Figure 3

Illustration of the network architecture. (a) The relationship between the network input and output. The network takes a 512 \(\times\) 512 image as input, and the output of the network is divided into two branches. The output target values for the classification branch are stored in a 3D tensor of size 64 \(\times\) 64 \(\times\) 2, which is used to judge whether the cell contains convex lines and whether the cell contains concave lines. The output target values for the location branch are stored in a 3D tensor of size 64 \(\times\) 64 \(\times\) 8, which is used to predict the locations of two sets of stair lines. (b) The architecture of our network. The backbone contains a focus module, several SE-ResNeXt blocks with dilated convolution and an ASPP module.