1 Segnet A Deep Convolutional Encoder Decoder Architecture

arXiv:2009.12942v1 [cs.CV] 27 Sep 2020

A similar architecture to [33] was proposed in SegNet [34], where the authors used VGG [29] as the backbone encoder, removed the fully connected layers, and added a symmetric decoder structure. The main difference is that every decoder layer uses the max pooling indices from the corresponding encoder layer, as opposed to concatenating it.

  Architecture, Decoder, Encoder, Gensets