Research map / Image generation / Text-to-image diffusion
Learning to Generate Stylized Handwritten Text via a Unified Representation of Style, Content, and Noise
Honglie Wang, Yan-Ming Zhang, Wangzi Yao, Fei Yin, Cheng-Lin Liu
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 |
|---|---|---|---|
| add aligned positional encoding (APE) over original PE (ablation) | InkSpire · IAM English lines | FID: 15.12 -> 9.31 | Improved |
| add rotation to APE (R-APE) (ablation) | InkSpire · IAM English lines | FID: 9.31 -> 7.92 | Improved |
| add aligned positional encoding (APE) (ablation) | InkSpire · ICDAR2013 Chinese lines | FID: 17.57 -> 11.75 | Improved |
| add rotation to APE (R-APE) (ablation) | InkSpire · ICDAR2013 Chinese lines | FID: 11.75 -> 10.98 | Improved |
| add R-APE on Chinese content accuracy (ablation) | InkSpire · ICDAR2013 Chinese lines | CR: 89.72 -> 92.92 | Improved |
| add R-APE on English Delta CER (ablation) | InkSpire · IAM English lines | Delta CER: 0.11 -> 0.01 | Improved |
| use random-size multi-region masking (R-Mask) instead of fixed top masking (ablation) | InkSpire · IAM English lines | FID: 8.73 -> 7.92 | Improved |
| use R-Mask instead of fixed top masking (ablation) | InkSpire · ICDAR2013 Chinese lines | FID: 11.57 -> 10.98 | Improved |
| use R-Mask instead of F-TopMask on Chinese CR (ablation) | InkSpire · ICDAR2013 Chinese lines | CR: 92.48 -> 92.92 | Improved |
| use frozen pretrained VAE instead of content encoder trained from scratch (ablation) | InkSpire · IAM English lines | FID: 20.48 -> 7.92 | Improved |
| use frozen VAE instead of content+style encoders from scratch (ablation) | InkSpire · IAM English lines | FID: 17.39 -> 7.92 | Improved |
| train content and style encoders from scratch (C+S-Enc) instead of frozen VAE (ablation) | InkSpire · ICDAR2013 Chinese lines | CR: 92.92 -> 94.74 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VATr — VATr++: Choose Your Words Wisely for Handwritten Text Generation (TPAMI 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Disentangling Writer and Character Styles for Handwriting Generation (CVPR 2023 · similarity 0.90)
- DiffInk: Glyph- and Style-Aware Latent Diffusion Transformer for Text to Online Handwriting Generation (ICLR 2026 · similarity 0.90)
- Decoupling Layout from Glyph in Online Chinese Handwriting Generation (ICLR 2025 · similarity 0.88)
- Zero-Shot Styled Text Image Generation, but Make It Autoregressive (CVPR 2025 · similarity 0.87)
- A Multi-Style Chinese Characters Writing Intelligent Tool Based on Small-scale Training Data (AAAI 2025 · similarity 0.86)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.85)
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.
