Skip to main content
. 2021 May 19;11(5):902. doi: 10.3390/diagnostics11050902

Figure 1.

Figure 1

U-Net-like segmentation architecture. The input image is first downscaled in four steps, then the low-resolution feature map is upscaled to the resolution of the input image. Additional feature maps from the downscaling part are incorporated at each upscaling layer.