Research map / Image generation / GANs
Improving GAN Equilibrium by Raising Spatial Awareness
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 |
|---|---|---|---|
| Add spatial encoding (SEL) to the generator (ablation) | StyleGAN2 · LSUN Cat | DI: 3.64 -> 3.12 | Helped |
| Add discriminator-attention alignment loss to SEL (ablation) | StyleGAN2 + SEL · LSUN Cat | FID: 7.82 -> 6.81 | Helped |
| Add EqGAN-SA spatial awareness | StyleGAN2 · FFHQ 256×256 | FID: 3.66 -> 2.96 | Helped |
| Add EqGAN-SA spatial awareness | SN-DCGAN · CIFAR-10 unconditional | FID: 23.72 -> 16.93 | Helped |
| Replace structured heatmaps with random Gaussian noise (ablation) | StyleGAN2 + SEL + alignment · LSUN Cat | FID: 8.36 -> 8.31 | No clear effect |
| Remove hierarchical constraint from multiresolution heatmap sampling (ablation) | StyleGAN2 + SEL + alignment · LSUN Cat | FID: 8.36 -> 7.29 | Helped |
| Use hierarchical heatmap sampling instead of independent sampling (ablation) | StyleGAN2 + SEL + alignment · LSUN Cat | FID: 7.29 -> 6.81 | Helped |
| Flatten the heatmap and concatenate it with the latent code, losing its 2D structure (ablation) | StyleGAN2 · LSUN Cat | FID: 8.36 -> 8.63 | Hurt |
| Encode the heatmap with SEL_concat instead of flattening it (ablation) | StyleGAN2 · LSUN Cat | FID: 8.63 -> 7.02 | Helped |
| Use SEL_norm rather than SEL_concat (ablation) | StyleGAN2 · LSUN Cat | FID: 7.02 -> 6.81 | Helped |
Closest papers by meaning
- Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step (ICLR 2018 · similarity 0.87)
- GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium (NeurIPS 2017 · similarity 0.87)
- Your Local GAN: Designing Two Dimensional Local Attention Mechanisms for Generative Models (CVPR 2020 · similarity 0.86)
- Projected GANs Converge Faster (NeurIPS 2021 · similarity 0.86)
- Sequential Training of GANs Against GAN-Classifiers Reveals Correlated "Knowledge Gaps" Present Among Independently Trained GAN Instances (CVPR 2023 · similarity 0.86)
- GANSpace: Discovering Interpretable GAN Controls (NeurIPS 2020 · similarity 0.86)
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.