Research map / Vision-language and multimodal / Human and embodied modeling
Image-to-Character-to-Word Transformers for Accurate Scene Text Recognition
Chuhui Xue, Jiaxing Huang, Wenqing Zhang, Shijian Lu, Changhu Wang, Song 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 C2W module on top of I2C (ablation) | I2C · CUTE (synthetic) | word accuracy: 81.2 -> 93.1 | Improved |
| add C2W module on top of I2C (ablation) | I2C · Total-Text (synthetic) | word accuracy: 71.0 -> 79.8 | Improved |
| add C2W module on top of I2C (ablation) | I2C · IC15 (synthetic) | word accuracy: 75.2 -> 82.8 | Improved |
| add C2W module on top of I2C (ablation) | I2C · IIIT5K (synthetic) | word accuracy: 89.2 -> 94.3 | Improved |
| predict relative character positions (I2C) instead of DETR 2D coordinates (ablation) | DETR · IC15 | word accuracy: 60.1 -> 75.2 | Improved |
| add C2W on top of I2C to beat DETR (ablation) | DETR · IC13 | word accuracy: 76.7 -> 95.0 | Improved |
| two-stage I2C2W instead of one-stage I2W (ablation) | I2W · CUTE | word accuracy: 87.8 -> 93.1 | Improved |
| two-stage I2C2W instead of one-stage I2W (ablation) | I2W · IC15 | word accuracy: 79.8 -> 82.8 | Improved |
| fine-tune with real images | I2C2W · IC15 | word accuracy: 82.8 -> 85.3 | Improved |
| fine-tune with real images | I2C2W · IIIT5K | word accuracy: 94.3 -> 94.5 | No clear effect |
| reduce N from 25 to 10 character candidates (ablation) | I2C2W · Total-Text | word accuracy: 79.8 -> 76.7 | Hurt |
| increase N from 25 to 100 character candidates (ablation) | I2C2W · IC13 | word accuracy: 95.0 -> 92.8 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RobustScanner — RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition (ECCV 2020) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.89)
- Scene Text Recognition from Two-Dimensional Perspective (AAAI 2019 · similarity 0.89)
- Char-Net: A Character-Aware Neural Network for Distorted Scene Text Recognition (AAAI 2018 · similarity 0.89)
- Instruction-Guided Scene Text Recognition (TPAMI 2025 · similarity 0.89)
- RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition (ECCV 2020 · similarity 0.88)
- Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing (NeurIPS 2024 · similarity 0.87)
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.
