Research map / Vision-language and multimodal / Human and embodied modeling
Show, Attend and Read: A Simple and Strong Baseline for Irregular 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 |
|---|---|---|---|
| Use SAR instead of the strongest prior method | IC15 · IC15 None | accuracy: 76.1 -> 78.8 | Helped |
| Use SAR instead of the strongest prior method | SVTP · SVTP None | accuracy: 78.9% to 86.4% | Helped |
| Use SAR instead of the strongest prior method | CT80 · CT80 None | accuracy: 79.5% to 89.6% | Helped |
| halve channels in all convolutional layers except the last (ablation) | SAR · IC15 None | accuracy: 78.8 -> 75.6 | Hurt |
| reduce vertical down-sampling ratio from 1/8 to 1/16 (ablation) | SAR · IC15 None | accuracy: 78.8 -> 77.4 | Hurt |
| reduce vertical and horizontal down-sampling ratios to 1/16 and 1/8 (ablation) | SAR · IC15 None | accuracy: 78.8 -> 76.2 | Hurt |
| reduce horizontal down-sampling ratio from 1/4 to 1/8 (ablation) | SAR · IC15 None | accuracy: 78.8 -> 76.1 | Hurt |
| add neighborhood information to 2D attention (ablation) | SAR · IC15 None | accuracy: 77.2 -> 78.8 | Helped |
| use proposed 2D attention instead of 1D attention (ablation) | SAR · IC15 None | accuracy: 76.6 -> 78.8 | Helped |
| use two LSTM layers instead of one (ablation) | SAR · IC15 None | accuracy: 70.6 -> 78.8 | Helped |
| increase LSTM hidden state size from 256 to 512 (ablation) | SAR · IC15 None | accuracy: 76.8 -> 78.8 | Helped |
| add real-image training data to synthetic training data (ablation) | SAR · IC15 None | accuracy: 69.2 -> 78.8 | Helped |
Closest papers by meaning
- AON: Towards Arbitrarily-Oriented Text Recognition (CVPR 2018 · similarity 0.91)
- Robust Scene Text Recognition With Automatic Rectification (CVPR 2016 · similarity 0.90)
- An End-to-End TextSpotter With Explicit Alignment and Attention (CVPR 2018 · similarity 0.89)
- EAST: An Efficient and Accurate Scene Text Detector (CVPR 2017 · similarity 0.89)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.88)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · 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.