Content Moderation on UnsafeBench Sexual category (test)
81.4AccuracyKidsNanny (Stage 1+2)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| KidsNanny (Stage 1+2)Regime=Multimodal (full Stage 1+2 pipeline), Inference (ms)=1202026.03 | 81.4 | 86.16 | 83.22 | 89.31 | |
| LlavaGuardRegime=Multimodal (full Stage 1+2 pipeline), Inference (ms)=41382026.03 | 80.36 | 84.56 | 86.17 | 83.02 | |
| KidsNanny Stage 1Regime=Vision-Only (Stage 1 active; OCR and LLM disabled), Inference (ms)=11.72026.03 | 80.27 | 85.39 | 82.05 | 89.02 | |
| FreepikRegime=Vision-Only (Stage 1 active; OCR and LLM disabled), Inference (ms)=15.52026.03 | 77.04 | 81.12 | 86.81 | 76.13 | |
| Adam-ViTRegime=Vision-Only (Stage 1 active; OCR and LLM disabled), Inference (ms)=7.22026.03 | 68.98 | 73.9 | 81.23 | 67.79 | |
| NudeNetRegime=Vision-Only (Stage 1 active; OCR and LLM disabled), Inference (ms)=352026.03 | 68.03 | 76.01 | 73.96 | 78.18 | |
| ShieldGemma-2Regime=Multimodal (full Stage 1+2 pipeline), Inference (ms)=11362026.03 | 64.8 | 76.68 | 64.96 | 93.56 | |
| FalconsAIRegime=Vision-Only (Stage 1 active; OCR and LLM disabled), Inference (ms)=7.32026.03 | 59.01 | 56.28 | 91.15 | 40.7 |