Table 1 Detailed architecture of the ResNeXt-50 model.

Input

–

300 × 300

Conv1

7 × 7 Conv, 64, stride = 2

150 × 150

Max pool

3 × 3 max pool, stride = 2

–

Conv2_x

\(\left\{ {\begin{array}{*{20}c} {1 \times 1 \;{\text{Conv}}, 128} \\ {3 \times 3\;{\text{Conv}}, 128} \\ {1 \times 1\;{\text{Conv}}, 256} \\ \end{array} } \right\} \times 3, \;{\text{C}} = 32\)

75 × 75

Down-sampling

\(1 \times 1\) Conv, stride = 2

–

Conv3_x

\(\left\{ {\begin{array}{*{20}c} {1 \times 1 \;{\text{Conv}}, 256} \\ {3 \times 3\;{\text{Conv}}, 256} \\ {1 \times 1\;{\text{Conv}}, 512} \\ \end{array} } \right\} \times 4,\;{\text{C}} = 32\)

38 × 38

Down-sampling

\(1 \times 1\) Conv, stride = 2

–

Conv4_x

\(\left\{ {\begin{array}{*{20}c} {1 \times 1\;{\text{Conv}}, 512} \\ {3 \times 3\;{\text{Conv}}, 512} \\ {1 \times 1\;{\text{Conv}}, 1024} \\ \end{array} } \right\} \times 6, \;{\text{C}} = 32\)

19 × 19

Down-sampling

\(1 \times 1\) Conv, stride = 2

–

Conv5_x

\(\left\{ {\begin{array}{*{20}c} {1 \times 1\;{\text{Conv}}, 1024} \\ {3 \times 3\;{\text{Conv}}, 1024} \\ {1 \times 1\;{\text{Conv}}, 2048} \\ \end{array} } \right\} \times 3, \;{\text{C}} = 32\)

10 × 10

Global average pool

fc, Softmax

1 × 1