Research map / Image generation / Text-to-image diffusion
Zero-Shot Paragraph-level Handwriting Imitation with Latent Diffusion Models
Martin Mayr, Marcel Dreier, Florian Kordon, Mathias Seuret, Jochen Zöllner, Fei Wu, Andreas Maier, Vincent Christlein
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 HTR and writer-identification losses to the VAE (ablation) | latent diffusion VAE · IAM paragraph reconstruction | CER_P (%): 14.50 -> 3.19 | Improved |
| add HTR and writer-identification losses to the VAE (ablation) | latent diffusion VAE · IAM paragraph reconstruction | CER_L (%): 9.95 -> 2.29 | Improved |
| use trained compressed VAE instead of Stable Diffusion VAE (ablation) | latent diffusion · IAM paragraph reconstruction | CER_P (%): 5.27 -> 3.19 | Improved |
| add ranked resampling by writer identification (rank_WI) (ablation) | Ours · IAM, query synth / key genuine | mAP (%): 54.73 -> 61.06 | Improved |
| add ranked resampling by HTR (rank_HTR) (ablation) | Ours · IAM, query synth / key genuine | mAP (%): 54.73 -> 56.57 | Improved |
| add combined ranked resampling (rank_HTR+WI) (ablation) | Ours · IAM, query synth / key genuine | mAP (%): 54.73 -> 59.40 | Improved |
| add combined ranked resampling (rank_HTR+WI) (ablation) | Ours · IAM, query synth / key genuine | top-1 (%): 50.23 -> 54.66 | Improved |
| use cosine noise scheduler instead of linear (ablation) | Ours · IAM, query synth / key genuine | mAP (%): 59.40 -> 64.21; CER_L 4.95 -> 7.03 | Mixed |
| remove new-line tokens from the target text (ablation) | Ours · IAM, query synth / key genuine | CER_L (%): 4.95 -> 34.84 | Hurt |
| remove new-line tokens from the target text (ablation) | Ours · IAM, query synth / key genuine | mAP (%): 59.40 -> 66.44 | Mixed |
| our paragraph model instead of VATr | paragraph imitation · IAM, query synth / key genuine | mAP (%): 14.40 -> 54.73 | Improved |
| our model instead of best GAN baseline on OOD data | paragraph imitation · CVL test, query synth / key genuine | mAP (%): 8.37 -> 17.11 | Improved |
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 6 of 6 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
- DiffInk: Glyph- and Style-Aware Latent Diffusion Transformer for Text to Online Handwriting Generation (ICLR 2026 · similarity 0.89)
- One-Shot Diffusion Mimicker for Handwritten Text Generation (ECCV 2024 · similarity 0.89)
- Decoupling Layout from Glyph in Online Chinese Handwriting Generation (ICLR 2025 · similarity 0.87)
- Layout-Agnostic Scene Text Image Synthesis with Diffusion Models (CVPR 2024 · similarity 0.85)
- Z*: Zero-shot Style Transfer via Attention Reweighting (CVPR 2024 · similarity 0.85)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.84)
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.
