Mastering Computer Vision with TensorFlow 2.x

Encoder-decoder network

An encoder is a neural network that takes an image and generates a feature vector. The decoder does the reverse of the encoder; it takes a feature vector and generates an image from it. The encoder and decoder are trained together to optimize a combined loss function.

An encoder-decoder network results in faster computation in the encoder path as features do not have to be dilated in the encoder path and sharp objects are recovered in the decoder path. The encoder-decoder network contains an encoder module that captures higher semantic information such as shape within the image. It does this by gradually reducing the feature maps. The decoder module, on the other hand, preserves the spatial information and sharper ...