核心概念
UniTable unifies training paradigm and objectives for state-of-the-art performance in Table Structure Recognition.
統計資料
UniTable achieves SOTA 99.18% accuracy on SynthTabNet when pretrained on 2M images.
SSP helps mitigate performance drop when replacing CNN backbone with linear projection.
引述
"UniTable unifies the training objectives of all three TSR tasks into a unified task-agnostic training objective: language modeling."