Safety violation detection on CAMMA-MVOR Synthetic
98.4Accuracy (Common)Ours (FT, Ensemble)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ours (FT, Ensemble)Fine-tuning (FT)=true, Strategy=Ensemble2025.06 | 98.4 | 98.41 | 97.49 | 98.21 | |
| GLM-4.6V-FlashBase Model=true2025.06 | 93.33 | 86.84 | 69.72 | 88.31 | |
| Ours (FT, Qwen-Screen)Fine-tuning (FT)=true, Base Model=Qwen-Screen2025.06 | 82.9 | 89.43 | 97.14 | 86.96 | |
| BaichuanMed-OCR-7BBase Model=true2025.06 | 76.51 | 59.14 | 7.41 | 49.59 | |
| Qwen2.5-VL-7B-InstructBase Model=true2025.06 | 59.53 | 64.92 | 8.37 | 42.52 | |
| Llama-3.2-11B-Vision-InstructBase Model=true2025.06 | 45.4 | 65.75 | 22.04 | 44.56 |