Safety Conflict Detection on Synthetic OR-VSKC constructed from 4D-OR (test)
99.27Common Conflict Detection RateOurs (FT, Ensemble)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ours (FT, Ensemble)Fine-tuned=true, Strategy=Ensemble2025.06 | 99.27 | 95.14 | 99.64 | 97.56 | |
| Ours (FT, Qwen-Screen)Fine-tuned=true, Screening Model=Qwen-Screen2025.06 | 92.14 | 96.11 | 98.67 | 95.54 | |
| Qwen2.5-VL-7B-Instruct2025.06 | 72.07 | 56.79 | 65.09 | 64.01 | |
| GPT-4o2025.06 | 71.75 | 71.71 | 60.14 | 68.51 | |
| BaichuanMed-OCR-7B2025.06 | 69.83 | 40.46 | 56.75 | 55.69 | |
| Qwen-vl-plus2025.06 | 63.16 | 65.48 | 63.94 | 64.4 | |
| Llama-3.2-11B-Vision-Instruct2025.06 | 62.19 | 64.59 | 64.03 | 63.68 | |
| GLM-4.6V-Flash2025.06 | 34.26 | 25.12 | 37.25 | 31.6 |