Geometric Probing on Three-Cell Experiment 1.0 (test)
93.4DistanceLOOPE (XG + Xc)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LOOPE (XG + Xc)Backbone=DeiT-Base, Positional Encoding=LOOPE2025.04 | 93.4 | 95.8 | 93.3 | 94.6 | 94.3 | |
| LFF(Fourier)Backbone=DeiT-Base, Positional Encoding=Learned Frequency Features2025.04 | 93.1 | 96.7 | 92.4 | 93.9 | 94 | |
| Inception-V3Model Type=CNN2025.04 | 92.8 | 96.1 | 93.7 | 95.3 | 94.5 | |
| 1D SinusoidBackbone=DeiT-Base, Positional Encoding=1D Sinusoid2025.04 | 90.9 | 94.9 | 91.3 | 95.3 | 93.1 | |
| ResNet-50Model Type=CNN2025.04 | 90.8 | 85.6 | 89.7 | 92.9 | 89.8 | |
| Static (XG)Backbone=DeiT-Base, Positional Encoding=Static Grid2025.04 | 88.8 | 95 | 89.9 | 93.8 | 91.9 | |
| LearnableBackbone=DeiT-Base, Positional Encoding=Learnable2025.04 | 85.6 | 89.3 | 84 | 92.2 | 87.8 | |
| 2D SinusoidBackbone=DeiT-Base, Positional Encoding=2D Sinusoid2025.04 | 83.6 | 60 | 72.6 | 92.3 | 77.1 | |
| RPEBackbone=DeiT-Base, Positional Encoding=Relative Positional Encoding2025.04 | 73.7 | 62.6 | 71 | 73.9 | 70.3 | |
| CPEBackbone=DeiT-Base, Positional Encoding=Conditional Positional Encoding2025.04 | 72.8 | 63.2 | 72.6 | 75 | 70.9 | |
| No PEBackbone=DeiT-Base, Positional Encoding=None2025.04 | 61.9 | 53.1 | 60.1 | 62.6 | 59.4 |