Research map / Image generation / Text-to-image diffusion
Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods
Xingsong Ye, Yongkun Du, Jiaxin Zhang, Haojie Zhang, Chong Sun, Chen Li, Jing LYU, 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 2M WATER-S synthetic data to real WATER-R training data (ablation) | WATERec · WordArt-Bench | accuracy (%): 88.55 -> 90.40 | Improved |
| add WATER-S to real data across architectures (ablation) | SVTRv2 / ABINet / SVTRv2-AR · WordArt-Bench | accuracy (%): 86.56 -> 88.68; 84.64 -> 87.03; 87.36 -> 90.14 | Improved |
| scale WATER-S from 2M to 3M (ablation) | WATERec · WordArt-Bench | accuracy (%): 90.40 -> 90.07 | No clear effect |
| combine 0.5M WATER-T and 0.5M WATER-Z instead of 1M of either (ablation) | WATERec · WordArt-Bench | accuracy (%): 89.94 vs 89.81 (T) and 89.41 (Z) | Improved |
| use WATER-T instead of SynthText or SynthTIGER (1M each) (ablation) | WATERec · WordArt-Bench | accuracy (%): 89.81 vs 88.95 and 88.55 | Improved |
| use WATER-Z instead of TextSSR (1M each) (ablation) | WATERec · WordArt-Bench | accuracy (%): 89.41 vs 88.09 | Improved |
| remove position encoding under arbitrary-shape input (ablation) | WATERec · WordArt-Bench | accuracy (%): 88.55 -> 49.57 | Hurt |
| use RoPE instead of APE or SPE with arbitrary-shape input (ablation) | WATERec · WordArt-Bench | accuracy (%): 87.69 (APE), 87.29 (SPE) -> 88.55 | Improved |
| allow arbitrary-shape input instead of fixed 256 tokens (with RoPE) (ablation) | WATERec · Union14M-Benchmark AVG | accuracy (%): 86.99 -> 88.14 | Improved |
| raise token range upper bound from 256 to 512 (ablation) | WATERec · WordArt-Bench | accuracy (%): 88.55 -> 88.82 | Improved |
| use arbitrary-shape input and AR decoder instead of fixed resolution | WATERec vs MAERec · Common benchmarks AVG | accuracy (%): 96.96 (MAERec) vs 96.69 | Hurt |
Closest papers by meaning
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.87)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.87)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.86)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.86)
- TextNeRF: A Novel Scene-Text Image Synthesis Method based on Neural Radiance Fields (CVPR 2024 · similarity 0.86)
- Layout-Agnostic Scene Text Image Synthesis with Diffusion Models (CVPR 2024 · similarity 0.86)
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.
