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

Figure 1

Schematic of the network architecture. The encoder network consists of residual convolution (blue arrow) and downsampling (yellow arrow) processes. The decoder network comprises upsampling (green arrow), a feature transformation model (purple arrow), and 3D convolution (red arrow). The upsampling and downsampling blocks share features through skip connections (dashed arrows). Finally, the probability vector is output through the softmax layer. The number next to the feature map indicates the spatial resolution and number of channels of the feature maps.