Just use its main idea, the model is not the same as the origin unet++ mentioned in paper paper : UNet++: A Nested U-Net Architecture for Medical Image Segmentation
y0 y1 y2
↑ ↑ ↑
input → x1 → x11 → x12 → x
↘ ↗ ↗ ↗
x2 → x21 → x
↘ ↗ ↗
x3 → x
↘ ↗
x4
Notice: For ease of drawing , I have omitted some of the connections ,including x2 to x , x1 to x12 , x1 to x , x11 to x