Text-to-Image Generation on Pick-a-Pic Unique V1 (val)
21.48PickScorePoly-DPO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Poly-DPOParadigm=Off-Policy2026.04 | 21.48 | 28.32 | 5.68 | 0.671 | |
| SPOParadigm=On-Policy2026.04 | 21.22 | 25.83 | 5.93 | 0.168 | |
| Diffusion-KTOParadigm=Off-Policy2026.04 | 21.12 | 28.19 | 5.68 | 0.642 | |
| DDPOParadigm=On-Policy2026.04 | 21.06 | 24.91 | 5.59 | 0.082 | |
| Diffusion-DPOParadigm=Off-Policy2026.04 | 20.99 | 25.54 | 5.6 | 0.302 | |
| D3POParadigm=On-Policy2026.04 | 20.76 | 23.97 | 5.53 | 0.124 | |
| SD1.5Paradigm=-2026.04 | 20.56 | 24.05 | 5.47 | 0.008 |