Preference Prediction on PickScore (test)
73.4AccuracyDRM
Evaluation Results
| Method | Links | |
|---|---|---|
| DRMWeights=Pre-trained, Epoch=1, Size=5122026.05 | 73.4 | |
| HPSv3 - 7B2026.05 | 72.8 | |
| DRMWeights=Pre-trained, Epoch=1, Size=2562026.05 | 72.1 | |
| PickScore2026.05 | 70.5 | |
| AutoRubric-T2I (on PickScore)Base Model=Gemini-3-Flash, Judging Protocol=Pointwise2026.05 | 70.3 | |
| HPS2026.05 | 66.7 | |
| HPSv22026.05 | 63.8 | |
| HPSv3 - 2B2026.05 | 63.6 | |
| AutoRubric-T2I (on PickScore)Base Model=Qwen3-VL-8B, Judging Protocol=Pointwise2026.05 | 63.2 | |
| MPS2026.05 | 63.1 | |
| Qwen3-VL-8BJudging Protocol=Pairwise2026.05 | 62.2 | |
| AutoRubric-T2I (on HPSv3)Base Model=Qwen3-VL-8B, Judging Protocol=Pointwise2026.05 | 61.7 | |
| ImageReward2026.05 | 61.1 | |
| CLIP ViT-H/142026.05 | 60.8 | |
| DRMWeights=Random, Epoch=2, Size=2562026.05 | 59.5 | |
| DRMWeights=Random, Epoch=3, Size=2562026.05 | 59 | |
| DRMWeights=Random, Epoch=1, Size=2562026.05 | 57.5 | |
| Aesthetic Score Predictor2026.05 | 56.8 | |
| Qwen3-VL-8BJudging Protocol=Pointwise2026.05 | 35.4 |