Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Learning for Text Recognition: A Critical Survey
Carlos Penarrubia, Jose J. Valero-Mas, Jorge Calvo-Zaragoza
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 |
|---|---|---|---|
| use SSL-pretrained ViT (SSM-Small) instead of CNN SSL (SimAN) | self-supervised text recognition · STR average WAcc | WAcc: 77.3 -> 93.8 | Improved |
| scale DiG from Tiny to Base | DiG · STR average WAcc | WAcc: 90.6 -> 92.9 | Improved |
| scale CCD from Tiny to Base | CCD · STR average WAcc | WAcc: 92.6 -> 94.0 | Improved |
| use attention decoder instead of CTC in PerSec-ViT | PerSec-ViT · STR average WAcc | WAcc: 76.3 -> 79.2 | Improved |
| use attention decoder instead of CTC in PerSec-CNN | PerSec-CNN · STR average WAcc | WAcc: 74.9 -> 77.6 | Improved |
| increase pre-training data (Text-DIAE 4M versus DiG/CCD 32.77M) | MIM pre-training · IIIT | WAcc: 86.1 -> 96.7 | Improved |
| add SSL pre-training before fine-tuning on real data versus from scratch (ablation) | text recognizers · STR average, ARD real data fine-tuning | WAcc | Improved |
| add SSL pre-training before fine-tuning on synthetic data versus from scratch (ablation) | text recognizers · STR average, synthetic data fine-tuning | WAcc | Improved |
Closest papers by meaning
- A Survey on Self-Supervised Learning: Algorithms, Applications, and Future Trends (TPAMI 2024 · similarity 0.90)
- Self-Supervised Learning of Graph Neural Networks: A Unified Review (TPAMI 2022 · similarity 0.88)
- TextAdaIN: Paying Attention to Shortcut Learning in Text Recognizers (ECCV 2022 · similarity 0.88)
- Self-Supervised Character-to-Character Distillation for Text Recognition (ICCV 2023 · similarity 0.87)
- Self-Distillation Regularized Connectionist Temporal Classification Loss for Text Recognition: A Simple Yet Effective Approach (AAAI 2024 · similarity 0.85)
- Clustering Properties of Self-Supervised Learning (ICML 2025 · similarity 0.85)
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.
