Image Classification on 12 Fine-Grained Datasets Suite
83.4Accuracy (Food-101)Scala-B
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Scala-Br=1.00, Arch=ViT-B, Protocol=Linear Probing2024.12 | 83.4 | 97.3 | 86.6 | 69.7 | 74.3 | 63 | 87.6 | 78 | 94.6 | 95.1 | 97.3 | 84.1 | 84.2 | |
| Scala-Br=0.50, Arch=ViT-S, Protocol=Linear Probing2024.12 | 78.6 | 95.5 | 81.9 | 65.9 | 64.4 | 55.2 | 86.1 | 74.9 | 94 | 93.9 | 96 | 81.1 | 80.6 | |
| DeiT-Br=1.00, Arch=ViT-B, Protocol=Linear Probing2024.12 | 76.9 | 95.6 | 81.1 | 63 | 57.1 | 49.1 | 83 | 70.3 | 92.8 | 90.7 | 90.7 | 72.8 | 77 | |
| Scala-Br=0.25, Arch=ViT-Ti, Protocol=Linear Probing2024.12 | 72.8 | 93.6 | 78.1 | 61.4 | 55.4 | 49.6 | 84.2 | 72.1 | 92.9 | 92.1 | 93.4 | 76.7 | 76.8 | |
| DeiT-Sr=1.00, Arch=ViT-S, Protocol=Linear Probing2024.12 | 72.1 | 94.1 | 77.7 | 60.5 | 49.4 | 44.5 | 83.3 | 69.5 | 92.1 | 90.3 | 88.8 | 71.9 | 74.5 | |
| DeiT-Tir=1.00, Arch=ViT-Ti, Protocol=Linear Probing2024.12 | 65 | 90.6 | 72.5 | 56.1 | 39.7 | 39.9 | 81.6 | 67 | 91.1 | 88.8 | 85.1 | 69.4 | 70.6 | |
| DINOv2Arch=ViT-g/14, Protocol=Linear evaluation of frozen features2023.04 | 0.947 | 0.995 | 0.944 | 0.787 | 0.914 | 0.872 | 0.89 | 0.845 | 0.967 | 0.976 | 0.997 | 0.916 | 0.921 | |
| OpenCLIPArch=ViT-G/14, Protocol=Linear evaluation of frozen features2023.04 | 0.945 | 0.987 | 0.91 | 0.84 | 0.961 | 0.802 | 0.893 | 0.86 | 0.957 | 0.981 | 0.995 | 0.899 | 0.919 | |
| DINOv2Arch=ViT-L/14, Protocol=Linear evaluation of frozen features2023.04 | 0.943 | 0.993 | 0.934 | 0.787 | 0.901 | 0.815 | 0.883 | 0.84 | 0.966 | 0.975 | 0.997 | 0.905 | 0.912 | |
| DINOv2Arch=ViT-B/14, Protocol=Linear evaluation of frozen features2023.04 | 0.928 | 0.987 | 0.913 | 0.773 | 0.882 | 0.794 | 0.882 | 0.833 | 0.962 | 0.961 | 0.996 | 0.896 | 0.901 | |
| iBOTArch=ViT-L/16, Protocol=Linear evaluation of frozen features2023.04 | 0.91 | 0.99 | 0.928 | 0.756 | 0.718 | 0.724 | 0.89 | 0.807 | 0.877 | 0.975 | 0.996 | 0.821 | 0.866 | |
| DINOv2Arch=ViT-S/14, Protocol=Linear evaluation of frozen features2023.04 | 0.891 | 0.977 | 0.875 | 0.744 | 0.816 | 0.74 | 0.878 | 0.806 | 0.951 | 0.97 | 0.996 | 0.881 | 0.877 | |
| DINOArch=ViT-B/8, Protocol=Linear evaluation of frozen features2023.04 | 0.851 | 0.972 | 0.869 | 0.703 | 0.766 | 0.706 | 0.867 | 0.796 | 0.932 | 0.954 | 0.976 | 0.817 | 0.851 | |
| MAEArch=ViT-H/14, Protocol=Linear evaluation of frozen features2023.04 | 0.784 | 0.961 | 0.839 | 0.639 | 0.561 | 0.634 | 0.843 | 0.754 | 0.894 | 0.959 | 0.923 | 0.572 | 0.78 |