LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking

Benchmark Model Rank Results
document-image-classification-on-rvl-cdipLayoutLMV3Large#2Accuracy: 95.93%Parameters: 368M
document-image-classification-on-rvl-cdipLayoutLMv3BASE#7Accuracy: 95.44%Parameters: 133M
document-layout-analysis-on-publaynet-valLayoutLMv3-B#3Overall: 0.951Text: 0.945Title: 0.906List: 0.955Table: 0.979
entity-linking-on-ec-funsdLayoutLMv3 (large)#4F1: 78.14
entity-linking-on-ec-funsdLayoutLMv3 (base)#7F1: 67.47
key-information-extraction-on-cordLayoutLMv3 Large#3F1: 97.46
key-value-pair-extraction-on-rfund-enLayoutLMv3#7key-value pair F1: 57.66
key-value-pair-extraction-on-sibrLayoutLMv3_base_chinese#4key-value pair F1: 73.51
named-entity-recognition-ner-on-cord-rLayoutLMv3#3F1: 82.72
named-entity-recognition-ner-on-funsd-rLayoutLMv3#2F1: 78.77
relation-extraction-on-funsdLayoutLMv3 large#3F1: 80.35
semantic-entity-labeling-on-ec-funsdLayoutLMv3 (large)#3F1: 83.88
semantic-entity-labeling-on-ec-funsdLayoutLMv3 (base)#7F1: 82.30
semantic-entity-labeling-on-funsdLayoutLMv3 Large#3F1: 92.08