Complicated Table Structure Recognition

TL;DR

This work targets PDF table structure recognition, a challenging task when many tables contain spanning cells. It introduces GraphTSR, a graph neural network that treats table cells as vertices and cell relations as edges, using alternating edge-to-vertex and vertex-to-edge attention blocks to predict vertical/horizontal relations. To support evaluation, the authors curate SciTSR, a large-scale dataset of 15,000 PDF tables with structure labels derived from LaTeX, and provide detailed experimental results showing strong gains over baselines, especially on complicated tables. The approach demonstrates superior generalization across datasets and underscores the importance of modeling local graph structure with attention mechanisms for accurate reconstruction of complex table schemes. These contributions advance automatic table understanding in PDFs and offer a practical resource for future research.

Abstract

The task of table structure recognition aims to recognize the internal structure of a table, which is a key step to make machines understand tables. Currently, there are lots of studies on this task for different file formats such as ASCII text and HTML. It also attracts lots of attention to recognize the table structures in PDF files. However, it is hard for the existing methods to accurately recognize the structure of complicated tables in PDF files. The complicated tables contain spanning cells which occupy at least two columns or rows. To address the issue, we propose a novel graph neural network for recognizing the table structure in PDF files, named GraphTSR. Specifically, it takes table cells as input, and then recognizes the table structures by predicting relations among cells. Moreover, to evaluate the task better, we construct a large-scale table structure recognition dataset from scientific papers, named SciTSR, which contains 15,000 tables from PDF files and their corresponding structure labels. Extensive experiments demonstrate that our proposed model is highly effective for complicated tables and outperforms state-of-the-art baselines over a benchmark dataset and our new constructed dataset.

Figures (6)

• Figure 1: An intuitive example of a complicated table with spanning cells. The example table is shown in (a), and (b) is the real structure of the dashed box area. The recognized structure by existing methods are shown in (c) - (e). Note that in (c), the four cells on the right side are incorrectly recognized as a single cell.
• Figure 2: Overview of our method. Given a table in PDF as input, our method recognize its structure by the following four steps: (a) Pre-processing: obtaining cell contents and their corresponding bounding box from PDF; (b) Graph construction: building an undirected graph on these cells; (c) Relation prediction: predicting adjacent relations by our proposed GraphTSR; (d) Post-processing: recovering table structure from the labeled graph.
• Figure 3: The architecture of our proposed GraphTSR.
• Figure 4: The construction pipeline of SciTSR dataset.
• Figure 5: A sample from results on SciTSR-COMP. Cells are marked with different colors to distinguish from each other.