Optical Character Recognition on WuDao rendered text images (test)
0.996ROUGE (R=2)Deepseek-OCR
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Deepseek-OCRScale=3B2026.01 | 0.996 | 0.941 | 0.986 | 0.947 | 0.927 | |
| Qwen3-VL-SFTScale=8B2026.01 | 0.988 | 0.946 | 0.987 | 0.93 | 0.908 | |
| VIST2Scale=8B2026.01 | 0.981 | 0.95 | 0.975 | 0.925 | 0.912 | |
| VIST2Scale=0.6B2026.01 | 0.972 | 0.983 | 0.938 | 0.915 | 0.89 | |
| MinerU2.5Scale=1.2B2026.01 | 0.967 | 0.617 | 0.619 | 0.621 | 0.568 | |
| VIST2Scale=4B2026.01 | 0.961 | 0.942 | 0.97 | 0.928 | 0.905 | |
| PaddleOCR-VLScale=0.9B2026.01 | 0.922 | 0.851 | 0.817 | 0.809 | 0.852 | |
| DotsOCRScale=3B2026.01 | 0.841 | 0.83 | 0.846 | 0.76 | 0.518 | |
| Qwen3-VLScale=8B2026.01 | 0.521 | 0.368 | 0.318 | 0.37 | 0.302 |