Skip to main content

Table 1 The structure of 3D-ResNet

From: Classifying the tracing difficulty of 3D neuron image blocks based on deep learning

Stage Component Output size
Convolution \(3\times 3\times 3\), 64, stride(1,2,2) \(32\times 32\times 32\)
Max pooling \(3\times 3\times 3\), 64, stride(2,2,2) \(16\times 16\times 16\)
Residual layer 1 Dropout=0.2, unit-A(64), unit-A(64) \(16\times 16\times 16\)
Residual layer 2 Dropout=0.2, unit-B(128), unit-A(128) \(8\times 8\times 8\)
Residual layer 3 Dropout=0.2, unit-B(256), unit-A(256) \(4\times 4\times 4\)
Residual layer 4 Dropout=0.2, unit-B(512), unit-A(512) \(2\times 2\times 2\)
Average pooling \(2\times 2\times 2\), stride(2,2,2) \(1\times 1\times 1\)
Classification layer Fully-connected, softmax 2
  1. * Unit-A(n) consists of a 3\(\times\)3\(\times\)3 convolution with a 1\(\times\)1\(\times\)1 stride and n channels, a batch normalization, and an activation function (ReLU). Unit-B(n) has the same structure as unit-A(n) but a different convolution step size 2\(\times\)2\(\times\)2