gan-face-generate 1.0.1

generate a random face created by DCGAN trained with DigiFace1M database

DCGAN

Deep Convolutional Generative Adversarial Networks (DCGAN) is a class of generative adversarial networks (GAN) introduced by Radford et. al. in the paper Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. The generator and discriminator of DCGAN are contructed of convolutional and convolutional-transpose layers.

Usage

Install the package (Consider installing in a virtual environment)

pip install gan_face_generate

Create your own GAN-generated face

gan_face_generate

Dataset

DigiFace-1M is a generated dataset for training face recognition models. The face images are high quality and thus are qualified to train a GAN network. There are two additional advantages of the dataset:

• Ethical considerations: The use of existing datasets that were collected from web images without explicit consent. In contrast, digital faces in DigiFace-1M are generated using a generative model constructed from high-quality head scans of a limited number of individuals obtained with consent.

• Data bias - DigiFace-1M is generated in a controlled pipeline, so that the racial distribution is guaranteed to be balance.

Results

Loss in Training Process

Generated Images compared with Dataset Images

Noise Arithmetic

Reference

• PyTorch DCGAN TUTORIAL
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• DigiFace-1M: 1 Million Digital Face Images for Face Recognition