Generate a confrontation network | Python implements GAN against a generative neural network (GAN) to generate face images
Table of contents
Process overview
basic introduction
In this project, new face images will be generated using a Generative Adversarial Neural Network (GAN), specifically a Deep Convolutional Generative Adversarial Network structure (DCGAN).
This project uses: mnist dataset http://yann.lecun.com/exdb/mnist/;
CelebA face dataset http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html.
Note that our parameters are best set to multiples of 2, such as 4, 8, 16, 32, 64. This allows tensorflow to optimize during calculations and make your model training faster. Batch size mainly affects the quality of images generated by your GAN. Here are some suggestions for parameter settings: