Explanation on T-SROIE dense text for robustness (test)
84.9CSSDocShield
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DocShield2026.04 | 84.9 | 77 | |
| Gemini-2.5-Pro2026.04 | 79.8 | 69.8 | |
| qwen3-vl-8B2026.04 | 75.9 | 66 | |
| InternVL-3.5-8B2026.04 | 74.7 | 67.5 | |
| gpt-4o-o3-04162026.04 | 73.3 | 65.5 | |
| Qwen2.5-VL-7B2026.04 | 65.4 | 66.3 | |
| DeepSeekVL-7B2026.04 | 60.6 | 53.7 | |
| FakeShield2026.04 | 51.7 | 52.9 |