Text-to-Image Generation on Stable Diffusion Alignment Prompts 1.5 (test)
0.7873ImageRewardAdjoint Matching
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Adjoint MatchingReward multiplier (λ)=10^22025.12 | 0.7873 | 0.2792 | 0.2791 | 0.3363 | |
| VMPO-DiffBackbone=SD1.5, Fine-tuned with=ImageReward2026.02 | 0.4716 | 0.2624 | 0.2645 | 0.3005 | |
| ITMReward multiplier (λ)=12025.12 | 0.4465 | 0.2794 | 0.2659 | 0.3383 | |
| GRPOBackbone=SD1.5, Fine-tuned with=ImageReward, group size=82026.02 | 0.249 | 0.272 | 0.242 | 0.3755 | |
| Adjoint MatchingReward multiplier (λ)=12025.12 | 0.217 | 0.2754 | 0.2576 | 0.3826 | |
| SD 1.5Mode=Base2025.12 | 0.1873 | 0.2746 | 0.2566 | 0.3849 | |
| VMPO-R2GBackbone=SD1.5, Fine-tuned with=ImageReward2026.02 | 0.1698 | 0.2721 | 0.2423 | 0.4008 | |
| SD1.5 (Base)Backbone=SD1.5, Status=Base Model2026.02 | 0.0331 | 0.2717 | 0.2368 | 0.4389 |