Face-Generation-in-Pytorch

In this project, I defined and trained a DCGAN on a CelebA dataset. The goal of this project is to get a generator network to generate new images of faces that look as realistic as possble.
Below are some examples of newly generated faces[after only 12 epohcs] :

Model architecture :

• Discriminator: 5-Layer CNN - Given a face image, distinguishes it as a real or a fake (generated) image
• 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.

Hyperparameters :

• Batch Size = 20
• Generated Image Size = 32 x 32
• num_epochs = 12
• learning_rate = 0.0005 with Adam optimizer : [beta1, beta2] = [0.1, 0.999]
• Length 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
• added batch normalization

Training :

Training involved alternating between training the discriminator and the generator.

• epochs : 12

Dataset :

• CelebA dataset [http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html]