Research map / Image generation / Image restoration
ESIR: End-To-End Scene Text Recognition via Iterative Image Rectification
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 |
|---|---|---|---|
| Use iterative rectification rather than no rectification (ablation) | ESIR · ICDAR2015, SVTP, CUTE | accuracy: 73.9 -> 76.9; 73.2 -> 79.6; 73.4 -> 83.3 | Helped |
| Increase rectification iterations from 1 to 5 (ablation) | ESIR · ICDAR2015, SVTP, CUTE | accuracy: 75.8 -> 76.9; 77.3 -> 79.6; 78.8 -> 83.3 | Helped |
| Increase line segments from 5 to 20 (ablation) | ESIR · ICDAR2015, SVTP, CUTE | accuracy: 75.8 -> 76.9; 78.0 -> 79.6; 81.7 -> 83.3 | Helped |
| Replace VGG with ResNet backbone | ESIR · ICDAR2013, ICDAR2015, IIIT5K, SVT, SVTP, CUTE | accuracy: 87.4 -> 89.1; 68.4 -> 70.1; 95.8 -> 97.8; 92.9 -> 96.1; 81.3 -> 82.9; 96.7 -> 97.1; 84.5 -> 85.9; 73.8 -> 75.8; 68.4 -> | Helped |
| Add SynthText training data to Synth90K | ESIR [ResNet] · ICDAR2013, ICDAR2015, IIIT5K, SVT, SVTP, CUTE | accuracy: 89.1 -> 91.3; 70.1 -> 76.9; 97.8 -> 99.6; 96.1 -> 98.8; 82.9 -> 93.3; 97.1 -> 97.4; 85.9 -> 90.2; 75.8 -> 79.6; 72.1 -> | Helped |
| Use ESIR rather than the strongest listed competing method | Scene text recognition · ICDAR2013 | accuracy: 93.3 vs 91.3 | Hurt |
| Use ESIR rather than the strongest listed competing method | Scene text recognition · IIIT5K (no lexicon) | accuracy: 93.4 vs 93.3 | Hurt |
| Use ESIR rather than the strongest listed competing method | Scene text recognition · SVTP | accuracy: 78.5 -> 79.6 | Helped |
| Use ESIR rather than the strongest listed competing method | Scene text recognition · CUTE | accuracy: 79.5 -> 83.3 | Helped |
| Use ESIR rather than the strongest listed competing method | Scene text recognition · IIIT5K (50-word lexicon), IIIT5K (1K-word lexicon), SVT (50-word lexicon) | accuracy: 99.6; 98.8; 97.4 | No clear effect |
| Use ESIR rather than the strongest listed competing method | Scene text recognition · ICDAR2015, SVT (no lexicon) | accuracy: 76.1 -> 76.9; 89.5 -> 90.2 | Helped |
Closest papers by meaning
- Robust Scene Text Recognition With Automatic Rectification (CVPR 2016 · similarity 0.88)
- Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition (AAAI 2019 · similarity 0.87)
- EAST: An Efficient and Accurate Scene Text Detector (CVPR 2017 · similarity 0.87)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.87)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.85)
- TextNeRF: A Novel Scene-Text Image Synthesis Method based on Neural Radiance Fields (CVPR 2024 · 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.