Fig. 8: The proposed Connect-UNets architecture.
From: Connected-UNets: a deep learning architecture for breast mass segmentation
Architecture shows two cascaded encoders (i.e. down-sample pathway) (red arrows) and decoders (i.e. up-sample pathway) (yellow arrows), all alternately connected via skip connection (i.e. dashed lines) and ASPP blocks. An input image is fed to the first block, and a segmentation (binary) image is returned by the last block. Encoders are represented by Convolution layer + Batch Normalization (blue blocks), and Activation layer (dark blue blocks). Decoders are represented by Transposed convolution (green blocks), and Convolution layer + Activation layer (light blue blocks).
