Deep Convolution Generative Adversarial Networks

PyTorch-DCGAN has been deprecated. Please see DCGAN-PyTorch, which includes implementations for all models in DCGAN.

Introduction

This example implements the paper Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

The implementation is very close to the Torch implementation main.py

After every 100 training iterations, the files real_samples.png and fake_samples.png are written to disk with the samples from the generative model.

After every epoch, models are saved to: netG_epoch_%d.pth and netD_epoch_%d.pth

Configure

• PyTorch > 1.3.0
• GTX 1080 Ti

Load dataset

• baidu netdisk password：g5qa

download data put on ./data/ folder.

Thanks 何之源

data/
└── faces/
    ├── 0000fdee4208b8b7e12074c920bc6166-0.jpg
    ├── 0001a0fca4e9d2193afea712421693be-0.jpg
    ├── 0001d9ed32d932d298e1ff9cc5b7a2ab-0.jpg
    ├── 0001d9ed32d932d298e1ff9cc5b7a2ab-1.jpg
    ├── 00028d3882ec183e0f55ff29827527d3-0.jpg
    ├── 00028d3882ec183e0f55ff29827527d3-1.jpg
    ├── 000333906d04217408bb0d501f298448-0.jpg
    ├── 0005027ac1dcc32835a37be806f226cb-0.jpg

Purpose

Use a stable DCGAN structure to generate avatar images of anime girls.

Usage

• train

if you want pretrain generate model, click it netg_200.pth

if you want pretrain discriminate model, click it netd_200.pth

please rename model name. netd_200.pth -> D.pth and netg_200.pth -> G.pth

start run:

python main.py --dataroot ./data --cuda

if you n’t have GPU, run

python main.py --dataroot ./data

• test

python main.py --mode test --out_images ./result

Example

• epoch 1

• epoch 30

• epoch 100

• epoch 200