Fig. 2From: Hierarchical combinatorial deep learning architecture for pancreas segmentation of medical computed tomography cancer imagesArchitecture of our network. Part (a) shows the main structure of our network. In the feature extract network, each color box stands for a conv layer, and the conv layers are divided into 5 different stages in different colors. Furtherly, each stage is connected to a features fusing layer. After that, an up-sampling structure is used to de-convolute the extracted features to the initial size. Part (b) and (c) separately show the up-sampling structure of the RCF network and oursBack to article page