To address this, we develop a new, more
comprehensive dataset for table extraction, called PubTables-1M.