Research map / Image generation / GANs
Analyzing and Improving the Image Quality of StyleGAN
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 |
|---|---|---|---|
| Replace AdaIN instance normalization with weight demodulation | StyleGAN generator · StyleGAN generated images | artifact prevalence | Helped |
| Replace AdaIN instance normalization with weight demodulation | StyleGAN generator · FID and precision/recall | FID | No clear effect |
| Remove normalization entirely (ablation) | StyleGAN generator · generated images | artifact prevalence | Helped |
| Perform R1 regularization once every 16 minibatches instead of every minibatch (ablation) | StyleGAN training · training | effectiveness of R1 regularization | No clear effect |
| Add path length regularization | StyleGAN2 generator · generator mapping | PPL | Helped |
| Add path length regularization (ablation) | StyleGAN2 generator · datasets less structured than FFHQ | FID | Mixed |
| Use spectral normalization in addition to the proposed contributions or instead of them (ablation) | StyleGAN2 generator · reported datasets | FID | Hurt |
| Use skip connections in the generator instead of the original feedforward generator (ablation) | StyleGAN generator · reported architecture configurations | PPL | Helped |
| Use a residual discriminator instead of the original feedforward discriminator (ablation) | StyleGAN discriminator · reported architecture configurations | FID | Helped |
| Use a residual architecture in the generator (ablation) | StyleGAN generator · reported architecture configurations | FID | Hurt |
| Disable progressive growing and use a skip generator with a residual discriminator | StyleGAN2 config e · reported benchmarks | FID | Helped |
| Double feature maps in resolutions 64²–1024² in both networks (ablation) | StyleGAN2 · LSUN | FID and recall | Helped |
Closest papers by meaning
- Content-Aware GAN Compression (CVPR 2021 · similarity 0.90)
- StyleSwin: Transformer-Based GAN for High-Resolution Image Generation (CVPR 2022 · similarity 0.90)
- Robust Unsupervised StyleGAN Image Restoration (CVPR 2023 · similarity 0.90)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.88)
- Style Quantization for Data-Efficient GAN Training (CVPR 2025 · similarity 0.88)
- WeditGAN: Few-Shot Image Generation via Latent Space Relocation (AAAI 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.