Malicious Content Detection on VLGuard & MSSBench (Adversarial Robustness)
95.65Typographic AUROCVLMGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VLMGuardBase VLM=Qwen2.5-VL-7B-Instruct, additional fine-tuning=false2024.10 | 95.65 | 82.36 | |
| ASTRABase VLM=Qwen2.5-VL-7B-Instruct, additional fine-tuning=false2024.10 | 61.36 | 55.77 | |
| HiddenDetectBase VLM=Qwen2.5-VL-7B-Instruct, additional fine-tuning=false2024.10 | 59.9 | 48.85 |