PubTables-1M: Towards comprehensive table extraction from unstructured documents
- Brandon Smock ,
- Rohith Pesala ,
- Robin Abraham
Recently, significant progress has been made applying machine learning to the problem of table structure inference and extraction from unstructured documents. However, one of the greatest challenges remains the creation of datasets with complete, unambiguous ground truth at scale. To address this, we develop a new, more comprehensive dataset for table extraction, called PubTables-1M. PubTables-1M contains nearly one million tables from scientific articles, supports multiple input modalities, and contains detailed header and location information for table structures, making it useful for a wide variety of modeling approaches. It also addresses a significant source of ground truth inconsistency observed in prior datasets called oversegmentation, using a novel canonicalization procedure. We demonstrate that these improvements lead to a significant increase in training performance and a more reliable estimate of model performance at evaluation for table structure recognition. Further, we show that transformer-based object detection models trained on PubTables-1M produce excellent results for all three tasks of detection, structure recognition, and functional analysis without the need for any special customization for these tasks.
论文与出版物下载
PubTables
4 6 月, 2021
This project is a large dataset, along with baseline trained machine learning models, for the tasks of table detection and table structure recognition in scientific PDF documents. Current datasets for table structure recognition are small and pre-processed in ways that make them applicable only to a specific model architecture, which has limited progress in data-driven methods for this task. The goal of releasing this dataset is to provide a new large standard benchmark for evaluation and a dataset for training that is large enough for deep models to learn effectively. Doing so would enable significant progress to be made toward machine learning methods for these tasks. The dataset for release is derived entirely from a public dataset, the PubMed Open Access dataset of over one million scientific articles, and specifically from the Commercial-Use Collection subset.