Research map / Image generation / Text-to-image diffusion
TextSSR: Diffusion-based Data Synthesis for Scene Text Recognition
Xingsong Ye, Yongkun Du, Yunbo Tao, Zhineng Chen
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 region-centric processing (ablation) | TextSSR generator · IC13 (ablation) | SeqAcc (%): 47.55 -> 82.99 | Improved |
| add character-level position prior (ablation) | TextSSR generator · IC13 (ablation) | SeqAcc (%): 77.21 -> 82.99 | Improved |
| add VAE fine-tuning (ablation) | TextSSR generator · IC13 (ablation) | SeqAcc (%) / FID-R: SeqAcc 80.26 -> 82.99; FID-R 45.83 -> 50.76 | Mixed |
| add character-level glyph prior (ablation) | TextSSR generator · IC13 (ablation) | SeqAcc (%): 76.23 -> 77.21 | Improved |
| anagram-based generation instead of equal-length substitution (ablation) | TextSSR data · 30k data, CRNN, six benchmarks avg | SeqAcc (%): 18.95 -> 50.33 | Improved |
| scale TextSSR data from 30k to 240k (ablation) | CRNN · six STR benchmarks avg | SeqAcc (%): 50.33 -> 64.44 | Improved |
| double data to 60k (x2) (ablation) | CRNN · six STR benchmarks avg | SeqAcc (%): 50.33 -> 50.29 | No clear effect |
| increase ST rendered data from 4m to 6.98m | CRNN · six STR benchmarks avg | SeqAcc (%): 81.90 -> 82.06 | No clear effect |
| add 1.05m TextSSR to ST | CRNN · six STR benchmarks avg | SeqAcc (%): 81.90 -> 85.33 | Improved |
| add 3.55m TextSSR-F to ST + TextOCR | MAERec · six STR benchmarks avg | SeqAcc (%): 94.57 -> 95.46 | Improved |
| use TextSSR instead of SynthAdd for contextless words | CRNN · six STR benchmarks avg | SeqAcc (%): 81.99 -> 85.33 | Improved |
| add TextSSR to ST + TextOCR on SVTP | CRNN · SVTP | SeqAcc (%): +0.00 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports TextSSR as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Real In, Real Out: What If We Only Use Real Data for Scene Text Editing? (NeurIPS 2026 · better on 1 of 1 shared result)
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
- Layout-Agnostic Scene Text Image Synthesis with Diffusion Models (CVPR 2024 · similarity 0.91)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.90)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.90)
- TextNeRF: A Novel Scene-Text Image Synthesis Method based on Neural Radiance Fields (CVPR 2024 · similarity 0.89)
- Brush Your Text: Synthesize Any Scene Text on Images via Diffusion Model (AAAI 2024 · similarity 0.89)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 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.
