According to the paper, it should be conv(8)-conv(8)-pool(2)- conv(16)-conv(16)-pool(2)- conv(32)-conv(32)-pool(8)[ 8-8-16-16-32-32]. The code is 8-16-32, and each layer may be missing a convolution kernel. Perhaps the wrong version of the code is transmitted, or the simplified version of the network can achieve similar results.
According to the paper, it should be conv(8)-conv(8)-pool(2)- conv(16)-conv(16)-pool(2)- conv(32)-conv(32)-pool(8)[ 8-8-16-16-32-32]. The code is 8-16-32, and each layer may be missing a convolution kernel. Perhaps the wrong version of the code is transmitted, or the simplified version of the network can achieve similar results.