Malicious Prompt Detection on VLGuard & MLLMGuard
96.82AUROCVLMGuard
Evaluation results
| Method | Links | |
|---|---|---|
| VLMGuardBackbone=Qwen2.5-VL-72B-Instruct2024.10 | 96.82 | |
| GuardRankBackbone=LLaVA-1.6-7B, Training Supervision=Fully supervised2024.10 | 92.17 | |
| VLMGuardBackbone=LLaVA-1.6-7B, Training Supervision=Unsupervised (no labeled data)2024.10 | 91.94 | |
| ASTRABackbone=Qwen2.5-VL-72B-Instruct2024.10 | 85.83 | |
| HiddenDetectBackbone=Qwen2.5-VL-72B-Instruct2024.10 | 78.72 |