MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding

Multimodal pre-training with text, layout, and image has made significant progress for Visually Rich Document Understanding (VRDU), especially the fixed-layout documents such as scanned document images.