For biomedical level image segmentation, the design of the U-shaped network structure has a great impact on the backbone design of the subsequent segmentation network.
Structure:
Left side: contracting path, consisting of 4 blocks, all of which are convolution + downsampling structures, ultimately resulting in a 32*32 feature map.
Right: Expansive path, also composed of 4 blocks, is upsampled through deconvolution, and each block is fused with the features of the corresponding block on the left.
During training, increase the weight of the boundary loss.
U-Net Overview
Supongo que te gusta
Origin blog.csdn.net/jiafeier_555/article/details/128907886
Recomendado
Clasificación