Fig. 2

Overview of network architecture consisting of four convolutional blocks and a final dense layer. Layers in each block are indicated above each arrow. The final convolutional block (red arrow) deviates from the first three blocks (yellow arrows) by not containing a max pooling layer. The information above each box represents the size of the resulting activation maps, which are difference between the classification and regression networks. Values used are given in Additional file 1: Table S2 and S3