CAE has been used by researchers for semantic segmentation. An interesting read is a 2015 paper by Badrinayanan et al, Segnet: a Deep Convolutional Encoder-Decoder Architecture for Image Segmentation (https://arxiv.org/pdf/1511.00561.pdf). The network uses the convolutional layers of VGG16 as its Encoder Network and contains a hierarchy of decoders, one corresponding to each encoder as its Decoder Network. The decoders use the max-pooling indices received from the corresponding Encoder and perform nonlinear upsampling of the input feature map. The link of the paper is given in the See Also section of this recipe along with the GitHub link.