Text-conditional image generation on LAION-10k
33.2CLIP ScoreLDM–T2I
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LDM–T2IDetection Threshold (τ1)=fixed2026.03 | 33.2 | — | — | 10.4 | 5.3 | 3.1 | |
| DCRDetection Threshold (τ1)=fixed2026.03 | 30.5 | — | — | 7.9 | 3.2 | 2.9 | |
| SD2.1-baseDetection Threshold (τ1)=fixed2026.03 | 27.8 | — | — | 8.3 | 7.4 | 3.3 | |
| RAPTADetection Threshold (τ1)=fixed2026.03 | 23.1 | — | — | 8.1 | 2.6 | 1.6 | |
| BaselineText mitigation=None2025.02 | 0.306 | 0.378 | 0.649 | 18.18 | — | — | |
| Ambient DiffusionText mitigation=S232025.02 | 0.306 | 0.308 | 0.547 | 21.3 | — | — | |
| Ambient DiffusionText mitigation=None2025.02 | 0.305 | 0.373 | 0.636 | 18.34 | — | — | |
| S23Text mitigation=S232025.02 | 0.302 | 0.319 | 0.573 | 20.55 | — | — | |
| W24Text mitigation=W242025.02 | 0.293 | 0.208 | 0.3 | 21.44 | — | — | |
| Ambient DiffusionText mitigation=W242025.02 | 0.293 | 0.192 | 0.267 | 20.74 | — | — |