Research map / Image generation / Text-to-image diffusion
Learn to Augment: Joint Data Augmentation and Network Optimization for Text Recognition
Canjie Luo, Yuanzhi Zhu, Lianwen Jin, Yongpan Wang
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 |
|---|---|---|---|
| Add learnable text augmentation to ASTER | ASTER · CT80 | word accuracy: 79.9 -> 84.4 | Improved |
| Add learnable text augmentation to ASTER | ASTER · SVT-P | word accuracy: 77.7 -> 79.2 | Improved |
| Add learnable text augmentation to ASTER | ASTER · IC15 | word accuracy: 75.8 -> 76.1 | Improved |
| Add custom fiducial-point augmentation | IAM recognizer · IAM unconstrained | WER: 19.12 -> 14.04 | Improved |
| Add custom fiducial-point augmentation | RIMES recognizer · RIMES unconstrained | WER: 13.83 -> 9.23 | Improved |
| Add learnable agent to random-direction augmentation (ablation) | Text recognizer · Real-50k | word accuracy: 63.4 -> 66.5 | Improved |
| Replace affine augmentation with custom text augmentation (ablation) | Text recognizer · Real-50k | word accuracy: 58.6 -> 63.4 | Improved |
| Replace rigid transformation with custom text augmentation (ablation) | Text recognizer · Real-50k | word accuracy: 58.7 -> 63.4 | Improved |
| Add learnable augmentation and agent rather than no augmentation (ablation) | Text recognizer · Syn-10k | word accuracy: 7.7 -> 21.7 | Improved |
| Set patch number to 3 instead of 2 (ablation) | Text augmentation · CT80 | word accuracy: 14.6 -> 16.3 | Improved |
| Set patch number to 3 instead of 2 (ablation) | Text augmentation · IIIT5K | word accuracy: 29.8 -> 29.4 | Hurt |
| Increase moving radius from 5 to 10 (ablation) | Text augmentation · IIIT5K | word accuracy: 20.3 -> 29.4 | Improved |
Closest papers by meaning
- AON: Towards Arbitrarily-Oriented Text Recognition (CVPR 2018 · similarity 0.86)
- TextAdaIN: Paying Attention to Shortcut Learning in Text Recognizers (ECCV 2022 · similarity 0.84)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.84)
- On the Generalization Effects of Linear Transformations in Data Augmentation (ICML 2020 · similarity 0.84)
- Learning Data Manipulation for Augmentation and Weighting (NeurIPS 2019 · similarity 0.84)
- What Makes Better Augmentation Strategies? Augment Difficult but Not too Different (ICLR 2022 · similarity 0.84)
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.
