Table 1:

The detail structure and parameters of the i3D network that we are using.

Stage	Details	Output Size
Conv1	5 × 7 × 7, 64, stride 1, 2, 2	32 × 112 × 112 × 64
Maxpool1	2 × 3 × 3, stride 2, 2, 2	32 × 56 × 56 × 64
Res2	$\left(\begin{array}{l}3\times 1\times 1,64\hfill \\ 1\times 3\times 3,64\hfill \\ 1\times 1\times 1,256\hfill \end{array}\right)\times 3$	16 × 56 × 56 × 256
Maxpool2	2 × 3 × 3, stride 2, 2, 2	8 × 28 × 28 × 256
Res3	$\left(\begin{array}{l}3\times 1\times 1,128\hfill \\ 1\times 3\times 3,128\hfill \\ 1\times 1\times 1,512\hfill \end{array}\right)\times 4$	8 × 28 × 28 × 512
Res4	$\left(\begin{array}{l}3\times 1\times 1,256\hfill \\ 1\times 3\times 3,256\hfill \\ 1\times 1\times 1,1024\hfill \end{array}\right)\times 23$	8 × 14 × 14 × 1024
Res5	$\left(\begin{array}{l}3\times 1\times 1,512\hfill \\ 1\times 3\times 3,512\hfill \\ 1\times 1\times \mathrm{1,2048}\hfill \end{array}\right)\times 3$	8 × 7 × 7 × 2048