulness of controls | Helped |
| Apply a PCA direction only to selected StyleGAN layers rather than globally (ablation) | StyleGAN and StyleGAN2 · Layer-wise image edits | specificity of edits | Helped |
| Vary BigGAN Skip-z inputs separately by layer (ablation) | BigGAN512-deep · Layer-wise image edits | layer-wise controllability | Helped |
| Use BigGAN layer-wise Skip-z edits rather than StyleGAN layer-wise style edits | BigGAN512-deep · Layer-wise image edits | edit disentanglement | Hurt |
| Randomize the first eight PCA coordinates while fixing the remaining coordinates (ablation) | StyleGAN2 · Cat image edits | pose and camera-angle variation | Helped |
| Randomize the remaining PCA coordinates while fixing the first eight (ablation) | StyleGAN2 · Cat image edits | appearance variation with pose consistency | Helped |
| Replace PCA directions with random directions (ablation) | StyleGAN and StyleGAN2 · GAN latent-space edits | distinctiveness of selected directions | Hurt |
| Restrict the latent representation to the first 100 PCA components (ablation) | StyleGAN2 FFHQ · FFHQ | overall image appearance | No clear effect |
| Apply the wrinkles edit to a child's face | StyleGAN2 FFHQ · FFHQ | wrinkle-edit effect | No clear effect |
| Apply makeup or lipstick edits to male-presenting faces | StyleGAN2 FFHQ · FFHQ | makeup-edit effect | No clear effect |
Closest papers by meaning
- GANSpace: Discovering Interpretable GAN Controls (NeurIPS 2020 · similarity 0.88)
- Unsupervised Discovery of Interpretable Directions in the GAN Latent Space (ICML 2020 · similarity 0.84)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.84)
- [Re] Explaining in Style: Training a GAN to explain a classifier in StyleSpace (NeurIPS 2022 · similarity 0.84)
- PandA: Unsupervised Learning of Parts and Appearances in the Feature Maps of GANs (ICLR 2023 · similarity 0.83)
- Editing in Style: Uncovering the Local Semantics of GANs (CVPR 2020 · similarity 0.83)
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.