Table 7 Overview of classifiers with similar structure but different depth from Classifer-inRAC

Conv

7 × 7, 64, Stride = 2

Norm

Batch normalization

Pool

3 × 3 Max pooling, Stride = 2

RA-Conv

\(\left[ \begin{gathered} 3 \times 3,\;\;64 \hfill \\ 3 \times 3,\;\;64 \hfill \\ \end{gathered} \right] \times 2\)

\(\left[ \begin{gathered} 3 \times 3,\;\;64 \hfill \\ 3 \times 3,\;\;64 \hfill \\ \end{gathered} \right] \times 3\)

\(\left[ \begin{gathered} 3 \times 3,\;\;64 \hfill \\ 3 \times 3,\;\;64 \hfill \\ \end{gathered} \right] \times 2\)

RA-Conv

\(\left[ {\begin{array}{*{20}c} {3 \times 3,\;\;128} \\ {3 \times 3,\;\;128} \\ \end{array} } \right] \times 2\)

\(\left[ \begin{gathered} 3 \times 3,\;\;128 \hfill \\ 3 \times 3,\;\;128 \hfill \\ \end{gathered} \right] \times 4\)

\(\left[ \begin{gathered} 3 \times 3,\;\;128 \hfill \\ 3 \times 3,\;\;128 \hfill \\ \end{gathered} \right] \times 3\)

RA-Conv

\(\left[ \begin{gathered} 3 \times 3,\;\;256 \hfill \\ 3 \times 3,\;\;256 \hfill \\ \end{gathered} \right] \times 2\)

\(\left[ \begin{gathered} 3 \times 3,\;\;256 \hfill \\ 3 \times 3,\;\;256 \hfill \\ \end{gathered} \right] \times 6\)

\(\left[ \begin{gathered} 3 \times 3,\;\;256 \hfill \\ 3 \times 3,\;\;256 \hfill \\ \end{gathered} \right] \times 5\)

RA-Conv

\(\left[ \begin{gathered} 3 \times 3,\;\;512 \hfill \\ 3 \times 3,\;\;512 \hfill \\ \end{gathered} \right] \times 2\)

\(\left[ \begin{gathered} 3 \times 3,\;\;512 \hfill \\ 3 \times 3,\;\;512 \hfill \\ \end{gathered} \right] \times 3\)

\(\left[ \begin{gathered} 3 \times 3,\;\;512 \hfill \\ 3 \times 3,\;\;512 \hfill \\ \end{gathered} \right] \times 2\)

Global average pooling (GAP)

Softmax