Synthetic Data Augmentation for Table Detection: Re-evaluating TableNet’s Performance with Automatically Generated Document Images
Krishna Sahukara, Zineddine Bettouche, Andreas M. Fischer · 2025
Document pages captured by smartphones or scanners often contain tables, yet manual extraction is slow and error-prone. We introduce an automated LaTeX-based pipeline that synthesises realistic two-column pages with visually diverse table layouts and aligned ground-truth masks. The generated corpus augments the real-world Marmot benchmark and enables a systematic resolution study of TableNet. Training TableNet on our synthetic data achieves a pixel-wise XOR error of 4.04 % on our synthetic test set with a 256 ×256 input resolution, and 4.33 % with 1024 ×1024. The best performance on the Marmot benchmark is 9.18 % (at 256 ×256). Cutting manual annotation effort by more than 75 %, the proposed framework allows for a detailed analysis of detection accuracy under different conditions.