NSFW
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| NSFW Concept Generation | NSFW-200 Violence v2.1 (test) | ASR-166 | 70 | |
| NSFW Concept Generation | NSFW-200 Sex v2.1 (test) | ASR-162 | 70 | |
| Text Safety Filter Bypass | NSFW-200 | Text Bypass Rate31 | 13 | |
| Concept Unlearning Preservation | NSFW | CSDR12.76 | 12 | |
| Adversarial Robustness | NSFW | ASR4.69 | 11 | |
| Image Safety Filter Bypass | NSFW-200 | Image Bypass Rate69.4 | 7 | |
| Harmful prompt detection | NSFW56k | Accuracy99 | 6 | |
| NSFW Detection | NSFW56k | Acceptance Rate (ASR)97 | 5 | |
| Text Safety Filter Bypass | NSFW-200 Target Filter: DeepSeek-V3 1.0 (cross-filter transfer) | Bypass Rate87.5 | 3 | |
| Text Safety Filter Bypass | NSFW-200 Target Filter: GPT-4.1 1.0 (cross-filter transfer) | Bypass Rate97.5 | 3 | |
| Text Safety Filter Bypass | NSFW-200 Target Filter: ShieldLM 1.0 (cross-filter transfer) | Bypass Rate92.5 | 3 | |
| Text-to-Image Generation | NSFW-200 | CLIP Score0.2762 | 3 | |
| Jailbreaking | NSFW-200 1.0 (test) | Bypass Rate90.5 | 3 | |
| NSFW Safety Evaluation | NSFW | Metric- | 0 |
Showing 14 of 14 rows