Table 3.

Output feature maps at different stages of our CMixNet architecture.

Stage	Output	Weights
Initial input size	96 × 96 × 96, 24	3 × 3 × 3, 24
First CMixNet block	48 × 48 × 48, 48	$\left\{\begin{array}{c}1\times 1\times 1,24\\ 3\times 3\times 3,24,\\ 1\times 1\times 1,32\end{array}\begin{array}{c}(\mathrm{stride}-2)\\ C=32\end{array}\right\}\times 2$
Second CMixNet block	24 × 24 × 24, 72	$\left\{\begin{array}{c}1\times 1\times 1,48\\ 3\times 3\times 3,48,\\ 1\times 1\times 1,56\end{array}\begin{array}{c}(\mathrm{stride}-2)\\ C=32\end{array}\right\}\times 2$
Third CMixNet block	12 × 12 × 12, 96	$\left\{\begin{array}{c}1\times 1\times 1,72\\ 3\times 3\times 3,72,\\ 1\times 1\times 1,80\end{array}\begin{array}{c}(\mathrm{stride}-2)\\ C=32\end{array}\right\}\times 2$
Fourth CMixNet block	6 × 6 × 6, 120	$\left\{\begin{array}{c}1\times 1\times 1,96\\ 3\times 3\times 3,96,\\ 1\times 1\times 1,104\end{array}\begin{array}{c}(\mathrm{stride}-2)\\ C=32\end{array}\right\}\times 2$
Upsampling/Deconv. 1	12 × 12 × 12, 216	2 × 2 × 2, 216
Fifth CMixNet block	12 × 12 × 12, 152	$\left\{\begin{array}{c}1\times 1\times 1,128\\ 3\times 3\times 3,128,\\ 1\times 1\times 1,136\end{array}\begin{array}{c}(\mathrm{stride}-2)\\ C=32\end{array}\right\}\times 2$
Upsampling/Deconv. 2	24 × 24 × 24, 224	2 × 2 × 2, 152
Sixth CMixNet block	24 × 24 × 24, 248	$\left\{\begin{array}{c}1\times 1\times 1,224\\ 3\times 3\times 3,224,\\ 1\times 1\times 1,232\end{array}\begin{array}{c}(\mathrm{stride}-2)\\ C=32\end{array}\right\}\times 2$
Output	24 × 24 × 24, 15	Dropout, p = 0.5 1 × 1 × 1, 64 1 × 1 × 1, 15