Table 1 Comparison of algorithm performance across diverse datasets
From: Interpretable capsule networks via self attention routing on spatially invariant feature surfaces
Model Name | Dataset Name | #Param. (M) | Batch Time (ms) | Training Accuracy (%) | Testing Accuracy (%) | GPU Friendliness | Spatial Interpretability |
|---|---|---|---|---|---|---|---|
ResNet18 | MNIST | 11.18 | 15 | 100.00 | 99.59 | Yes | No |
MSTAR | 11.17 | 17 | 99.79 | 95.60 | Yes | No | |
USCS | 11.17 | 17 | 99.98 | 97.81 | Yes | No | |
AlexNet | MNIST | 57.03 | 6 | 99.63 | 99.35 | Yes | No |
MSTAR | 57.03 | 6 | 98.21 | 88.70 | Yes | No | |
USCS | 57.02 | 7 | 99.17 | 99.43 | Yes | No | |
CapsNet | MNIST | 6.80 | 24 | 99.92 | 99.48 | No | No |
MSTAR | 21.47 | 53 | 99.93 | 93.32 | No | No | |
USCS | 5.90 | 22 | 99.98 | 99.91 | No | No | |
Efficient-CapsNet | MNIST | 1.48 | 12 | 99.57 | 99.39 | Yes | No |
MSTAR | 0.74 | 12 | 99.82 | 93.46 | Yes | No | |
USCS | 1.18 | 11 | 99.98 | 99.95 | Yes | No | |
SISA-CapsNet (Ours) | MNIST | 8.85 | 11 | 98.72 | 98.73 | Yes | Yes |
MSTAR | 153.60 | 52 | 99.75 | 93.35 | Yes | Yes | |
USCS | 32.74 | 18 | 99.99 | 99.93 | Yes | Yes |
