Research map / Image generation / GANs
StyleSwin: Transformer-Based GAN for High-Resolution Image Generation
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 style injection to the Swin baseline (ablation) | StyleSwin · FFHQ 256×256 | FID: 15.03 -> 8.40 | Helped |
| add double attention after style injection (ablation) | StyleSwin · FFHQ 256×256 | FID: 8.40 -> 7.86 | Helped |
| add a wavelet discriminator (ablation) | StyleSwin · FFHQ 256×256 | FID: 7.86 -> 6.34 | Helped |
| add sinusoidal positional encoding (ablation) | StyleSwin with a wavelet discriminator · FFHQ 256×256 | FID: 6.34 -> 5.76 | Helped |
| increase model size (ablation) | StyleSwin with SPE · FFHQ 256×256 | FID: 5.76 -> 5.50 | Helped |
| add balanced consistency regularization (ablation) | StyleSwin with a larger model · FFHQ 256×256 | FID: 5.50 -> 2.81; by 2.69 | Helped |
| replace the baseline with AdaIN style injection (ablation) | StyleSwin · FFHQ 256×256 | FID: 15.03 -> 6.34 | Helped |
| use AdaLN instead of AdaIN (ablation) | StyleSwin · FFHQ 256×256 | FID: 6.34 -> 6.95 | Hurt |
| use AdaBN instead of AdaIN (ablation) | StyleSwin · FFHQ 256×256 | FID: 6.34 -> $>$ 100 | Hurt |
| use AdaRMSNorm instead of AdaIN (ablation) | StyleSwin · FFHQ 256×256 | FID: 6.34 -> 7.43 | Hurt |
| use modulated MLP instead of AdaIN (ablation) | StyleSwin · FFHQ 256×256 | FID: 6.34 -> 7.09 | Hurt |
| use cross-attention instead of AdaIN (ablation) | StyleSwin · FFHQ 256×256 | FID: 6.34 -> 6.59 | Hurt |
Closest papers by meaning
- Styleformer: Transformer Based Generative Adversarial Networks With Style Vector (CVPR 2022 · similarity 0.91)
- Improved Transformer for High-Resolution GANs (NeurIPS 2021 · similarity 0.91)
- Analyzing and Improving the Image Quality of StyleGAN (CVPR 2020 · similarity 0.90)
- Generative Adversarial Transformers (ICML 2021 · similarity 0.90)
- Native-Resolution Image Synthesis (NeurIPS 2025 · similarity 0.87)
- Exploring DCN-like architecture for fast image generation with arbitrary resolution (NeurIPS 2024 · 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.