Research map / Vision-language and multimodal / Human and embodied modeling
Symmetry-Constrained Rectification Network for Scene Text Recognition
Mingkun Yang, Yushuo Guan, Minghui Liao, Xin He, Kaigui Bian, Song Bai, Cong Yao, Xiang Bai
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 |
|---|---|---|---|
| add the symmetry-constrained rectification module on shared features (ablation) | recognition baseline · IC15 (SynthText only) | accuracy: 67.3 -> 68.8 | Improved |
| add the rectification module (ablation) | recognition baseline · SVTP (SynthText only) | accuracy: 66.5 -> 68.2 | Improved |
| add the rectification module (ablation) | recognition baseline · CUTE80 (SynthText only) | accuracy: 80.6 -> 81.9 | Improved |
| add the geometry loss without the rectification module (ablation) | recognition baseline · IC15 (SynthText only) | accuracy: 67.3 -> 67.0 | No clear effect |
| add the rectification module on top of the multi-loss variant (ablation) | multi-loss baseline · IC03 (SynthText only) | accuracy: 92.1 -> 91.2 | Hurt |
| add ScRN rectification (ablation) | recognition baseline · IC15 (Synth90k+SynthText) | accuracy: 76.9 -> 78.7 | Improved |
| add ScRN rectification (ablation) | recognition baseline · SVTP (Synth90k+SynthText) | accuracy: 77.7 -> 80.8 | Improved |
| add ScRN rectification (ablation) | recognition baseline · CUTE80 (Synth90k+SynthText) | accuracy: 84.4 -> 87.5 | Improved |
| replace the STN rectifier with ScRN (ablation) | STN_baseline · SVTP (Synth90k+SynthText) | accuracy: 79.4 -> 80.8 | Improved |
| replace the STN rectifier with ScRN (ablation) | STN_baseline · CUTE80 (Synth90k+SynthText) | accuracy: 85.8 -> 87.5 | Improved |
| replace the supervised STN rectifier with ScRN (ablation) | STN_supervision · IC15 (Synth90k+SynthText) | accuracy: 78.5 -> 78.7 | Improved |
| add extra supervision to the STN control points (ablation) | STN_baseline · CUTE80 (Synth90k+SynthText) | accuracy: 85.8 -> 86.5 | Improved |
Closest papers by meaning
- Robust Scene Text Recognition With Automatic Rectification (CVPR 2016 · similarity 0.89)
- ASTER: An Attentional Scene Text Recognizer with Flexible Rectification (TPAMI 2018 · similarity 0.87)
- SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition (AAAI 2021 · similarity 0.87)
- Arbitrary Shape Scene Text Detection With Adaptive Text Region Representation (CVPR 2019 · similarity 0.87)
- ESIR: End-To-End Scene Text Recognition via Iterative Image Rectification (CVPR 2019 · similarity 0.87)
- AON: Towards Arbitrarily-Oriented Text Recognition (CVPR 2018 · 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.
