Table 3 Accuracy per stage and statistical features of their filters for EfficientNet-B0 trained on ImageNet.
From: Towards a universal mechanism for successful deep learning
EfficientNet-B0 on ImageNet | |||||||
|---|---|---|---|---|---|---|---|
Stage | \({N}_{f}\) | Filter’s outputs | FC size | Accuracy | \(n\) | \({N}_{c}\) | \({C}_{s}\) |
9 | 1280 | 1 × 1 | 1280 | 0.750 | 1057.7 | 3.5 | 4.4 |
7 | 192 | 1 × 1 | 192 | 0.489 | 5729.6 | 8.3 | 3.2 |
5 | 80 | 2 × 2 | 320 | 0.187 | 7168.0 | 8.9 | 2.6 |
4 | 40 | 4 × 4 | 640 | 0.136 | 13,416.9 | 16.5 | 2.5 |
3 | 24 | 8 × 8 | 1536 | 0.065 | 6471.0 | 13.2 | 1.9 |
1 | 32 | 16 × 16 | 8192 | 0.022 | 50,381.9 | 8.5 | 7.4 |
