Tables to LaTeX: structure and content extraction from scientific tables

Scientific documents contain tables that list important information in a concise fashion. Structure and content extraction from embedded within PDF research is very challenging task due to the existence of visual features like spanning cells mathematical symbols equations. Most existing table structure identification methods tend ignore these academic writing features. In this paper, we adapt transformer-based language modeling paradigm for scientific extraction. Specifically, proposed model converts tabular image its corresponding LaTeX source code. Overall, outperform current state-of-the-art baselines achieve an exact match accuracy 70.35 49.69% on extraction, respectively. Further analysis demonstrates models efficiently identify number rows columns, alphanumeric characters, tokens, symbols.