shockwavemk / Face-Generation-with-DCGAN-in-Pytorch

Face Generation in Pytorch using DCGAN with a discriminator and a generator composed of CNN blocks

Face Generation using DCGAN (Pytorch)

This is my implementation of a face generation algorithm, which takes in a large number of face images and generates a new face image made from scratch from a latent vector z (random uniform distribution in [0,1]). Below are some examples of newly generated faces:

There are 2 main components of this model:

1. Discriminator: 4-Layer CNN - Given a face image, distinguishes it as a real or a fake (generated) image

2. Generator: 4-Layer CNN - Given a latent vector z, generates a new face image from learned weights from images in training set. It tries to trick the Dircriminator to think that the generated image is REAL.

Repository

This repository contains:

• Face_Generation.py : Complete code for pre-processing and batching data, building DCGAN, training DCGAN, and visualizing the generated faces

Datasets

Datasets necessary for this implementation can be downloaded by clicking here.

List of Hyperparameters used:

• Batch Size = 128
• Generated Image Size = 32 x 32
• Eength of latent vector z = 100
• Number of Filters in Discriminator's first hidden layer = 32
• Number of Filters in Generator's first hidden layer = 32
• Initial Learning Rate, [beta1, beta2] = 0.0002, [0.5, 0.999]
• Number of Epochs = 50

Sources

I referenced the following sources for building & debugging the final model :

• https://github.com/udacity/P1_Facial_Keypoints/tree/master