Text-to-Image Alignment on Decomposable-Captions-4k
90.25Human Preference ScoreEvaluate-and-Refine (PW + CA)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Evaluate-and-Refine (PW + CA)Refinement=Prompt-weighting and Cross-attention updates, Inference Time (s)=12.24 s2023.07 | 90.25 | 74.16 | — | — | — | |
| Evaluate-and-Refine (PW)Refinement=Prompt-weighting, Inference Time (s)=10.32 s2023.07 | 89.53 | 70.28 | — | — | — | |
| Attend-and-ExciteInference Time (s)=8.59 s2023.07 | 85.94 | 65.5 | 23.6 | 62.3 | 14.1 | |
| Structure-DiffusionInference Time (s)=11.51 s2023.07 | 74.93 | 45.23 | 37.5 | 54.6 | 7.9 | |
| Stable-DiffusionInference Time (s)=3.54 s2023.07 | 72.98 | 43.66 | 41.7 | 50.1 | 8.2 | |
| Composable-DiffusionInference Time (s)=10.89 s2023.07 | 70.28 | 37.72 | 57.1 | 38.5 | 4.4 |