Research map / Vision-language and multimodal / Human and embodied modeling
What's Wrong with Synthetic Data for Scene Text Recognition? A Strong Synthetic Engine with Diverse Simulations and Self-Evolution
Xingsong Ye, Yongkun Du, JiaXin Zhang, 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 |
|---|---|---|---|
| replace existing 2D synthetic data (36M) with UnionST-S (5M) | SVTRv2-AR · Union14M-Benchmark | average accuracy: 73.36 -> 83.00 | Improved |
| scale UnionST-S from 5M to 10M samples (ablation) | SVTRv2-AR · Union14M-Benchmark | average accuracy: 83.00 -> 83.29 | No clear effect |
| add pseudo-labeled corpus (UnionST-P) to UnionST-S (ablation) | SVTRv2-AR · Union14M-Benchmark | average accuracy: 83.00 -> 84.86 | Improved |
| first ISR iteration with high-confidence pseudo-labels (ablation) | UnionST-SP · Union14M-Benchmark | average accuracy: 84.86 -> 89.12 | Improved |
| second ISR iteration (ablation) | UnionST-SP · Union14M-Benchmark | average accuracy: 89.12 -> 89.81 | Improved |
| third ISR iteration (ablation) | UnionST-SP · Union14M-Benchmark | average accuracy: 89.81 -> 89.81 | No clear effect |
| add 290K manually annotated hard samples after ISR (ablation) | UnionST-SP + ISR · Union14M-Benchmark | average accuracy: 89.81 -> 91.23 | Improved |
| fine-tune on real data after UnionST-SP pre-training | SVTRv2-AR · Union14M-Benchmark | average accuracy: 84.86 -> 91.39 | Improved |
| fine-tune UnionST-S model on only 1% real labels | SVTRv2-AR · Union14M-Benchmark | average accuracy: 87.26 vs 87.22 (full real data) | Improved |
| add curved placement (ablation) | UnionST engine (0.5M) · Union14M Curve subset | accuracy: 19.83 -> 46.70 | Improved |
| add multi-oriented placement (ablation) | UnionST engine (0.5M) · Union14M-Benchmark | average accuracy: 47.35 -> 63.82 | Improved |
| add corpus augmentation (ablation) | UnionST engine (0.5M) · Union14M-Benchmark | average accuracy: 63.82 -> 70.60 | Improved |
Closest papers by meaning
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.87)
- Image Captioning with Multi-Context Synthetic Data (AAAI 2024 · similarity 0.86)
- Scaling Laws of Synthetic Images for Model Training ... for Now (CVPR 2024 · similarity 0.86)
- TextNeRF: A Novel Scene-Text Image Synthesis Method based on Neural Radiance Fields (CVPR 2024 · similarity 0.86)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.86)
- The Unmet Promise of Synthetic Training Images: Using Retrieved Real Images Performs Better (NeurIPS 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.
