3D Image Classification on MedMNIST 3D v2 (test)
0.998Organ AccuracyTransVW
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TransVWBackbone=ResNet-182024.06 | 0.998 | 0.898 | 0.731 | 0.835 | 0.905 | 0.811 | 0.863 | — | — | — | — | — | — | — | |
| CDSSL-P3DBackbone=ResNet-182024.06 | 0.998 | 0.908 | 0.754 | 0.88 | 0.947 | 0.835 | 0.888 | — | — | — | — | — | — | — | |
| ResNet-18+3DBackbone=ResNet-182024.06 | 0.996 | 0.863 | 0.712 | 0.827 | 0.874 | 0.82 | 0.848 | — | — | — | — | — | — | — | |
| UniMiSSBackbone=PVT-small2024.06 | 0.996 | 0.894 | 0.724 | 0.853 | 0.927 | 0.847 | 0.874 | — | — | — | — | — | — | — | |
| DINOBackbone=ResNet-182024.06 | 0.995 | 0.89 | 0.707 | 0.847 | 0.918 | 0.81 | 0.861 | — | — | — | — | — | — | — | |
| SimSiamBackbone=ResNet-182024.06 | 0.995 | 0.874 | 0.738 | 0.837 | 0.876 | 0.765 | 0.848 | — | — | — | — | — | — | — | |
| PCRLv2Backbone=ResNet-182024.06 | 0.995 | 0.894 | 0.74 | 0.853 | 0.93 | 0.798 | 0.868 | — | — | — | — | — | — | — | |
| CDSSL-P3DBackbone=PVT-small2024.06 | 0.993 | 0.93 | 0.761 | 0.857 | 0.96 | 0.912 | 0.902 | — | — | — | — | — | — | — | |
| Rand. init.Backbone=PVT-small2024.06 | 0.98 | 0.876 | 0.651 | 0.824 | 0.907 | 0.77 | 0.835 | — | — | — | — | — | — | — | |
| PCmBackbone=ViT-B/16, GPU memory=33 GB, Time=40 min, Training Protocol=Full fine-tuning2026.03 | 0.975 | 0.881 | 0.527 | 0.815 | 0.919 | 0.927 | 0.841 | 1 | 0.918 | 0.728 | 0.866 | 0.946 | 0.96 | 0.903 | |
| PCgBackbone=ViT-L/16, GPU memory=87 GB, Time=120 min, Training Protocol=Full fine-tuning2026.03 | 0.975 | 0.87 | 0.537 | 0.838 | 0.926 | 0.901 | 0.841 | 0.999 | 0.94 | 0.738 | 0.877 | 0.958 | 0.971 | 0.914 | |
| 3DBackbone=ViT-L/16, GPU memory=87 GB, Time=120 min, Training Protocol=Full fine-tuning2026.03 | 0.972 | 0.883 | 0.584 | 0.848 | 0.944 | 0.93 | 0.86 | 0.998 | 0.936 | 0.766 | 0.889 | 0.967 | 0.973 | 0.922 | |
| PCgBackbone=ViT-B/16, GPU memory=33 GB, Time=40 min, Training Protocol=Full fine-tuning2026.03 | 0.97 | 0.885 | 0.531 | 0.848 | 0.927 | 0.933 | 0.849 | 0.999 | 0.946 | 0.731 | 0.891 | 0.938 | 0.969 | 0.912 | |
| PCmBackbone=ViT-L/16, GPU memory=87 GB, Time=120 min, Training Protocol=Full fine-tuning2026.03 | 0.97 | 0.869 | 0.571 | 0.83 | 0.936 | 0.919 | 0.849 | 0.999 | 0.932 | 0.739 | 0.875 | 0.951 | 0.966 | 0.91 | |
