Research map / Label-efficient and robust learning / Semi-supervised learning
What if We Only Use Real Datasets for Scene Text Recognition? Toward Scene Text Recognition With Fewer Labels
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train on 11 real labeled datasets instead of MJ+ST synthetic data | CRNN · Six benchmark datasets (total) | accuracy: 74.8 vs. 75.8 | No clear effect |
| Train on 11 real labeled datasets instead of MJ+ST synthetic data | TRBA · Six benchmark datasets (total) | accuracy: 86.6% over 85.7% | Helped |
| Add data augmentation to real-labeled training (ablation) | CRNN · Six benchmark datasets (total) | accuracy: +5.2% | Helped |
| Add data augmentation to real-labeled training (ablation) | TRBA · Six benchmark datasets (total) | accuracy: +0.9% | Helped |
| Apply the best augmentation setting to synthetic-data training (ablation) | CRNN and TRBA Baseline-synth · Six benchmark datasets (total) | accuracy: 75.8 -> 73.4; 85.7 -> 85.2 | Hurt |
| Add Pseudo-Label (ablation) | CRNN with best augmentation · Six benchmark datasets (total) | accuracy: +2.8 | Helped |
| Add Pseudo-Label (ablation) | TRBA with best augmentation · Six benchmark datasets (total) | accuracy: +1.7 | Helped |
| Add Mean Teacher (ablation) | CRNN with best augmentation · Six benchmark datasets (total) | accuracy: -0.2 | Hurt |
| Add Mean Teacher (ablation) | TRBA with best augmentation · Six benchmark datasets (total) | accuracy: -0.4 | Hurt |
| Pretrain the feature encoder with RotNet (ablation) | CRNN with best augmentation · Six benchmark datasets (total) | accuracy: +1.3% | Helped |
| Pretrain the feature encoder with MoCo (ablation) | TRBA with best augmentation · Six benchmark datasets (total) | accuracy: -0.8% | Hurt |
| Combine Pseudo-Label with RotNet pretraining (ablation) | CRNN with best augmentation · Six benchmark datasets (total) | accuracy: +3.4% | Helped |
Closest papers by meaning
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.88)
- An Empirical Study of Scaling Law for Scene Text Recognition (CVPR 2024 · similarity 0.88)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.88)
- Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation (CVPR 2025 · similarity 0.87)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.87)
- Out of Length Text Recognition with Sub-String Matching (AAAI 2025 · 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.