Research map / Vision-language and multimodal / Human and embodied modeling
Adaptive Text Recognition through Visual Matching
Chuhan Zhang, Ankush Gupta, Andrew Zisserman
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 decoder on top of the visual similarity map (ablation) | visual matching model · FontSynth (train R) | CER (%): 49.0 -> 9.4 | Improved |
| add position encoding to similarity disambiguation (ablation) | visual matching model · FontSynth (train R) | CER (%): 11.8 -> 9.4 | Improved |
| add self-attention to similarity disambiguation (ablation) | visual matching model without position encoding · FontSynth (train R) | CER (%): 23.9 -> 11.8 | Improved |
| add aggregator embedding (ablation) | visual matching model · FontSynth (train R) | CER (%): 22.9 -> 9.4 | Improved |
| add aggregator embedding (ablation) | visual matching model · FontSynth (train R+B+L+I) | CER (%): 8.5 -> 5.6 | Improved |
| add position encoding (ablation) | visual matching model · FontSynth (train R+B+L+I) | CER (%): 7.9 -> 5.6 | Improved |
| add Omniglot-Seq to training data | visual matching model · FontSynth test | CER (%): 5.6 -> 3.5 | Improved |
| increase training fonts from 1 attribute (R) to 4 attributes | visual matching model · FontSynth test | CER (%): 9.4 -> 5.6 | Improved |
| use test-font glyphs instead of cross-font exemplars | visual matching model · FontSynth (train R) | CER (%): 11.0 -> 9.4 | Improved |
| add 6-gram language model | visual matching model · Google1000 English | CER (%): 3.1 -> 2.4 | Improved |
| add language model to Attn. Baek et al. | Attn. Baek et al. · Google1000 English | CER (%): 5.4 -> 5.4 | No clear effect |
| add Italic split to training fonts for CTC Baek et al. | CTC Baek et al. · FontSynth | CER (%): 10.4 -> 10.4 | No clear effect |
Closest papers by meaning
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning (NeurIPS 2025 · similarity 0.88)
- PP-OCRv5: A Specialized 5M-Parameter Model Rivaling Billion-Parameter Vision-Language Models on OCR Tasks (CVPR 2026 · similarity 0.87)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.87)
- Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition (CVPR 2025 · similarity 0.87)
- On the Generalization of Handwritten Text Recognition Models (CVPR 2025 · similarity 0.86)
- ASCIIEval: Benchmarking Models' Visual Perception in Text Strings via ASCII Art (ICLR 2026 · 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.