| ViViTTraining Protocol=Full fine-tuning2026.03 | 0.965 | 0.872 | 0.51 | 0.829 | 0.94 | 0.905 | 0.837 | 0.999 | 0.883 | 0.699 | 0.863 | 0.951 | 0.935 | 0.888 | |
| PCmBackbone=ViT-S/16, GPU memory=17 GB, Time=15 min, Training Protocol=Full fine-tuning2026.03 | 0.962 | 0.874 | 0.564 | 0.854 | 0.934 | 0.918 | 0.851 | 0.999 | 0.934 | 0.738 | 0.892 | 0.949 | 0.951 | 0.91 | |
| PCgBackbone=ViT-S/16, GPU memory=17 GB, Time=15 min, Training Protocol=Full fine-tuning2026.03 | 0.962 | 0.878 | 0.519 | 0.846 | 0.936 | 0.922 | 0.844 | 0.999 | 0.932 | 0.722 | 0.885 | 0.941 | 0.957 | 0.906 | |
| 3DBackbone=ViT-B/16, GPU memory=33 GB, Time=40 min, Training Protocol=Full fine-tuning2026.03 | 0.962 | 0.877 | 0.526 | 0.851 | 0.92 | 0.917 | 0.842 | 0.998 | 0.927 | 0.742 | 0.89 | 0.949 | 0.961 | 0.911 | |
| 3DBackbone=ViT-S/16, GPU memory=17 GB, Time=15 min, Training Protocol=Full fine-tuning2026.03 | 0.957 | 0.878 | 0.55 | 0.856 | 0.919 | 0.927 | 0.848 | 0.998 | 0.938 | 0.76 | 0.897 | 0.949 | 0.965 | 0.918 | |
| PlaneCycle (PCm)Backbone=ViT-B/16, Evaluation protocol=Linear probing, GPU memory=4.3 GB, Training Time=20 min2026.03 | 0.942 | 0.874 | 0.471 | 0.826 | 0.906 | 0.849 | 0.812 | 99.8 | 91 | 65.5 | 85.6 | 90.3 | 87.4 | 86.6 | |
| PlaneCycle (PCg)Backbone=ViT-B/16, Evaluation protocol=Linear probing, GPU memory=4.3 GB, Training Time=20 min2026.03 | 0.934 | 0.886 | 0.514 | 0.834 | 0.909 | 0.857 | 0.82 | 99.7 | 92.7 | 69.4 | 86.9 | 90.5 | 88.9 | 87.8 | |
| PlaneCycle (PCm)Backbone=ViT-L/16, Evaluation protocol=Linear probing, GPU memory=7.0 GB, Training Time=60 min2026.03 | 0.915 | 0.868 | 0.474 | 0.833 | 0.892 | 0.834 | 0.802 | 99.5 | 90.9 | 65.6 | 84 | 85.2 | 86.7 | 85.3 | |
| PlaneCycle (PCm)Backbone=ViT-S/16, Evaluation protocol=Linear probing, GPU memory=2.5 GB, Training Time=10 min2026.03 | 0.902 | 0.864 | 0.458 | 0.834 | 0.882 | 0.83 | 0.795 | 99.4 | 88.9 | 62.9 | 85 | 87.6 | 85.3 | 84.8 | |
| R-ACSBackbone=ResNet-50, Evaluation protocol=fully fine-tuned2026.03 | 0.889 | 0.841 | 0.517 | 0.758 | 0.858 | 0.709 | 0.762 | 99.4 | 88.6 | 75 | 82.8 | 91.2 | 71.9 | 84.8 | |
| R-3DBackbone=ResNet-50, Evaluation protocol=fully fine-tuned2026.03 | 0.883 | 0.847 | 0.494 | 0.745 | 0.918 | 0.795 | 0.78 | 99.4 | 87.5 | 72.5 | 82.8 | 90.7 | 85.1 | 86.3 | |
