Nikronic / CoarseNet

Modified version of U-net based on "Convolutional Networks for Biomedical Image Segmentation (Ronneberger et al., 2015)" paper.
https://arxiv.org/abs/1505.04597
MIT License
25 stars 2 forks source link

Number of encoder and decoder layers are not match. How to apply skip connections? #4

Closed Nikronic closed 5 years ago

Nikronic commented 5 years ago

This is the structure of encoder and decoder of modified version of UNet (our CoarseNet): image

But as we know in the original paper of UNet, each layer of encoder is concatenated to the corresponding layer in decoder. So number of layers should be same.

So I do not know how to apply these skip-connections. One idea is to duplicate concatenation of layers with same size in encoder and decoder. Another idea is so to separate last layers from decoder and use them as final layers without any skip-connections.