Safety Evaluation on Toxic Image Dataset Section 3 1.0 (test)
20.65PornTGA
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TGABasic LLM=Mistral-7B, Safety of LLM=Medium2024.10 | 20.65 | 9.48 | 22.73 | 17.92 | 17.29 | 30.83 | 29.42 | |
| Unlearn-FigSBasic LLM=Mistral-7B, Safety of LLM=Medium2024.10 | 8.76 | 4.27 | 16.98 | 14.31 | 10.1 | 21.42 | 18.55 | |
| Qwen-VL-chatBasic LLM=Qwen-7B, Safety of LLM=Strong2024.10 | 4.23 | 1.46 | 5.15 | 5.48 | 4.41 | 5.72 | 5.4 | |
| InstructBlipBasic LLM=Vicuna-7B, Safety of LLM=Weak2024.10 | 1.28 | 0.12 | 0.57 | 0 | 0 | 0.75 | 0.23 | |
| LLaVA-1.5Basic LLM=Vicuna-7B, Safety of LLM=Weak2024.10 | 1.2 | 0.37 | 0.57 | 0.19 | 0.76 | 1.22 | 0.35 | |
| BLIP-2Basic LLM=Vicuna-13B, Safety of LLM=Weak2024.10 | 1.17 | 0 | 0 | 0 | 0 | 0.98 | 0.06 | |
| LLaVA-1.6Basic LLM=Mistral-7B, Safety of LLM=Medium2024.10 | 1.05 | 0.56 | 0.78 | 0.25 | 0.17 | 1.95 | 1.22 |