Table 1 The specific configuration of the NR-Network
From: Noise-resistant network: a deep-learning method for face recognition under noise
Name | Type | Filter size/stride | Output size | #Params |
|---|---|---|---|---|
Conv1 | Convolution | 5 × 5/1 | 60 × 60 × 20 | 500 |
Pool1 | Max pooling | 3 × 3/3 | 20 × 20 × 20 | – |
Conv2 | Convolution | 3 × 3/1 | 18 × 18 × 40 | 7 K |
Pool2 | Max pooling | 2 × 2/2 | 9 × 9 × 40 | – |
Conv31 | Convolution | 3 × 3/1 | 7 × 7 × 40 | 14 K |
Pool3 | Max pooling | 2 × 2/2 | 5 × 5 × 40 | – |
Conv321 | Convolution | 1 × 1/1 | 5 × 5 × 20 | 800 |
Conv322 | Convolution | 3 × 3/1 | 5 × 5 × 20 | 7 K |
Conv323 | Convolution | 5 × 5/11 | 5 × 5 × 20 | 19 K |
Conc1 | Concat | – | 5 × 5 × 60 | – |
Conv4 | Convolution | 3 × 3/1 | 3 × 33 × 40 | 21 K |
Pool4 | Max pooling | 2 × 2/1 | 3 × 3 × 60 | – |
Conc2 | Concat | – | 3 × 3 × 100 | – |
Pool51 | Average pooling | 3 × 3/1 | 1 × 1 × 100 | – |
Pool52 | Average pooling | 5 × 5/5 | 2 × 2 × 40 | – |
Pool53 | Average pooling | 3 × 3/2 | 2 × 2 × 60 | – |
Conc3 | Concat | – | 500 |  |
Fc5 | Fully connected | Â | 256 | 125Â K |
Total | – | – | – | 194 K |