Human Feedback Score Prediction on RichHF-18K (test)
0.487Text-image Alignment PLCCRAHF
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| RAHFarchitecture variant=multi-head2023.12 | 0.487 | 0.666 | 0.654 | 0.605 | 0.591 | 0.5 | 0.582 | 0.561 | |
| RAHFinput configuration=augmented prompt2023.12 | 0.474 | 0.693 | 0.681 | 0.6 | 0.589 | 0.496 | 0.58 | 0.562 | |
| CLIPevaluation protocol=fine-tuned2023.12 | 0.398 | 0.39 | 0.378 | 0.357 | 0.36 | 0.39 | 0.353 | 0.352 | |
| PickScoreevaluation protocol=off-the-shelf2023.12 | 0.346 | 0.0098 | 0.028 | 0.131 | 0.14 | 0.34 | 0.202 | 0.226 | |
| CLIPevaluation protocol=off-the-shelf2023.12 | 0.185 | — | — | — | — | 0.13 | — | — | |
| ResNet-502023.12 | 0.108 | 0.495 | 0.487 | 0.37 | 0.363 | 0.119 | 0.337 | 0.308 |