Research map / Architectures and efficiency / Attention and transformers
CoMER: Modeling Coverage for Transformer-Based Handwritten Mathematical Expression Recognition
Wenqi Zhao, Liangcai Gao
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 scale augmentation to BTTR training (ablation) | BTTR · CROHME 2014 test | ExpRate (%): 53.45 -> 55.17 | Improved |
| add self-coverage ARM to the transformer decoder (ablation) | BTTR (baseline) · CROHME 2014 test | ExpRate (%): 55.17 -> 57.51 (+2.34) | Improved |
| add cross-coverage ARM to the transformer decoder (ablation) | BTTR (baseline) · CROHME 2014 test | ExpRate (%): 55.17 -> 58.11 (+2.94) | Improved |
| combine self-coverage and cross-coverage (fusion) (ablation) | CoMER (Self) · CROHME 2014 test | ExpRate (%): 57.51 -> 59.33 | Improved |
| add self-coverage to cross-coverage CoMER (fusion) (ablation) | CoMER (Cross) · CROHME 2014 test | ExpRate (%): 58.11 -> 59.33 | Improved |
| replace vanilla transformer decoder with CoMER (ARM + fusion coverage) | BTTR (baseline) · CROHME 2016 test | ExpRate (%): 56.58 -> 59.81 | Improved |
| replace vanilla transformer decoder with CoMER (ARM + fusion coverage) | BTTR (baseline) · CROHME 2019 test | ExpRate (%): 59.55 -> 62.97 | Improved |
| replace vanilla transformer decoder with CoMER (ARM + fusion coverage) | BTTR (baseline) · CROHME 2014 test | ExpRate (%): 55.17 -> 59.33 | Improved |
| use CoMER instead of the best RNN coverage model (Ding et al.) | CoMER · CROHME 2014 test | ExpRate (%): 58.72 -> 59.33 | Improved |
| use CoMER instead of Ding et al. | CoMER · CROHME 2019 test | ExpRate (%): 61.38 -> 62.97 | Improved |
| use CoMER instead of Ding et al. on <=3 errors | CoMER · CROHME 2014 test | ExpRate with <=3 errors (%): 78.60 -> 77.89 | Hurt |
| add coverage ARM (Fusion) for long expressions (ablation) | BTTR (baseline) · CROHME 2014 test, sequences longer than 50 | accuracy by length: up to 5x BTTR (baseline) | Improved |
Later papers that compared with this method
2 later accepted papers report CoMER as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- TAMER: Tree-Aware Transformer for Handwritten Mathematical Expression Recognition (AAAI 2025 · better on 0 of 1 shared result)
- PosFormer: Recognizing Complex Handwritten Mathematical Expression with Position Forest Transformer (ECCV 2024 · better on 6 of 6 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
- Uni-MuMER: Unified Multi-Task Fine-Tuning of Vision-Language Model for Handwritten Mathematical Expression Recognition (NeurIPS 2025 · similarity 0.87)
- TAMER: Tree-Aware Transformer for Handwritten Mathematical Expression Recognition (AAAI 2025 · similarity 0.86)
- SSAN: A Symbol Spatial-Aware Network for Handwritten Mathematical Expression Recognition (AAAI 2025 · similarity 0.85)
- Generating Handwritten Mathematical Expressions From Symbol Graphs: An End-to-End Pipeline (CVPR 2024 · similarity 0.82)
- Separations in the Representational Capabilities of Transformers and Recurrent Architectures (NeurIPS 2024 · similarity 0.81)
- Seq-VCR: Preventing Collapse in Intermediate Transformer Representations for Enhanced Reasoning (ICLR 2025 · similarity 0.80)
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.
