Human preference prediction on ImageReward 466 prompts (test)
65.14Preference AccuracyImageReward
Evaluation results
| Method | Links | |
|---|---|---|
| ImageRewardBackbone=BLIP (ViT-L), Training Set Size=8k2023.04 | 65.14 | |
| BLIP Score2023.04 | 57.76 | |
| Aesthetic Score2023.04 | 57.35 | |
| CLIP Score2023.04 | 54.82 |