Reward Modeling on PAM∃LA held-out users (test)
0.4514User SROCCPAM∃LA
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PAM∃LA2026.04 | 0.4514 | 0.5269 | 0.4722 | 0.6116 | 66.31 | 67.98 | |
| HPSv32026.04 | 0.4019 | 0.5076 | 0.4444 | 0.588 | 64.27 | 67.73 | |
| ImageReward2026.04 | 0.2841 | 0.2314 | 0.2855 | 0.2044 | 59.78 | 57.62 | |
| Q-AlignVariant=Aesthetics2026.04 | 0.2677 | 0.3273 | 0.2906 | 0.3606 | 59.32 | 61.09 | |
| Q-AlignVariant=IQA2026.04 | 0.2497 | 0.329 | 0.2416 | 0.3244 | 58.65 | 60.96 | |
| DeQA2026.04 | 0.2371 | 0.2864 | 0.2105 | 0.2741 | 58.18 | 59.5 | |
| LAION2026.04 | 0.1516 | 0.1511 | 0.1471 | 0.1426 | 51.1 | 51.56 |