Research map / Architectures and efficiency / Attention and transformers
UniMERNet: A Universal Network for Real-World Mathematical Expression Recognition
Zhuangcheng Gu, Guang Liang, Bin Wang, Zhiyuan Zhao, Qintong Zhang, Weijia Li, Chao Xu, Bo Zhang, Botian Shi, Jiang Wu, Wentao Zhang, Conghui He
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 global attention with Raster-Scan attention (ablation) | UniMERNet encoder · UniMER-Test CPE | CDM: 0.943 -> 0.952 | Improved |
| replace global attention with Raster-Scan attention (ablation) | UniMERNet encoder · encoder efficiency | FPS (bs=1): 2.11 -> 10.52 | Improved |
| replace window attention with Raster-Scan attention (ablation) | UniMERNet encoder · UniMER-Test CPE | CDM: 0.931 -> 0.952 | Improved |
| use horizontal-then-vertical instead of vertical-then-horizontal order (ablation) | R-S Attention · UniMER-Test CPE | CDM: 0.948 -> 0.952 | Improved |
| add ConvEnhance module (ablation) | UniMERNet encoder · UniMER-Test CPE | CDM: 0.952 -> 0.955 | Improved |
| add ConvEnhance module (ablation) | UniMERNet encoder · UniMER-Test SCE | CDM: 0.937 -> 0.939 | No clear effect |
| train on UniMER-1M instead of Pix2tex+HWE (ablation) | UniMERNet · UniMER-Test CPE | CDM: 0.665 -> 0.955 | Improved |
| train on UniMER-1M instead of Pix2tex+HWE (ablation) | UniMERNet · UniMER-Test SPE | CDM: 0.983 -> 0.991 | Improved |
| add handwritten data to Pix2tex training (ablation) | UniMERNet · UniMER-Test CPE | CDM: 0.667 -> 0.665 | No clear effect |
| increase input resolution to 384x1344 | UniMERNet · UniMER-Test CPE | CDM: 0.955 -> 0.972 | Improved |
| compare UniMERNet with PosFormer on HME100K | UniMERNet · HME100K | ExpRate: 69.5 -> 68.5 | No clear effect |
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.
- PosFormer — PosFormer: Recognizing Complex Handwritten Mathematical Expression with Position Forest Transformer (ECCV 2024) · better on 2 of 3 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 0.90)
- Complex Mathematical Expression Recognition: Benchmark, Large-Scale Dataset and Strong Baseline (AAAI 2026 · similarity 0.90)
- Handwritten Mathematical Expression Recognition via Attention Aggregation Based Bi-directional Mutual Learning (AAAI 2022 · similarity 0.89)
- Seeing Symbols, Missing Structure: A Real-World Handwritten Mathematical Expression Recognition Benchmark for Large Models (ICML 2026 · similarity 0.88)
- SSAN: A Symbol Spatial-Aware Network for Handwritten Mathematical Expression Recognition (AAAI 2025 · similarity 0.86)
- TAMER: Tree-Aware Transformer for Handwritten Mathematical Expression Recognition (AAAI 2025 · similarity 0.85)
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.