| 2DBackbone=ViT-B/16, Evaluation protocol=Linear probing, GPU memory=4.3 GB, Training Time=20 min2026.03 | 0.883 | 0.858 | 0.467 | 0.822 | 0.895 | 0.852 | 0.79 | 98.9 | 88.9 | 63.6 | 85 | 88.9 | 86.7 | 85.1 | |
| PlaneCycle (PCg)Backbone=ViT-S/16, Evaluation protocol=Linear probing, GPU memory=2.5 GB, Training Time=10 min2026.03 | 0.882 | 0.87 | 0.52 | 0.833 | 0.888 | 0.853 | 0.808 | 99.2 | 90.7 | 69.6 | 85.9 | 86.2 | 88.6 | 86.7 | |
| 2DBackbone=ViT-B/16, GPU memory=33 GB, Time=40 min, Training Protocol=Full fine-tuning2026.03 | 0.878 | 0.868 | 0.449 | 0.805 | 0.894 | 0.885 | 0.797 | 0.99 | 0.908 | 0.611 | 0.806 | 0.82 | 0.941 | 0.846 | |
| PlaneCycle (PCg)Backbone=ViT-L/16, Evaluation protocol=Linear probing, GPU memory=7.0 GB, Training Time=60 min2026.03 | 0.877 | 0.855 | 0.517 | 0.826 | 0.9 | 0.865 | 0.807 | 99.2 | 90.9 | 69.3 | 85.2 | 90.1 | 88.9 | 87.3 | |
| 2DBackbone=ViT-S/16, GPU memory=17 GB, Time=15 min, Training Protocol=Full fine-tuning2026.03 | 0.858 | 0.855 | 0.441 | 0.808 | 0.901 | 0.882 | 0.791 | 0.986 | 0.891 | 0.622 | 0.817 | 0.842 | 0.93 | 0.848 | |
| 2DBackbone=ViT-L/16, GPU memory=87 GB, Time=120 min, Training Protocol=Full fine-tuning2026.03 | 0.846 | 0.868 | 0.438 | 0.801 | 0.897 | 0.868 | 0.786 | 0.986 | 0.902 | 0.646 | 0.811 | 0.853 | 0.905 | 0.85 | |
| 2DBackbone=ViT-S/16, Evaluation protocol=Linear probing, GPU memory=2.5 GB, Training Time=10 min2026.03 | 0.845 | 0.884 | 0.482 | 0.839 | 0.887 | 0.838 | 0.796 | 98.6 | 89.2 | 64.1 | 87 | 84.7 | 85.7 | 84.9 | |
| 3DBackbone=ViT-L/16, Evaluation protocol=Linear probing, GPU memory=7.0 GB, Training Time=60 min2026.03 | 0.833 | 0.856 | 0.47 | 0.779 | 0.889 | 0.791 | 0.77 | 98.6 | 87.5 | 69.9 | 77.6 | 82.4 | 79.5 | 82.6 | |
| 2DBackbone=ViT-L/16, Evaluation protocol=Linear probing, GPU memory=7.0 GB, Training Time=60 min2026.03 | 0.815 | 0.881 | 0.443 | 0.824 | 0.888 | 0.816 | 0.778 | 98.2 | 90.1 | 63.3 | 86.2 | 81.5 | 86.1 | 84.2 | |
| 3DBackbone=ViT-B/16, Evaluation protocol=Linear probing, GPU memory=4.3 GB, Training Time=20 min2026.03 | 0.776 | 0.831 | 0.546 | 0.79 | 0.887 | 0.805 | 0.772 | 97.8 | 81.7 | 70.5 | 80.5 | 77.5 | 82.3 | 81.7 | |
| 3DBackbone=ViT-S/16, Evaluation protocol=Linear probing, GPU memory=2.5 GB, Training Time=10 min2026.03 | 0.75 | 0.815 | 0.508 | 0.772 | 0.887 | 0.756 | 0.748 | 97.4 | 77.7 | 70.5 | 77.7 | 73.9 | 74.5 | 78.6 |