Inference model architecture..
| Type | Patch size/stride | Output size |
| Convolution (f21) | 11 × 11/4 | 96 × 55 × 55 |
| Max pooling (f22) | 3 × 3/2 | 96 × 27 × 27 |
| Convolution (f23) | 5 × 5/1 | 256 × 27 × 27 |
| Max pooling (f24) | 3 × 3/2 | 256 × 13 × 13 |
| Convolution (f25) | 3 × 3/1 | 384 × 13 × 13 |
| Convolution (f26) | 3 × 3/1 | 384 × 13 × 13 |
| Convolution (f3) | 3 × 3/1 | 256 × 13 × 13 |
| Max pooling | 3 × 3/2 | 256 × 6 × 6 |
| Full connectivity (f4) | 1 × 4096 | |
| Full connectivity (f5) | 1 × 404 |