Research map / Image generation / Editing and style transfer
Styleformer: Transformer Based Generative Adversarial Networks With Style Vector
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 the convolution-based generator with a style-conditioned Transformer generator | StyleGAN2-ADA generator · CIFAR-10 | FID: 2.92 -> 2.82 | Helped |
| Apply Linformer to self-attention | Styleformer · CelebA | FID: 3.92 to 3.36 | Helped |
| Apply Linformer to self-attention | Styleformer · CelebA | Memory per GPU: 14668MiB -> 5316MiB | Helped |
| Apply Linformer to self-attention | Styleformer · CelebA | Speed: 6.46 -> 4.93 | Helped |
| Apply Linformer to self-attention | Styleformer · LSUN church | FID: OOM -> 7.99 | Helped |
| Combine low-resolution Styleformer with high-resolution StyleGAN2 blocks | Styleformer-C · CLEVR | FID: 16.05 -> 11.67 | Helped |
| Combine low-resolution Styleformer with high-resolution StyleGAN2 blocks | Styleformer-C · Cityscapes | FID: 8.35 -> 5.99 | Helped |
| Use separate style vectors for input and value (ablation) | Styleformer attention style injection · CIFAR-10 ablation | FID: 10.27 -> 8.56 | Helped |
| Add the input style vector alongside the value style vector (ablation) | Styleformer attention style injection · CIFAR-10 ablation | FID: 11.40 -> 8.56 | Helped |
| Add the value style vector alongside the input style vector (ablation) | Styleformer attention style injection · CIFAR-10 ablation | FID: 11.01 -> 8.56 | Helped |
| Use the modified residual connection with demodulation (ablation) | Styleformer encoder · CIFAR-10 ablation | FID: 19.09 -> 8.56 | Helped |
| Use the modified residual connection rather than Residual A (ablation) | Styleformer encoder · CIFAR-10 ablation | FID: 14.70 -> 8.56 | Helped |
Closest papers by meaning
- StyleSwin: Transformer-Based GAN for High-Resolution Image Generation (CVPR 2022 · similarity 0.91)
- Generative Adversarial Transformers (ICML 2021 · similarity 0.87)
- Analyzing and Improving the Image Quality of StyleGAN (CVPR 2020 · similarity 0.87)
- StyleNeRF: A Style-based 3D Aware Generator for High-resolution Image Synthesis (ICLR 2022 · similarity 0.87)
- Compositional Transformers for Scene Generation (NeurIPS 2021 · similarity 0.87)
- StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesis (ICML 2023 · 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.