LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding

Benchmark Model Rank Results
document-image-classification-on-rvl-cdipLiLT[EN-R]BASE#5Accuracy: 95.68%
key-value-pair-extraction-on-sibrLiLT ([InfoXLM]_base)#5key-value pair F1: 72.76
key-value-pair-extraction-on-rfund-enLiLT ([EN-R]_base)#8key-value pair F1: 54.33
key-value-pair-extraction-on-rfund-enLiLT ([InfoXLM]_base)#10key-value pair F1: 52.18
key-information-extraction-on-cordLILT#9F1: 96.07
semantic-entity-labeling-on-funsdLILT#9F1: 88.41