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.
Install the package (Consider installing in a virtual environment)
pip install gan_face_generateCreate your own GAN-generated face
gan_face_generateThe package works with python >= 3.9, <3.12. Check CI Result
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.
