AI Paper Atlas

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

CVPR 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded scene text and structure

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 changedWhereMeasuredEffect
replace existing 2D synthetic data (36M) with UnionST-S (5M)SVTRv2-AR · Union14M-Benchmarkaverage accuracy: 73.36 -> 83.00Improved
scale UnionST-S from 5M to 10M samples (ablation)SVTRv2-AR · Union14M-Benchmarkaverage accuracy: 83.00 -> 83.29No clear effect
add pseudo-labeled corpus (UnionST-P) to UnionST-S (ablation)SVTRv2-AR · Union14M-Benchmarkaverage accuracy: 83.00 -> 84.86Improved
first ISR iteration with high-confidence pseudo-labels (ablation)UnionST-SP · Union14M-Benchmarkaverage accuracy: 84.86 -> 89.12Improved
second ISR iteration (ablation)UnionST-SP · Union14M-Benchmarkaverage accuracy: 89.12 -> 89.81Improved
third ISR iteration (ablation)UnionST-SP · Union14M-Benchmarkaverage accuracy: 89.81 -> 89.81No clear effect
add 290K manually annotated hard samples after ISR (ablation)UnionST-SP + ISR · Union14M-Benchmarkaverage accuracy: 89.81 -> 91.23Improved
fine-tune on real data after UnionST-SP pre-trainingSVTRv2-AR · Union14M-Benchmarkaverage accuracy: 84.86 -> 91.39Improved
fine-tune UnionST-S model on only 1% real labelsSVTRv2-AR · Union14M-Benchmarkaverage accuracy: 87.26 vs 87.22 (full real data)Improved
add curved placement (ablation)UnionST engine (0.5M) · Union14M Curve subsetaccuracy: 19.83 -> 46.70Improved
add multi-oriented placement (ablation)UnionST engine (0.5M) · Union14M-Benchmarkaverage accuracy: 47.35 -> 63.82Improved
add corpus augmentation (ablation)UnionST engine (0.5M) · Union14M-Benchmarkaverage accuracy: 63.82 -> 70.60Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy