DCGAN-pytorch

Implementation of Deep Convolutional Generative Adversarial Networks
Based on paper: Unsupervised representation learning with deep convolutional generative adversarial networks

DCGAN Model Architecture

https://arxiv.org/pdf/1511.06434.pdf

Install & Usage

Clone repo

Install requirements

Argparse

Default Args

$ python main.py -lr=0.0002 -dataset=MNIST -batch_size=128 -hidden_dim=128 -z_dim=100 -beta_1=0.5 -beta_2=0.999 -epochs=30 -optimizer=Adam -visual_batch_step=60 -denormalize_img=True

Prerequisites

_{^{certifi==2020.12.5
cffi==1.14.4
cycler==0.10.0
kiwisolver==1.3.0
matplotlib==3.3.2
mkl-fft==1.2.0
mkl-random==1.1.0
mkl-service==2.3.0
numpy==1.19.2
olefile==0.46
pandas==1.1.5
Pillow==8.0.1
pip==20.3.3
pycparser==2.20
pyparsing==2.4.7
python-dateutil==2.8.1
pytz==2020.4
setuptools==51.0.0.post20201207
sip==4.19.13
six==1.15.0
torch==1.6.0
torchvision==0.7.0
tornado==6.1
tqdm==4.54.1
wheel==0.36.2
wincertstore==0.2}}

Development Environments

Python 3.7.6
pytorch 1.6.0
NVIDIA RTX 2060
Ubuntu 20.04.1 LTS

Results

Generated image by same noise during training

MNIST	CelebA

Generated image after 30 epoch training

MNIST	CelebA

Interpolation

MNIST
CelebA

Reference

[1] Alec Radford & Luke Metz, Soumith Chintala.(2016). 'Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks' arXiv:1511.06434v2
[2] Ian J. Goodfellow et al. (2014).'Generative Adversarial Networks' arXiv:1406.2661v1
[3] Martin Arjovsky et al. (2017). 'Wasserstein GAN' arXiv:1701.07875