Research map / Image generation / GANs
Discovering Interpretable Latent Space Directions of GANs Beyond Binary Attributes
Check 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 positive target-attribute samples only instead of binary positive/negative labels | AdvStyle · Anime and human attribute editing | attribute disentanglement | Helped |
| Use separately learned attribute directions without an additional orthogonality constraint | AdvStyle · Multi-attribute editing | attribute disentanglement | Helped |
| Use positive-only adversarial direction learning instead of pseudo-binary labels | AdvStyle vs. InterFaceGAN · Non-binary style editing | style fidelity | Helped |
| Add the identity loss to adversarial direction learning (ablation) | AdvStyle · Open mouth editing | identity preservation | Helped |
| Add the distribution loss to adversarial direction learning (ablation) | AdvStyle · Open mouth editing | image detail preservation | Helped |
| Use adversarial loss alone, without identity or distribution losses (ablation) | AdvStyle · Open mouth editing | image quality and identity preservation | Hurt |
| Add layer-wise editing (ablation) | AdvStyle · Chinese celebrity style editing | skin color preservation | Helped |
| Use InterFaceGAN binary directions for non-binary styles | InterFaceGAN · Anime and human style editing | style fidelity | Hurt |
| Use InterFaceGAN directions for simultaneous attribute editing | InterFaceGAN · Multi-attribute face editing | attribute disentanglement | Hurt |
| Learn an involuntary negative direction while training on positive target examples | AdvStyle · Non-binary style editing | interpretability of reverse direction | Helped |
Closest papers by meaning
- Using latent space regression to analyze and leverage compositionality in GANs (ICLR 2021 · similarity 0.90)
- Interpreting the Latent Space of GANs for Semantic Face Editing (CVPR 2020 · similarity 0.90)
- GANSpace: Discovering Interpretable GAN Controls (NeurIPS 2020 · similarity 0.90)
- Low-Rank Subspaces in GANs (NeurIPS 2021 · similarity 0.89)
- Navigating the GAN Parameter Space for Semantic Image Editing (CVPR 2021 · similarity 0.89)
- NoiseCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions in Diffusion Models (CVPR 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.