Research map / Image generation / Text-to-image diffusion
Synthetically Supervised Feature Learning for Scene Text Recognition
Yang Liu, Zhaowen Wang, Hailin Jin, Ian Wassell
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 ideal rendered clean image as auxiliary supervision (ablation) | CRNN · SVT | accuracy (%): 80.8 -> 87.0 | Improved |
| use binarized image as clean target instead of none (ablation) | CRNN · SVT | accuracy (%): 80.8 -> 85.8 | Improved |
| use deskewed image as clean target instead of none (ablation) | CRNN · SVT | accuracy (%): 80.8 -> 84.7 | Improved |
| use ideal image instead of binarized image as clean target (ablation) | synthetically supervised model · SVT | accuracy (%): 85.8 -> 87.0 | Improved |
| add image generation loss Lg (ablation) | CRNN · SVT | accuracy (%): 80.8 -> 86.1 | Improved |
| add feature consistency loss Lf (ablation) | CRNN · SVT | accuracy (%): 80.8 -> 85.1 | Improved |
| add image discriminator adversarial loss Lga to image generation (ablation) | Image Generation variant · SVT | accuracy (%): 86.1 -> 84.7 | Hurt |
| add feature discriminator adversarial loss Lfa plus Lg and Lf (ablation) | Feature Matching variant · SVT | accuracy (%): 85.1 -> 87.0 | Improved |
| add synthetic clean-image supervision (full model vs CRNN) on irregular text | CRNN · SVT-Perspective | accuracy (%): 66.8 -> 73.9 | Improved |
| add synthetic clean-image supervision (full model vs CRNN) on curved text | CRNN · CUTE80 | accuracy (%): 54.9 -> 62.5 | Improved |
| add synthetic clean-image supervision on lexicon-free IIIT5K | CRNN · IIIT5K (lexicon-free) | accuracy (%): 78.2 -> 89.4 | Improved |
Closest papers by meaning
- Revisiting Scene Text Recognition: A Data Perspective (ICCV 2023 · similarity 0.90)
- What if We Only Use Real Datasets for Scene Text Recognition? Toward Scene Text Recognition With Fewer Labels (CVPR 2021 · similarity 0.90)
- TextSSR: Diffusion-based Data Synthesis for Scene Text Recognition (ICCV 2025 · similarity 0.90)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.88)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.88)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.88)
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.
