Robust table structure recognition with dynamic queries enhanced detection transformer

Jiawei Wang, Weihong Lin, Chixiang Ma, Mingze Li, Zheng Sun, Lei Sun, Qiang Huo

Abstract: Highlights•We are the first to formulate separation line prediction as a line regression problem. To demonstrate the effectiveness of this new formulation, we propose a new DETR based separation line prediction model, dubbed DQ-DETR, to predict the positions of points on separation lines from both distorted and undistorted table images with high localization accuracy.•We introduce the concept of dynamic queries into the DETR framework.•We propose a new prior-enhanced matching strategy to accelerate the convergence speed of DQ-DETR in training .•Our approach is robust to tables with complex structures, large blank spaces, empty or spanning cells and distorted or even curved tables.•Our approach achieves state-of-the-art performance on TSR public benchmarks.