Text-to-Image Preference Alignment on HPD v2.1
0.385HPS-v2.1 ScoreDynamic-TreeRPO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Dynamic-TreeRPONFE_pi_theta_old=13.8(Avg), NFE_pi_theta=4, Iteration Time (s)=1512025.09 | 0.385 | 0.251 | 1.77 | |
| MixGRPONFE_pi_theta_old=25, NFE_pi_theta=4, Iteration Time (s)=2402025.09 | 0.367 | 0.237 | 1.629 | |
| MixGRPO-FlashNFE_pi_theta_old=16, NFE_pi_theta=4, Iteration Time (s)=1662025.09 | 0.358 | 0.236 | 1.528 | |
| DanceGRPONFE_pi_theta_old=25, NFE_pi_theta=14, Iteration Time (s)=3232025.09 | 0.356 | 0.233 | 1.436 | |
| DanceGRPONFE_pi_theta_old=25, NFE_pi_theta=4, Iteration Time (s)=2412025.09 | 0.334 | 0.225 | 1.335 | |
| FLUX2025.09 | 0.313 | 0.227 | 1.088 |