Convolutional Neural Network

Neural Network s that use convolutions. The output of a convolutional layer is computed by sliding a dimensional kernel over the input.

Convolutions can be used to reduce the spacial size. In this case the part of the network is called the encoder.

Transpose Convolutions #

Are used to increase spatial size by using the kernel to upsample.

Encoder-Decoder #

Transpose convolutional layers can be used in combination with convolutional layers to create a encoder-decoder structure. Where the output is in the same size as the input but the network can do cool things inbetween.