DCGAN on Face Image Dataset

This project implements the DCGAN(Deep Convolutional Generative Adversarial Network) on the Face Image Dataset. The goal of this project is to generate fake face images which can be used for data augmentation.

Dataset

The dataset can be downloaded from Face Image Project. The basic structure of the face dataset folder (default name: combined):

   ├── aligned  <-- 29,437 train data
   |   ├── 01_F <-- This subfolder contains the images with gender 'F' and age `01`.
   |   ├── 01_M
   |   ├── 02_F
   |   ├── 02_M
   |   └── ...
   └── valid    <-- 3,681 test data
       ├── 01_F 
       ├── 01_M
       ├── 02_F
       └── ...

The size of each image is 128*128*3 with color channels RGB. In the project, each image was downsampled to 64*63*3.