3D M-Net: Object-Specific 3D Segmentation Network Based on a Single Projection

Table 1

Parametric structure of the essential components.

Layer	Parameters	Output size
ResConv block (k)	3 × 3 × k Conv + BN + ReLU	256^2 × k
	3 × 3 × k Conv + BN
	1 × 1 × k Conv
	ReLU
DownSample block (n)	ResConv block (512)	n^2 × 512
	ResConv block (512)
	2 × 2 max-pooling
UpSample block (n)	3 × 3 Deconv with 2 × 2 stride	n^2 × 512
	ResConv block (512)
	ResConv block (512)
Skip connect (n)	Concatenate + 1 × 1 × 512 Conv	n^2 × 512
Transformation module	1 × 1 × 512 Conv + ReLU	256^3 × 2
	Reshape
	1 × 1 × 1 × 2 Conv
3D Conv block (k)	3 × 3 × 3 × k Conv + BN + ReLU	256^3 × k

k denotes the number of filters in the convolution layers, and n denotes the output resolution of the downsampling or upsampling block.