Research map / Image generation / Super-resolution
Scene Text Image Super-resolution in the wild
Wenjia Wang, Enze Xie, Xuebo Liu, Wenhai Wang, Ding Liang, Chunhua Shen, 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 |
|---|---|---|---|
| replace SRResNet blocks with 5 sequential residual blocks (BLSTM) (ablation) | SRResNet · TextZoom (ASTER) | average accuracy: 51.3% -> 56.2% | Improved |
| add central alignment module (ablation) | 5xSRBs · TextZoom (ASTER) | average accuracy: 56.2% -> 57.8% | Improved |
| add gradient profile loss (ablation) | 5xSRBs + align · TextZoom (ASTER) | average accuracy: 57.8% -> 58.3% | Improved |
| add binary mask as input (ablation) | 5xSRB · TextZoom (ASTER) | accuracy easy/medium/hard: 73.9% -> 74.5%, 51.6% -> 53.3%, 36.0% -> 37.3% | Improved |
| train on real LR (TextZoom) instead of synthetic LR (ablation) | TSRN · TextZoom (ASTER) | average accuracy: 49.7% -> 58.3% | Improved |
| train on real LR instead of synthetic LR (ablation) | TSRN · TextZoom (CRNN) | average accuracy: 27.6% -> 41.4% | Improved |
| train on real LR instead of synthetic LR (ablation) | SRResNet · TextZoom (ASTER) | average accuracy: 48.9% -> 51.3% | Improved |
| use 32 hidden units in BLSTM instead of 0 (SRResNet) (ablation) | 5xSRBs · TextZoom | accuracy easy: 69.6% -> 74.5% | Improved |
| increase BLSTM hidden units from 32 to 128 (ablation) | 5xSRBs · TextZoom | accuracy easy: 74.5% -> 71.4% | Hurt |
| increase number of SRBs from 5 to 7 (ablation) | TSRN · TextZoom | accuracy easy: 74.5% -> 72.3% | Hurt |
| add central alignment on misaligned training data (ablation) | 5xSRB · TextZoom | average accuracy: 51.6% -> 57.4% | Improved |
| add central alignment module to SRResNet (ablation) | SRResNet · TextZoom | average accuracy: 51.7% -> 53.0% | Improved |
Closest papers by meaning
- Scene Text Telescope: Text-Focused Scene Image Super-Resolution (CVPR 2021 · similarity 0.91)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.86)
- Text-Aware Real-World Image Super-Resolution via Diffusion Model with Joint Segmentation Decoders (NeurIPS 2025 · similarity 0.85)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.84)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.84)
- Out of Length Text Recognition with Sub-String Matching (AAAI 2025 · 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.
