Research map / Image generation / GANs
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 latent directions | StyleGAN · StyleGAN latent edits | separation of pose and appearance | Helped |
| Randomize principal coordinates instead of random coordinates | StyleGAN2 · StyleGAN2 Cats | control of pose versus appearance | Helped |
| Apply a PCA direction only to selected synthesis layers | StyleGAN · StyleGAN2 FFHQ | edit disentanglement | Helped |
| Restrict the age-and-hairstyle direction to later layers | StyleGAN · StyleGAN2 FFHQ | lighting edit specificity | Helped |
| Apply BigGAN Skip-z inputs independently by layer | BigGAN512-deep · BigGAN image editing | layer-wise control | Helped |
| Use PCA in StyleGAN W space rather than PCA on feature tensors transferred to W (ablation) | StyleGAN · StyleGAN latent edits | usefulness of controls | Helped |
| Compute BigGAN PCA at the first linear layer rather than later layers (ablation) | BigGAN512-deep · BigGAN controls | usefulness of controls | Helped |
| Use random directions instead of PCA directions (ablation) | StyleGAN · StyleGAN latent edits | separation of pose and appearance | Hurt |
| Use early PCA components rather than later components (ablation) | StyleGAN and BigGAN · GAN image edits | control of geometric layout | Mixed |
| Apply the Wrinkles edit to a child face | StyleGAN2 FFHQ · FFHQ faces | wrinkle editing | No clear effect |
| Apply Makeup or Lipstick edits to male-presenting faces | StyleGAN2 FFHQ · FFHQ faces | makeup editing | No clear effect |
| Try to combine masculine and child attributes | StyleGAN2 FFHQ · FFHQ faces | attribute combination feasibility | Hurt |
Closest papers by meaning
- Navigating the GAN Parameter Space for Semantic Image Editing (CVPR 2021 · similarity 0.91)
- Discovering Interpretable Latent Space Directions of GANs Beyond Binary Attributes (CVPR 2021 · similarity 0.90)
- Exploiting Spatial Dimensions of Latent in GAN for Real-Time Image Editing (CVPR 2021 · similarity 0.89)
- A Geometric Analysis of Deep Generative Image Models and Its Applications (ICLR 2021 · similarity 0.89)
- [Re] GANSpace: Discovering Interpretable GAN Controls (NeurIPS 2022 · similarity 0.88)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · 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.