Likert-scale judgment correlation on VQ-Diffusion generated images
82Kendall Tau-cMID
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MIDBackbone=CLIP (ViT-L/14)2022.05 | 82 | 71.1 | |
| MIDBackbone=CLIP (ViT-B/32)2022.05 | 79.8 | 69.1 | |
| InfoNCEBackbone=CLIP (ViT-L/14)2022.05 | 78 | 67.6 | |
| InfoNCEBackbone=CLIP (ViT-B/32)2022.05 | 74.2 | 64.3 | |
| OFA-Captioning+CLIP-SBackbone=OFA-Large + CLIP (ViT-L/14)2022.05 | 74.2 | 64.3 | |
| OFA-Captioning+CLIP-SBackbone=OFA-Large + CLIP (ViT-B/32)2022.05 | 73.8 | 63.9 | |
| CLIP-SBackbone=CLIP (ViT-L/14)2022.05 | 70.9 | 61.4 | |
| CLIP-SBackbone=CLIP (ViT-B/32)2022.05 | 70.3 | 60.9 | |
| CLIP-R-PrecisionBackbone=CLIP (ViT-L/14)2022.05 | 68.5 | 56.5 | |
| CLIP-R-PrecisionBackbone=CLIP (ViT-B/32)2022.05 | 66.5 | 54.5 | |
| SOABackbone=YOLO-V32022.05 | 37 | 38.4 |