Leveraging Existing Table Structures and Content to Synthesize Realistic Data for Improving Table Recognition

A novel method for synthesizing high-quality, realistic table annotation data by leveraging the structure and content of existing complex tables, enabling the efficient creation of tables that closely replicate authentic styles found in the target domain.