Table 5 Semantic Structural Congruence (SSC) across datasets for different document layout models with input embeddings: OCR and Layout Only.
From: Representation learning approach for understanding structured documents
Dataset / Model | LayoutLMv36 | DocLayout-YOLO30 | DocSAM28 | LayoutLLM36 | DocLayLLM37 | D-REEL |
|---|---|---|---|---|---|---|
PRImA Newspaper Dataset34 | 57.14 | 60.27 | 58.63 | 63.42 | 64.19 | 73.05 |
German-Brazilian Newspapers (GBN)35 | 58.08 | 60.72 | 59.37 | 62.31 | 63.57 | 71.48 |
S2-VL Dataset10 | 83.52 | 86.11 | 86.43 | 88.59 | 89.03 | 90.47 |
IIIT AR 13K33 | 84.68 | 87.29 | 87.15 | 89.21 | 90.07 | 89.14 |
Publaynet32 | 84.03 | 86.42 | 89.17 | 88.26 | 89.08 | 89.31 |
