DCGAN
lastmod 2026-11-10created 2021-11-302 edits
DCGAN (Deep Convolutional GAN) is a GAN architecture whose generator and discriminator are built from convolutional neural networks. It is characterized by design guidelines that stabilize GAN training, such as using strided convolutions and transposed convolutions instead of pooling layers, applying batch normalization, and removing fully connected layers. This page collects the face image generation example from the official PyTorch tutorial.