Research map / Image generation / GANs
Latent Transformations via NeuralODEs for GAN-Based Image Editing
Valentin Khrulkov, Leyla Mirvakhabova, Ivan Oseledets, Artem Babenko
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 nonlinear Neural ODE flow instead of InterFaceGAN linear shift (attribute control, Q1) | Ours(nonlinear) · FFHQ | human preference: +34% | Improved |
| use nonlinear Neural ODE flow instead of InterFaceGAN linear shift (attribute control, Q1) | Ours(nonlinear) · Places365 | human preference: +47% | Improved |
| use nonlinear Neural ODE flow instead of a trainable constant (linear) flow (Q1) (ablation) | Ours(nonlinear) · FFHQ | human preference: +10% | Improved |
| use nonlinear Neural ODE flow instead of a trainable constant (linear) flow (Q1) (ablation) | Ours(nonlinear) · Places365 | human preference: +48% | Improved |
| use nonlinear flow instead of IF for identity preservation (Q2) | Ours(nonlinear) · Places365 | human preference: -20% | Hurt |
| use nonlinear flow instead of IF for identity preservation (Q2) | Ours(nonlinear) · FFHQ | human preference: +4% | Improved |
| use nonlinear flow instead of linear Neural ODE for identity preservation (Q2) (ablation) | Ours(nonlinear) · Places365 | human preference: +31% | Improved |
| use nonlinear flow instead of linear Neural ODE for identity preservation (Q2) (ablation) | Ours(nonlinear) · FFHQ | human preference: +5% | Improved |
| replace the L2 attribute-preservation loss with a FaceNet cosine identity loss (ablation) | Ours(nonlinear) · FFHQ (13 attributes) | identity preservation preference: 60% vs 40% | Improved |
| increase Neural ODE depth from 1 to 2 (ablation) | Ours(nonlinear) · FFHQ / Places365 | visual quality | No clear effect |
| use nonlinear flow for small localized attributes such as Smiling or Bushy_Eyebrows | Ours(nonlinear) · FFHQ | visual quality | No clear effect |
| use nonlinear flow for global content attributes | Ours(nonlinear) · FFHQ/Places365 | Spearman correlation: ~0.41 | Improved |
Closest papers by meaning
- Interpreting the Latent Space of GANs for Semantic Face Editing (CVPR 2020 · similarity 0.94)
- Image-to-Image Translation With Disentangled Latent Vectors for Face Editing (TPAMI 2023 · similarity 0.93)
- Enjoy Your Editing: Controllable GANs for Image Editing via Latent Space Navigation (ICLR 2021 · similarity 0.92)
- L2M-GAN: Learning To Manipulate Latent Space Semantics for Facial Attribute Editing (CVPR 2021 · similarity 0.92)
- Surrogate Gradient Field for Latent Space Manipulation (CVPR 2021 · similarity 0.91)
- SDGAN: Disentangling Semantic Manipulation for Facial Attribute Editing (AAAI 2024 · similarity 0.89)
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.
