Text-to-Image Generation on PartiPrompts 62 pairs SDXL-based
66.8PG-MAP Win RatePG-MAP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PG-MAPComparison Baseline=vs. NFE-matched UG, lambda (λ)=0.05, Number of raters=100, Total judgments=6,2002026.06 | 66.8 | 1,794 | 1.5 | |
| PG-MAPComparison Baseline=vs. SDXL static, lambda (λ)=0.05, Number of raters=100, Total judgments=6,2002026.06 | 60.2 | 1,458 | 5.9 | |
| PG-MAPComparison Baseline=vs. Tuned-CFG (w⋆=7.5), lambda (λ)=0.05, Number of raters=100, Total judgments=6,2002026.06 | 56 | 1,883 | 1.8 |