Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Implicit Glyph Attention for Text Recognition
Read the paperCheck 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 |
|---|---|---|---|
| add self-supervised glyph attention to a ViT-B recognizer | MGP-STR · MPSC | accuracy: 7.0% | Helped |
| add self-supervised glyph attention to a ViT-B recognizer | MGP-STR · ArbitText | accuracy: 10.3% | Helped |
| add self-supervised glyph attention to an SVTR recognizer | SVTR · MPSC and ArbitText | accuracy: 1.3% and 2.9% | Helped |
| add self-supervised glyph attention to a ResNet45 recognizer | PIMNet · MPSC and ArbitText | accuracy: 4.8% and 4.9% | Helped |
| add the joint GPC and GLAN structure (ablation) | SIGA_R baseline · six context benchmarks (average) | accuracy: 84.64 -> 91.65 | Helped |
| add the attention-based character fusion module (ablation) | SIGA_R with JS · six context benchmarks (average) | accuracy: 91.65 -> 92.10; 0.45% | Helped |
| add JS (ablation) | SRN · six context benchmarks (average) | accuracy: 86.04 -> 91.72 | Helped |
| add JS (ablation) | ABINet · six context benchmarks (average) | accuracy: 90.29 -> 91.63; 1.34% | Helped |
| use L_cor only instead of no alignment loss (ablation) | SIGA implicit attention alignment · TextSeg | attention overlap (Θ): 53.2 -> 55.2 | Helped |
| use L_dif only instead of no alignment loss (ablation) | SIGA implicit attention alignment · TextSeg | attention overlap (Θ): 53.2 -> 60.5 | Helped |
| combine L_cor and L_dif (ablation) | SIGA implicit attention alignment · TextSeg | attention overlap (Θ): 53.2 -> 63.6 | Helped |
| combine L_cor and L_dif (ablation) | SIGA implicit attention alignment · TextSeg | recognition accuracy: 69.1 -> 70.5 | Helped |
Closest papers by meaning
- GlyphSR: A Simple Glyph-Aware Framework for Scene Text Image Super-Resolution (AAAI 2025 · similarity 0.87)
- GlyphMastero: A Glyph Encoder for High-Fidelity Scene Text Editing (CVPR 2025 · similarity 0.86)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.84)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.84)
- Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation (CVPR 2025 · similarity 0.84)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.83)
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.