Document Layout Analysis on D4LA
70.3mAPDocLayout-YOLO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DocLayout-YOLOBackbone=v10m++, Pretrain Dataset=DocSynth, 300K2024.10 | 70.3 | 82.4 | — | |
| VGTBackbone=ViT-B, Pretrain Dataset=IIT-CDIP, 42M2024.10 | 68.8 | — | — | |
| YOLO-v10Backbone=v10m2024.10 | 68.6 | 80.7 | — | |
| DiT-Cascade-LBackbone=ViT-L, Pretrain Dataset=IIT-CDIP, 42M2024.10 | 68.2 | 80.1 | — | |
| DiT-Cascade-BBackbone=ViT-B, Pretrain Dataset=IIT-CDIP, 42M2024.10 | 67.7 | 79.8 | — | |
| DINOLabeling Configuration=Fully supervised gold standard baseline2026.06 | 66.2 | — | — | |
| DINO-4scaleBackbone=R50, Pretrain Dataset=ImageNet1K2024.10 | 64.7 | 76.9 | — | |
| LayoutLMv3-BBackbone=ViT-B, Pretrain Dataset=IIT-CDIP, 42M2024.10 | 60 | 72.6 | — | |
| Faster R-CNNLabeling Configuration=Fully supervised gold standard baseline2026.06 | 54.2 | — | — | |
| DINOLabeling Configuration=BBLP pseudo-labels2026.06 | 53.9 | — | — | |
| DINOLabeling Configuration=10% manually re-annotated baseline2026.06 | 47.1 | — | — | |
| Faster R-CNNLabeling Configuration=BBLP pseudo-labels2026.06 | 42 | — | — | |
| Faster R-CNNLabeling Configuration=10% manually re-annotated baseline2026.06 | 37.9 | — | — | |
| DiT2026.03 | — | — | 67.7 | |
| LayoutLMv32026.03 | — | — | 62.6 | |
| PromptDLAModel=LayoutLMv32026.03 | — | — | 63.1 | |
| PromptDLAModel=DiT2026.03 | — | — | 69.1 | |
| VGT2026.03 | — | — | 68.8 |