Binary Safety Hallucination Detection on EndoVis VQA In-template 18 (val)
98AccuracyQA-SNNE
Evaluation Results
| Method | Links | |
|---|---|---|
| QA-SNNEBackbone Model=MedGemma3.0, Evaluation Protocol=Zero-shot, Variant=Ent2025.11 | 98 | |
| QA-SNNEBackbone Model=MedGemma3.0, Evaluation Protocol=Zero-shot, Variant=CrossE2025.11 | 98 | |
| SNNEBackbone Model=PitVQA, Evaluation Protocol=Peft2025.11 | 98 | |
| VL-UBackbone Model=PitVQA, Evaluation Protocol=Peft2025.11 | 98 | |
| QA-SNNEBackbone Model=MedGemma3.0, Evaluation Protocol=Zero-shot, Variant=Emb2025.11 | 87 | |
| QA-SNNEBackbone Model=PitVQA, Evaluation Protocol=Peft, Variant=Emb2025.11 | 85 | |
| QA-SNNEBackbone Model=SurgicalGPT, Evaluation Protocol=Peft, Variant=Emb2025.11 | 84 | |
| SNNEBackbone Model=SurgicalGPT, Evaluation Protocol=Peft2025.11 | 82 | |
| VL-UBackbone Model=MedGemma3.0, Evaluation Protocol=Zero-shot2025.11 | 81 | |
| VL-UBackbone Model=Qwen2.5, Evaluation Protocol=Zero-shot2025.11 | 79 | |
| VL-UBackbone Model=Llama3.2, Evaluation Protocol=Zero-shot2025.11 | 74 | |
| QA-SNNEBackbone Model=Llama3.2, Evaluation Protocol=Zero-shot, Variant=CrossE2025.11 | 70 | |
| QA-SNNEBackbone Model=Llama3.2, Evaluation Protocol=Zero-shot, Variant=Ent2025.11 | 69 | |
| QA-SNNEBackbone Model=Llama3.2, Evaluation Protocol=Zero-shot, Variant=Emb2025.11 | 67 | |
| QA-SNNEBackbone Model=Qwen2.5, Evaluation Protocol=Zero-shot, Variant=Ent2025.11 | 67 | |
| SNNEBackbone Model=Llama3.2, Evaluation Protocol=Zero-shot2025.11 | 56 | |
| QA-SNNEBackbone Model=Qwen2.5, Evaluation Protocol=Zero-shot, Variant=CrossE2025.11 | 54 | |
| QA-SNNEBackbone Model=PitVQA, Evaluation Protocol=Peft, Variant=CrossE2025.11 | 51 | |
| VL-UBackbone Model=SurgicalGPT, Evaluation Protocol=Peft2025.11 | 39 | |
| QA-SNNEBackbone Model=SurgicalGPT, Evaluation Protocol=Peft, Variant=Ent2025.11 | 35 | |
| QA-SNNEBackbone Model=SurgicalGPT, Evaluation Protocol=Peft, Variant=CrossE2025.11 | 35 | |
| SNNEBackbone Model=MedGemma3.0, Evaluation Protocol=Zero-shot2025.11 | 22 | |
| SNNEBackbone Model=Qwen2.5, Evaluation Protocol=Zero-shot2025.11 | 20 | |
| QA-SNNEBackbone Model=Qwen2.5, Evaluation Protocol=Zero-shot, Variant=Emb2025.11 | 20 | |
| QA-SNNEBackbone Model=PitVQA, Evaluation Protocol=Peft, Variant=Ent2025.11 | 1 |