I2P
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Explicit Content Removal | I2P | Armpits Count2 | 28 | |
| Safe Text-to-Image Generation | I2P | Inappropriate Probability4 | 23 | |
| Concept Unlearning | I2P | I2P0.0011 | 17 | |
| Nudity Erasure | I2P 1.0 (test) | ASR (UD Attack)90.27 | 16 | |
| Inappropriate Content Erasing | I2P | I2P (%)17.8 | 14 | |
| Safe Text-to-Image Generation | I2P | ASR0.023 | 13 | |
| Erase Effectiveness | I2P sexual 1.0 (test) | Total Erased Count1,070 | 13 | |
| Common Robustness | I2P | ASR0.7 | 12 | |
| Nudity Unlearning | I2P | ESD71.83 | 11 | |
| Explicit Content Unlearning | I2P | Armpits153 | 11 | |
| Explicit Content Unlearning | I2P v1.5 (test) | Armpits162 | 10 | |
| NSFW Concept Erasure | I2P 4,703 potentially unsafe prompts | Total Success Count605 | 10 | |
| Nudity Detection | I2P (test) | Common Detections Count406 | 10 | |
| Text-to-Image Generation | I2P | Harmful Rate0.123 | 9 | |
| Text-to-Image Safety | I2P | Harmful Rate10.2 | 9 | |
| Explicit Content Erasure | I2P benchmark | NN Score0 | 9 | |
| Safe Generation Rate | I2P | GPT-4o Score83.88 | 9 | |
| Prompt-Image Alignment | I2P | CLIPScore0.8514 | 8 | |
| Concept Erasure | I2P | Nudity Rate1.83 | 7 | |
| Concept Attack | I2P Violence concept | FLUX.1 ASR91.06 | 6 | |
| Concept Attack | I2P Nudity concept | FLUX.1 ASR100 | 6 | |
| Text-to-Image Inappropriateness Evaluation | I2P | Prob Inappropriate (Hate)0.09 | 6 | |
| Concept Erasure | I2P | ASR (%)2.4 | 5 | |
| Safe Image Generation | I2P | Sexual Violation Frequency15 | 5 | |
| Safety-Sensitive Text-to-Image Generation | I2P Overall | Inappropriate Probability10 | 5 |
Showing 25 of 34 rows