Text-to-Image Generation on MemBench 3000 Memorized Prompts 1.0 (test)
50.3SSCD (Target)No mitigation
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| No mitigationBackbone=SD v1.4, Denoising Sampler=DDIM Sampling, Evaluation Protocol=Zero-Shot Generalization2026.02 | 50.3 | 47.51 | 3.67 | 22.023 | 30.05 | |
| Ren et al.Backbone=SD v1.4, Denoising Sampler=DDIM Sampling, Evaluation Protocol=Zero-Shot Generalization2026.02 | 21.84 | 15.08 | 4.65 | 20.1319 | 29.37 | |
| Wen et al.Backbone=SD v1.4, Denoising Sampler=DDIM Sampling, Evaluation Protocol=Zero-Shot Generalization2026.02 | 20.91 | 12.19 | 4.12 | 18.7822 | 29.98 | |
| Jain et al.Backbone=SD v1.4, Denoising Sampler=DDIM Sampling, Evaluation Protocol=Zero-Shot Generalization2026.02 | 17.75 | 7.26 | 17.82 | 38.5641 | 25.2 | |
| RADSBackbone=SD v1.4, Denoising Sampler=DDIM Sampling, Evaluation Protocol=Zero-Shot Generalization2026.02 | 14.49 | 10.06 | 5.16 | 26.7537 | 25.24 |