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

Table 2

Parametric structure of the entire network.

	Layer	Output size
Encoder network	ResConv block (64)	256^2 × 64
	ResConv block (128)	256^2 × 128
	ResConv block (256)	256^2 × 256
	ResConv block (512)	256^2 × 512
	DownSample block (128)	128^2 × 512
	DownSample block (64)	64^2 × 512
	DownSample block (32)	32^2 × 512
	DownSample block (16)	16^2 × 512
	DownSample block (8)	8^2 × 512
Decoder network	UpSample block (16)	16^2 × 512
	UpSample block (32)	32^2 × 512
	UpSample block (64)	64^2 × 512
	UpSample block (128)	128^2 × 512
	UpSample block (256)	256^2 × 512
	Transformation module	256^3 × 2
	3D Conv block (4)	256^3 × 4
	3D Conv block (8)	256^3 × 8
	3D Conv block (16)	256^3 × 16
	1 × 1 × 1 Conv + softmax	256^3 × 15