FCN

Concept

A Fully Convolutional neural network (FCN) is a normal convolutional neural networks(CNN), where the last fully connected layer is substituted by another convolution layer with a large "receptive field". The idea is to capture the global context of the scene. The model architecture see below figure:

Implementation

Our FCN classifier includes a custom implemented version of Dice Score and also we have added multiple layers and changed some in the original implementation. We found these changes had worked in favor of better dice scores. we have mainly used the up sampling functions from [2]. Then parameters where fine tuned using a manually implemented grid search.

Reference:

1. Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 3431-3440).
2. gitHub Repository https://github.com/JihongJu/keras-fcn