Reading Order Matters: Information Extraction from Visually-rich Documents by Token Path Prediction

Open paper
Benchmark Model Rank Results
entity-linking-on-funsdTPP (LayoutMask)#3F1: 79.20
key-information-extraction-on-cordTPP (LayoutMask)#4F1: 96.92
key-value-pair-extraction-on-rfund-enTPP (LayoutLMv3_base)#11key-value pair F1: 50.27
named-entity-recognition-ner-on-cord-rTPP (LayoutLMv3)#1F1: 91.85
named-entity-recognition-ner-on-cord-rTPP (LayoutMask)#2F1: 89.34
named-entity-recognition-ner-on-funsd-rTPP (LayoutLMv3)#1F1: 80.40
named-entity-recognition-ner-on-funsd-rTPP (LayoutMask)#3F1: 78.19
reading-order-detection-on-roorTPP (LayoutLMv3-base)#3Segment-level F1: 42.96
relation-extraction-on-funsdTPP (LayoutMask)#5F1: 79.20
semantic-entity-labeling-on-funsdTPP (LayoutMask)#9F1: 85.16

Rank counts only results with a code link.