Research map / Image generation / GANs
On the "steerability" of generative adversarial networks
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 |
|---|---|---|---|
| Learn a self-supervised linear latent direction for image edits | BigGAN · ImageNet transformations (zoom, shift, color) | transformation fidelity | Helped |
| Increase transformation magnitude beyond the model's supported range | BigGAN latent walks · ImageNet transformations | image realism | Hurt |
| Use a non-linear trajectory instead of a linear walk (ablation) | BigGAN · ImageNet transformations | FID | Hurt |
| Use a non-linear trajectory instead of a linear walk (ablation) | BigGAN · ImageNet transformations | transformation strength: 0.59 vs. 0.49 (Luminance); 0.28 vs. 0.49 (Shift X); 0.39 vs. 0.55 (Shift Y); 0.37 vs. 0.60 (Zoom) | Mixed |
| Add data augmentation during training (ablation) | DCGAN · MNIST transformations | L2 reconstruction error | Helped |
| Add data augmentation during training (ablation) | DCGAN · MNIST transformations | structural integrity | Helped |
| Jointly optimize generator weights and the latent walk (ablation) | DCGAN · MNIST transformations | L2 reconstruction error | Helped |
| Apply data augmentation and jointly optimize the generator and walk (ablation) | DCGAN · CIFAR10 transformations | realistic transformation quality | Hurt |
| Use a linear trajectory instead of a non-linear trajectory (ablation) | BigGAN · ImageNet transformations | image realism | Helped |
| Train on data with broader variability for the transformed attribute | GAN class distributions · ImageNet classes | achievable transformation range | Helped |
Closest papers by meaning
- A Geometric Analysis of Deep Generative Image Models and Its Applications (ICLR 2021 · similarity 0.91)
- GAN "Steerability" without optimization (ICLR 2021 · similarity 0.90)
- Constrained Generative Adversarial Networks for Interactive Image Generation (CVPR 2019 · similarity 0.89)
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.89)
- AdaGAN: Boosting Generative Models (NeurIPS 2017 · similarity 0.89)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · 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.