Layer Type | Sizes | |
---|---|---|
0 | Convolutional | M^{(0)}=d, T^{(0)}=T, |
S^{(0)}=(⌊T/6⌋+1)·2+1 | ||
1 | Max pooling | M^{(1)}=M^{(0)}·6, T^{(1)}=T^{(0)}, |
R^{(1)}=3 if applicable, else R^{(1)}=1 | ||
2 | Convolutional | M^{(2)}=M^{(1)}, T^{(2)}=⌈T^{(1)}/R^{(1)}⌉, |
S^{(2)}=(⌊T/12⌋+1)·2+1 | ||
M^{(3)}=M^{(2)}·3, T^{(3)}=T^{(2)} | ||
3 | Fully connected | N^{(3)}=M^{(3)}·T^{(3)} |
4 | Fully connected | \(N^{(4)} = \lfloor \sqrt {N^{(3)} \cdot N^{(5)}} \rfloor \) |
5 | Output | N^{(5)}=n |