Harm Recognition on Violence imagery dataset
100F1 ScoreLLaVAShield
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVAShieldSetting=Pixel-only2026.06 | 100 | |
| SingGuard-2BSetting=Pixel-only2026.06 | 100 | |
| SingGuard-4BSetting=Pixel-only2026.06 | 100 | |
| SingGuard-8BSetting=Pixel-only2026.06 | 100 | |
| LlavaGuardSetting=Pixel-only2026.06 | 99.13 | |
| Qwen3-VL-235BSetting=Pixel-only2026.06 | 90.31 | |
| Gemini3-ProSetting=Pixel-only2026.06 | 89.52 | |
| GPT-5.1Setting=Pixel-only2026.06 | 87.69 | |
| SafeGuard-VL-RLSetting=Pixel-only2026.06 | 86.83 | |
| GuardReasoner-VLSetting=Pixel-only2026.06 | 85.15 | |
| Qwen3-VL-8BSetting=Pixel-only2026.06 | 82.83 | |
| Qwen3-VL-4BSetting=Pixel-only2026.06 | 77.89 | |
| ShieldGemma-2Setting=Pixel-only2026.06 | 73.91 | |
| LlamaGuard4Setting=Pixel-only2026.06 | 67.24 | |
| LlamaGuard3-VisionSetting=Pixel-only2026.06 | 18.75 |