Research map / Image generation / Editing and style transfer
Diff-Font: Diffusion Model for Robust One-Shot Font Generation
Haibin He, Xinyuan Chen, Chaoyue Wang, Juhua Liu, Bo Du, Dacheng Tao, Yu Qiao
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 stroke count encoding condition (vs none) (ablation) | Diff-Font · Chinese font (small dataset) | FID: 28.83 -> 27.30 | Improved |
| use count encoding instead of one-bit stroke encoding (ablation) | Diff-Font · Chinese font (small dataset) | FID: 30.44 -> 27.30 | Improved |
| add one-bit stroke encoding condition (vs none) (ablation) | Diff-Font · Chinese font (small dataset) | FID: 28.83 -> 30.44 | Hurt |
| add one-bit stroke encoding condition (vs none) (ablation) | Diff-Font · Chinese font (small dataset) | SSIM: 0.74 -> 0.739 | Hurt |
| add stroke count encoding condition (vs none) (ablation) | Diff-Font · Chinese font (small dataset) | LPIPS: 0.127 -> 0.124 | Improved |
| raise content guidance scale s1 from 1 to 3 (s2=3 vs s2=1 fixed at 3) (ablation) | Diff-Font · Chinese font (large dataset) | FID: 16.67 -> 16.20 | Improved |
| raise stroke guidance scale s2 from 1 to 5 (s1=1) (ablation) | Diff-Font · Chinese font (large dataset) | FID: 16.67 -> 17.16 | Hurt |
| raise content scale from 3 to 5 (s2=1) (ablation) | Diff-Font · Chinese font (large dataset) | FID: 16.36 -> 16.18 | No clear effect |
| replace GAN image-to-image translation with conditional diffusion (vs DG-Font) | Diff-Font · Chinese font (small dataset) | FID: 43.44 -> 27.30 | Improved |
| replace GAN image-to-image translation with conditional diffusion (vs MX-Font) | Diff-Font · Chinese font (large dataset) | FID: 26.64 -> 16.20 | Improved |
| replace GAN methods with Diff-Font on Korean (vs DG-Font) | Diff-Font · Korean font | FID: 43.36 -> 10.69 | Improved |
| replace GAN methods with Diff-Font in human study (vs DG-Font) | Diff-Font · Chinese font human study | preference share: 24.53% -> 48.75% | 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.
- DG-Font — DG-Font: Deformable Generative Networks for Unsupervised Font Generation (CVPR 2021) · better on 7 of 7 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
- Generate Like Experts: Multi-Stage Font Generation by Incorporating Font Transfer Process into Diffusion Models (CVPR 2024 · similarity 0.92)
- FontDiffuser: One-Shot Font Generation via Denoising Diffusion with Multi-Scale Content Aggregation and Style Contrastive Learning (AAAI 2024 · similarity 0.92)
- Few-shot Font Generation with Localized Style Representations and Factorization (AAAI 2021 · similarity 0.91)
- VQ-FONT: Few-Shot Font Generation with Structure-Aware Enhancement and Quantization (AAAI 2024 · similarity 0.90)
- IF-Font: Ideographic Description Sequence-Following Font Generation (NeurIPS 2024 · similarity 0.90)
- DeepCalliFont: Few-Shot Chinese Calligraphy Font Synthesis by Integrating Dual-Modality Generative Models (AAAI 2024 · similarity 0.89)
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.
