Table 1.

Structure of wide residual networks

Group name	Output size	Block type = B(3,3)
Conv1	[64 × 64, 16]	[3 × 3, 16]
Conv2	[64 × 64, 16 × k]	$\left[\begin{array}{cc}3\times 3& 16\times k\\ 3\times 3& 16\times k\end{array}\right]\times N$
Conv3	[32 × 32, 32 × k]	$\left[\begin{array}{cc}3\times 3& 32\times k\\ 3\times 3& 32\times k\end{array}\right]\times N$
Conv4	[16 × 16, 64 × k]	$\left[\begin{array}{cc}3\times 3& 64\times k\\ 3\times 3& 64\times k\end{array}\right]\times N$
Avg-pool	[16 × 16, 64 × k]	[8 × 8]
Flatten	[ 1 × 16 × 16 × 64 × k]	N/A
Fc-gender	[1 × 2]	N/A
Fc-age	[1 × 101]	N/A

The Network width is determined by factor k, here k = 8. Groups of convolutions are shown in brackets where N is the number of blocks in group, here N = 2. Downsampling is performed by the first layers in groups conv3 and conv4. The final two fully connected layers perform prediction based on the flatten layer