This repository contains code for training a Deep Convolutional Generative Adversarial Network (DCGAN) using PyTorch. The implementation of the DCGAN was carried out with the primary objective of gaining a deeper understanding of GAN training. It was trained on the Pokémon dataset.
The model will be trained with the following hyperparameters, which are specified in config.yaml:
- num_epochs: 1000
- batch_size: 128
- image_size: 64
- latent_dim: 100
- ngf: 64
- ndf: 64
- nc: 3
- lr: 0.0002
- beta1: 0.5
- beta2: 0.999
The training process, using this hyperparameters, takes approximately one hour to complete.