《Dual Encoding U-Net for Retinal Vessel Segmentation》閱讀筆記-MICCAI2019

做者：Bo Wang1,2, Shuang Qiu2, and Huiguang He1,2,3

目的：Retinal Vessel Segmentation is an essential step for the early diagnosis of eye-related diseases, such as diabetes and hypertension. Segmentation of blood vessels requires both sizeable receptive field and rich spatial information.

方法：Dual Encoding U-Net (DEU-Net), 空間information和上下文information

該結構圖outputpatches居然和input同樣。

1. Spatial Path，連續四個卷積，用了大stride：7，12，9，6.參考了Large kernel matters-improve semantic segmentation by global convolutional network.， CVPR2017。
2. Context Path，就是inception block，google提出的
3. Attention Skip Module，最簡單的attention方式處理
4. Feature Fusion Module，這個方式我看到過，不知道爲何叫作feature fusion，其實連結處就是和attention residual for image classification那篇文章同樣.
5. Multiscale Predict Module，這個模塊沒看到過，主要是pixel shuffle（參考CVPR2016 Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network）這個操做。 

 

試驗結果：略