Harm Recognition on Facebook Hateful Memes
87.79F1 ScoreGPT-5.1
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5.1Setting=Pixel-only2026.06 | 87.79 | |
| Gemini3-ProSetting=Pixel-only2026.06 | 87.22 | |
| GuardReasoner-VLSetting=Pixel-only2026.06 | 86.75 | |
| SingGuard-8BSetting=Pixel-only2026.06 | 85.71 | |
| SingGuard-4BSetting=Pixel-only2026.06 | 85.04 | |
| SingGuard-2BSetting=Pixel-only2026.06 | 84.73 | |
| LLaVAShieldSetting=Pixel-only2026.06 | 84.6 | |
| Qwen3-VL-235BSetting=Pixel-only2026.06 | 83.89 | |
| ShieldGemma-2Setting=Pixel-only2026.06 | 82.09 | |
| Qwen3-VL-8BSetting=Pixel-only2026.06 | 80.56 | |
| Qwen3-VL-4BSetting=Pixel-only2026.06 | 79.59 | |
| SafeGuard-VL-RLSetting=Pixel-only2026.06 | 76.7 | |
| LlamaGuard4Setting=Pixel-only2026.06 | 67.43 | |
| LlavaGuardSetting=Pixel-only2026.06 | 65.87 | |
| LlamaGuard3-VisionSetting=Pixel-only2026.06 | 64.98 |