Text-to-Image Generation on Drawbench (HPSV2.1, PickScore, ImageReward)
31.67HPSV2.1ReFL + RSA-FT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ReFL + RSA-FTBackbone=SD 1.5, Resolution=512 x 5122026.03 | 31.67 | 21.7 | 0.671 | |
| DRTune + RSA-FTBackbone=SDXL, Resolution=1024 × 10242026.03 | 31.58 | 22.61 | 0.944 | |
| DRTune + RSA-FTBackbone=SD 1.5, Resolution=512 x 5122026.03 | 31.16 | 21.52 | 0.54 | |
| ReFLBackbone=SD 1.5, Resolution=512 x 5122026.03 | 31.08 | 21.57 | 0.536 | |
| DRTuneBackbone=SD 1.5, Resolution=512 x 5122026.03 | 30.63 | 21.34 | 0.477 | |
| ReFL + RSA-FTBackbone=SDXL, Resolution=1024 × 10242026.03 | 30.31 | 22.6 | 0.719 | |
| DRTuneBackbone=SDXL, Resolution=1024 × 10242026.03 | 30.04 | 22.49 | 0.804 | |
| DRTune + RSA-FTBackbone=SD3, Resolution=1024 x 10242026.03 | 29.89 | 22.61 | 0.979 | |
| ReFL + RSA-FTBackbone=SD3, Resolution=1024 x 10242026.03 | 29.81 | 22.51 | 0.953 | |
| Alignprop + RSA-FTBackbone=SD 1.5, Resolution=512 x 5122026.03 | 29.59 | 21.51 | 0.268 | |
| DRTuneBackbone=SD3, Resolution=1024 x 10242026.03 | 29.48 | 22.41 | 0.914 | |
| ReFLBackbone=SD3, Resolution=1024 x 10242026.03 | 29.45 | 22.38 | 0.921 | |
| ReFLBackbone=SDXL, Resolution=1024 × 10242026.03 | 29.45 | 22.41 | 0.705 | |
| VanillaBackbone=SD3, Resolution=1024 x 10242026.03 | 28.77 | 22.45 | 0.914 | |
| VanillaBackbone=SDXL, Resolution=1024 × 10242026.03 | 27.41 | 22.31 | 0.619 | |
| Draft-LV + RSA-FTBackbone=SD 1.5, Resolution=512 x 5122026.03 | 26.67 | 21.12 | 0.035 | |
| Draft-LVBackbone=SD 1.5, Resolution=512 x 5122026.03 | 25.59 | 20.96 | -0.062 | |
| AlignpropBackbone=SD 1.5, Resolution=512 x 5122026.03 | 25.12 | 20.98 | -0.033 | |
| VanillaBackbone=SD 1.5, Resolution=512 x 5122026.03 | 24.02 | 21.02 | -0.147 |