Tampered Text Classification on TFR CL (Cross-Language)
85.5AccuracyTextShield-R1
Evaluation Results
| Method | Links | |
|---|---|---|
| TextShield-R1Fine-tuning=Full training set images2026.02 | 85.5 | |
| SIDA*Fine-tuning=Full training set images, Base MLLM=Qwen2.5-VL-7B2026.02 | 85.2 | |
| Qwen2.5-VL-7BFine-tuning=Full training set images2026.02 | 85.1 | |
| FakeShield*Fine-tuning=Full training set images, Base MLLM=Qwen2.5-VL-7B2026.02 | 85.1 | |
| InternVL3-8BFine-tuning=Full training set images2026.02 | 84.3 | |
| MiniCPM_V_2.6Fine-tuning=Full training set images2026.02 | 81.5 | |
| Qwen2.5-VL-3BFine-tuning=Full training set images2026.02 | 80.9 | |
| InternVL3-2BFine-tuning=Full training set images2026.02 | 80.2 | |
| SIDAFine-tuning=Full training set images2026.02 | 70.4 | |
| FakeShieldFine-tuning=Full training set images2026.02 | 70.3 | |
| Qwen2.5-VL-7BFine-tuning=None2026.02 | 50.1 | |
| GPT4oFine-tuning=None2026.02 | 48.3 | |
| Qwen2.5-VL-3BFine-tuning=None2026.02 | 47.1 | |
| InternVL3-8BFine-tuning=None2026.02 | 45.1 | |
| InternVL3-2BFine-tuning=None2026.02 | 34.5 | |
| MiniCPM_V_2.6Fine-tuning=None2026.02 | 30.9 |