Skip to content

Latest commit

 

History

History

UNet

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

U-Net: Convolutional Networks for BiomedicalImage Segmentation

Contents

Summary

Convolutional Networks have limited success due to the size and availability of training sets. The typical use of Convolutional Networks is to predict the image as a single class label. However, in many visual tasks, the desired output should include localization, i.e., each pixel is to be assigned a class label.

Architecture

Layout

A more elegant architecture (fully convolutional network) is extended such that it works with very few training images and yeilds precise segmentations.

The main idea is to supplement a usual contracting network by successive layers, whose pooling operators are replaced by upsampling operators. The architecture consists of a Contracting Path and an Expansive Path.

Contracting Path

Consists of repeated application of two 3x3 convolutions followed by a ReLU unit and 2x2 max pooling operation with stride 2 for downsampling.

At each downsampling step, the number of feature chanels are doubled.

Expansive Path

Consists of upsampling of the feature map followed by 2x2 convolutions that halves the number of feature channels, a concatenation with the correspondingly cropped feature map from the contracting path, and two 3x3 convolutions followed by a ReLU.

Final Layer

A 1x1 convolution is used to map each 64 component feature vector to the desired number of classes.

Implementation