Table 4.
The testing accuracy and training time.
| Dataset | GTSRB | BTSC [3] | MNIST | SVHN [32] | Cifar10 | Cifar100 | |
|---|---|---|---|---|---|---|---|
| Property | Categories | 43 | 62 | 10 | 10 | 10 | 100 |
| Training samples | 39209 | 4591 | 60000 | 73257 | 50000 | 50000 | |
| Testing samples | 12630 | 2498 | 10000 | 26032 | 10000 | 10000 | |
|
| |||||||
| MicronNet | Testing accuracy | 97.686% | 80.388% | 99.49% | 92.16% | 34.829% | 10.333% |
| Training time | 1 s | 1.01 s | 0.55 s | 0.38 s | 0.85 s | 0.85 s | |
|
| |||||||
| MicronNet-BF | Testing accuracy | 99.383% | 82.122% | 99.58% | 93.35% | 78.67% | 49.93% |
| Training time | 1.41 s | 1.48 s | 0.60 s | 0.65 s | 1.29 s | 1.27 s | |
The bold numbers represent the one with the better test accuracy in MicronNet-BF and MicronNet.