Text-to-Image Alignment on Multi-reward evaluation scenario
34.98HPSV2.1RSA-FT + DRTune
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RSA-FT + DRTuneReward=HPS + Pick + Aes2026.03 | 34.98 | 24.12 | 1.268 | |
| RSA-FT + DRTuneReward=HPS + Pick2026.03 | 34.74 | 24.14 | 1.259 | |
| RSA-FT + ReFLReward=HPS + Pick + Aes2026.03 | 34.39 | 24.3 | 1.226 | |
| RSA-FT + ReFLReward=HPS + Pick2026.03 | 34.21 | 24.33 | 1.207 | |
| DRTuneReward=HPS + Pick2026.03 | 34.16 | 24.01 | 1.217 | |
| ReFLReward=HPS + Pick + Aes2026.03 | 33.99 | 24.25 | 1.198 | |
| DRTuneReward=HPS + Pick + Aes2026.03 | 33.91 | 23.9 | 1.192 | |
| ReFLReward=HPS + Pick2026.03 | 33.75 | 24.3 | 1.197 |