Research map / Image generation / Editing and style transfer
Improving Visual Quality of Image Synthesis by A Token-based Generator with Transformers
Yanhong Zeng, Huan Yang, Hongyang Chao, Jianbo Wang, Jianlong Fu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 StyleGAN2 style-based generator with TokenGAN | generator · FFHQ-256 | FID: 6.09 -> 5.41 | Improved |
| replace StyleGAN2 with TokenGAN | generator · FFHQ-1024 | FID: 5.20 -> 5.21 | No clear effect |
| replace StyleGAN2 with TokenGAN | generator · LSUN Church | FID: 5.88 -> 5.56 | Improved |
| replace StyleGAN2 with TokenGAN | generator · FFHQ-256 | Recall: 0.401 -> 0.447 | Improved |
| replace StyleGAN2 with TokenGAN | generator · LSUN Church | Precision: 0.587 -> 0.577 | Hurt |
| replace StyleGAN2 with TokenGAN | generator · FFHQ-1024 | Precision: 0.653 -> 0.651 | No clear effect |
| replace StyleGAN2 with TokenGAN | generator · FFHQ-256 inversion | MAE: 16.45 -> 13.43 | Improved |
| replace StyleGAN2 with TokenGAN | generator · FFHQ-256 inversion | LPIPS: 0.1539 -> 0.1238 | Improved |
| increase style tokens from 8 to 32 (ablation) | TokenGAN · FFHQ-256 | FID: 7.66 -> 6.81 | Improved |
| increase style tokens from 32 to 64 (ablation) | TokenGAN · FFHQ-256 | FID: 6.81 -> 7.60 | Hurt |
| increase content tokens from (16^2,16^2,32^2) to (64^2,64^2,128^2) (ablation) | TokenGAN · FFHQ-256 | FID: 18.54 -> 6.81 | Improved |
| increase content tokens from (32^2,32^2,64^2) to (64^2,64^2,128^2) (ablation) | TokenGAN · FFHQ-256 | FID: 15.69 -> 6.81 | Improved |
Closest papers by meaning
- ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis (NeurIPS 2024 · similarity 0.88)
- TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation (CVPR 2025 · similarity 0.88)
- Rethinking the Objectives of Vector-Quantized Tokenizers for Image Synthesis (CVPR 2024 · similarity 0.88)
- An Image is Worth 32 Tokens for Reconstruction and Generation (NeurIPS 2024 · similarity 0.88)
- OmniTokenizer: A Joint Image-Video Tokenizer for Visual Generation (NeurIPS 2024 · similarity 0.88)
- FlexTok: Resampling Images into 1D Token Sequences of Flexible Length (ICML 2025 · similarity 0.87)
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.
