UniMERNet：面向真实世界数学表达式识别的通用网络
UniMERNet: A Universal Network for Real-World Mathematical Expression Recognition
提出 Raster-Scan Attention，将全局注意力的 O(NH²W²D) 复杂度降至 O(NHWD(H+W))，可复用于其他需要处理长序列的视觉任务（如场景文本识别、文档理解），在保持精度的同时大幅降低显存和加速推理。
原文依据 reduces computational complexity from \mathcal O (NH^2W^2D) to \mathcal O (NHWD(H + W))
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论本文提出 UniMERNet，一种面向多样化真实场景的高精度、计算高效的数学表达式识别（MER）算法。为促进 UniMERNet 的训练，我们构建了 UniMER-1M，一个百万规模的数据集，其前所未有的多样性赋予模型稳健的泛化能力。通过深入分析，我们发现 Transformer 模型在 MER 任务中的注意力分布呈现出独特的光栅扫描模式（从左到右、从上到下），这与人类阅读习惯高度一致。基于这一关键发现，我们设计了一种创新的**光栅扫描注意力**机制，采用“先水平、后垂直”的顺序注意力计算策略。该方法不仅成功将计算复杂度从 \mathcal O (NH^2W^2D) 降低至 \mathcal O (NHWD(H + W))，还能更高效地捕获长程依赖，实现与全局注意力相当的识别性能。借助 UniMER-1M 和创新的注意力机制，UniMERNet 在四个真实场景中均达到最优性能，同时相比全局注意力显著降低计算资源：训练时显存节省超过 **1.2 倍**，推理时显存减少约 **10 倍**，速度提升 **5 倍**且精度略有提高。所有资源将公开释放，以进一步推动 MER 研究。
英文原文摘要 1461 字符
This paper introduces UniMERNet, a high-accuracy, computation-efficient algorithm for Mathematical Expression Recognition (MER) across diverse real-world scenarios. To facilitate UniMERNet's training, we constructed UniMER-1M, a million-scale dataset whose unprecedented diversity endows the model with robust generalization ability. Through in-depth analysis, we discover a distinctive raster-scan pattern (left-to-right, top-to-bottom) in the attention distribution of Transformer models for MER tasks, which closely aligns with human reading habits. Based on this key finding, we design an innovative **Raster-Scan Attention** mechanism that employs a "horizontal-first, vertical-second" sequential attention computation strategy. This approach not only successfully reduces computational complexity from \mathcal O (NH^2W^2D) to \mathcal O (NHWD(H + W)) , but also enables the model to capture long-range dependencies more efficiently, achieving recognition performance comparable to global attention. Leveraging both UniMER-1M and our innovative attention mechanism, UniMERNet achieves state-of-the-art performance across four real-world scenarios while significantly reducing computational resources compared to global attention: over **1.2x** memory savings during training, approximately **10x** memory reduction during inference, and **5x** speed with slightly improved accuracy. All resources will be publicly released to advance MER research further.