Research map / Image generation / GANs
Toward a Visual Concept Vocabulary for GAN Latent Space
Sarah Schwettmann, Evan Hernandez, David Bau, Samuel Klein, Jacob Andreas, Antonio Torralba
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 layer-selective directions instead of random directions (ablation) | annotation direction generation · BigGAN-Places direction annotations | distinct 1-grams: 2,316 -> 3,156 | Improved |
| use layer-selective directions instead of GANSpace directions (ablation) | annotation direction generation · BigGAN-Places direction annotations | distinct 1-grams: 2,975 -> 3,156 | Improved |
| use layer-selective directions instead of random directions (ablation) | annotation direction generation · BigGAN-Places direction annotations | distinct 3-grams: 22,938 -> 31,307 | Improved |
| use layer-selective directions instead of GANSpace directions (ablation) | annotation direction generation · BigGAN-Places direction annotations | inter-annotator BERTScore-R: 0.343 -> 0.393 | Improved |
| use layer-selective directions instead of random directions (ablation) | annotation direction generation · BigGAN-Places direction annotations | inter-annotator BLEU: 8.86 -> 7.17 | Hurt |
| apply distilled directions to new z (same class) | distilled concept directions · BigGAN-Places | human accuracy (4-way): more than 60% | Improved |
| apply distilled directions to unseen classes instead of the same class | distilled concept directions · BigGAN-Places | human accuracy (4-way): .66 -> .39 | Hurt |
| compose pairs of distilled directions instead of single ones | distilled concept directions · BigGAN-Places | human accuracy (4-way): .66 -> .44 | Hurt |
Closest papers by meaning
- Unsupervised Discovery of Interpretable Directions in the GAN Latent Space (ICML 2020 · similarity 0.91)
- Interpretable Generative Adversarial Networks (AAAI 2022 · similarity 0.89)
- FEditNet: Few-Shot Editing of Latent Semantics in GAN Spaces (AAAI 2023 · similarity 0.89)
- LatentCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions (ICCV 2021 · similarity 0.89)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.88)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · 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.
