论文标题
砂粒：网格表相似性度量指标用于表结构识别
GriTS: Grid table similarity metric for table structure recognition
论文作者
论文摘要
在本文中，我们提出了一类新的用于表结构识别（TSR）评估的度量，称为网格表相似性（Grits）。与先前的指标不同，Grits可以直接以其自然形式作为矩阵评估预测表的正确性。为了在矩阵之间创建相似性度量，我们将NP-HARD的二维最大公共子结构（2D-LC）问题推广到2D最相似的子结构（2D-MSS）问题，并提出了解决该问题的多项式启发式方法。该算法在矩阵之间的真实相似性上产生上层和下限。我们在大型现实世界数据集上使用评估表明，实际上，这些界限几乎没有区别。我们将沙粒与其他指标进行比较，并在经验上验证矩阵相似性比用于TSR性能评估的替代方案表现出更理想的行为。最后，grits在同一框架内统一了细胞拓扑识别，细胞位置识别和细胞含量识别的所有三个子任务，从而简化了评估，并可以在不同类型的TSR方法上进行更有意义的比较。代码将在https://github.com/microsoft/table-transformer上发布。
In this paper, we propose a new class of metric for table structure recognition (TSR) evaluation, called grid table similarity (GriTS). Unlike prior metrics, GriTS evaluates the correctness of a predicted table directly in its natural form as a matrix. To create a similarity measure between matrices, we generalize the two-dimensional largest common substructure (2D-LCS) problem, which is NP-hard, to the 2D most similar substructures (2D-MSS) problem and propose a polynomial-time heuristic for solving it. This algorithm produces both an upper and a lower bound on the true similarity between matrices. We show using evaluation on a large real-world dataset that in practice there is almost no difference between these bounds. We compare GriTS to other metrics and empirically validate that matrix similarity exhibits more desirable behavior than alternatives for TSR performance evaluation. Finally, GriTS unifies all three subtasks of cell topology recognition, cell location recognition, and cell content recognition within the same framework, which simplifies the evaluation and enables more meaningful comparisons across different types of TSR approaches. Code will be released at https://github.com/microsoft/table-transformer.
