TABLET: Table Structure Recognition using Encoder-only Transformers

TL;DR

This paper introduces TABLET, a Transformer-based model for accurate, fast, and scalable table structure recognition, especially for large, densely populated tables, outperforming existing methods in real-world scenarios.

Contribution

The paper presents a novel Split-Merge top-down approach using dual Transformer encoders for improved table structure recognition without unstable bounding box predictions.

Findings

Outperforms existing methods on FinTabNet and PubTabNet datasets.

Achieves high accuracy with reduced computational complexity.

Demonstrates robustness and scalability for industrial applications.

Abstract

To address the challenges of table structure recognition, we propose a novel Split-Merge-based top-down model optimized for large, densely populated tables. Our approach formulates row and column splitting as sequence labeling tasks, utilizing dual Transformer encoders to capture feature interactions. The merging process is framed as a grid cell classification task, leveraging an additional Transformer encoder to ensure accurate and coherent merging. By eliminating unstable bounding box predictions, our method reduces resolution loss and computational complexity, achieving high accuracy while maintaining fast processing speed. Extensive experiments on FinTabNet and PubTabNet demonstrate the superiority of our model over existing approaches, particularly in real-world applications. Our method offers a robust, scalable, and efficient solution for large-scale table recognition, making it well-suited for industrial deployment.