pytorch-dcgan

Introduction

Pytorch implementation of DCGAN with high extensibility. You can easily extend this code to add more models and datasets.

Dependencies

• Python 3
• Pytorch 0.4.0

Usage

Train

The default config in config.py is set for 'mnist' dataset.

To train the dcgan on 'mnist':

python main.py

To train the dcgan on 'celeba':

python main.py --dataset celeba --dataset_root [celeba dataset folder] --c_dim 3 --height 64 --width 64

To train the dcgan on your own dataset:

• implement a dataset loader file in ./datasets/
• modify get_data_loader() function in utils.py
• train the model

PS: height and width must be a multiple of 16, see config.py to find more details.

Test

Generate images with trained model.

If 'mnist' dataset:

python main.py --mode test --generator_restored ./results/checkpoints/Generator-final.pt

If 'celeba' dataset or your own dataset:

python main.py --mode test --c_dim [c] --height [h] --width [w] --generator_restored [model path]

Generated Result

mnist

celeba