Generative Adversarial Networks Projects

The discriminator network

Similar to the generator network, the discriminator network is a deep convolutional neural network, containing a series of downsampling convolutional layers. The downsampling layers generate feature maps from images, whether they are real images from the real data distribution or images generated by the generator network. Then, we concatenate the feature maps to the text embedding. We use compression and spatial replication to convert a text embedding to the format required for concatenation. The spatial compression and replication include a fully connected layer, which is used to compress the text embedding to an  ...