Human Preference Image Retrieval on HPIR
74.4AccuracyDataComp + RLFT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DataComp + RLFTTraining Phase=Alignment fine-tuning2024.06 | 74.4 | 71.9 | |
| CLIP + RLFTTraining Phase=Alignment fine-tuning2024.06 | 73.1 | 71.7 | |
| DataCompTraining Phase=Pre-training2024.06 | 71.8 | 62.7 | |
| Ours-RLFTTraining Phase=Alignment fine-tuning2024.06 | 71.7 | 67.6 | |
| CLIPTraining Phase=Pre-training2024.06 | 68.1 | 62.1 | |
| Ours-PTTraining Phase=Pre-training2024.06 | 66.2 | 59.4 |