type | patch size/stride | output size | 1 × 1 | 3 × 3 | two sequential 3 × 3 | maxpool |
convolution | 7 × 7/2 | 112 × 112 × 32 |
|
|
|
|
max pool | 3 × 3/2 | 56 × 56 × 32 |
|
|
|
|
batch norm |
| 56 × 56 × 32 |
|
|
|
|
convolution | 3 × 3/1 | 56 × 56 × 128 | 64 |
|
|
|
batch norm |
| 56 × 56 × 128 |
|
|
|
|
max pool | 3 × 3/2 | 28 × 28 × 128 |
|
|
|
|
SI 1 |
| 28 × 28 × 256 | 64 | 96 | 32 | 64 |
SI 2 |
| 28 × 28 × 416 | 112 | 128 | 64 | 112 |
max pool | 3 × 3/2 | 14 × 14 × 416 |
|
|
|
|
SI 3 |
| 14 × 14 × 576 | 160 | 160 | 96 | 160 |
SI 4 |
| 14 × 14 × 740 | 208 | 196 | 128 | 208 |
max pool | 3 × 3/2 | 7 × 7 × 740 |
|
|
|
|
SI 5 |
| 7 × 7 × 900 | 256 | 228 | 160 | 256 |
SI 6 |
| 7 × 7 × 1010 | 304 | 206 | 196 | 304 |
average pool | 7 × 7/1 | 1 × 1 × 1010 |
|
|
|
|
dropout (40%) |
| 1 × 1 × 1010 |
|
|
|
|
linear |
| 1 × 1 × class num |
|
|
|
|
softmax |
| 1 × 1 × class num |
|
|
|
|