Multipath Lightweight Deep Network Using Randomly Selected Dilated Convolution

. 2021 Nov 26;21(23):7862. doi: 10.3390/s21237862

Algorithm 1 MLDN in nth denseblock in CondenseNet

Input: $x_{n}$
for $l = 1$ to $L_{n}$ do
$x_{n, l, 2 p - 1}, x_{n, l, 2 p} = split (x_{n, l})$ where $1 \leq p \leq P / 2$
for $p = 1$ to $P / 2$ do
$y_{n, l, 2 p - 1} = cat (x_{n, l, 2 p - 1}, H_{n, l, 2 p} (x_{n, l, 2 p}))$
$y_{n, l, 2 p} = cat (x_{n, l, 2 p}, H_{n, l, 2 p - 1} (x_{n, l, 2 p - 1}))$
end for
$x_{n, l + 1} = cat (y_{n, l, 1}, y_{n, l, 2}, \dots, y_{n, l, 2 p - 1}, y_{n, l, 2 p})$
end for
Output: $x_{n + 1} = x_{n, L_{n} + 1}$