/DCGAN-Pytorch

The implementation of Deep Convolutional Generative Adversarial Networks using Pytorch

Primary LanguagePython

DCAGN-Pytorch

This is the implementation of DCGAN (Deep Convolutional Generative Adversarial Networks) using pytorch and we generate the mnist numbers and Anime Avatar in this project.

DCGAN

Development Environment

  • pytorch 1.4
  • Python 3.7
  • torchvision 1.1
  • CUDA 10.0
  • tqdm
  • NVIDIA 2080ti
  • Ubuntu 18.04
  • OpenCV

Content

|-- data
|   |-- MNIST                 # Mnist dataset
|       |-- processed
|       |   |-- test.pt
|       |   |-- training.pt
|       |-- raw
|-- face_train.py             # train to generate the Anime Avatar
|-- generate_gif.py           # generate gif images
|-- mnist_train.py            # train to generate the mnist numbers
|-- model                   
|   |-- face_dcgan.py         # DCGAN (3 channels)
|   |-- mnist_dcgan.py        # DCGAN (1 channel)
|-- preprocess.py             # Crop images
|-- result                    # result gif
|   |-- comic.gif
|   |-- mnist.gif

Quick Start

  • Train the Anime Avatar (you can modify the epochs and other args) 
    # Berfore the train, create the g_loss.txt and d_loss.txt
$ cd DCGAN-pytorh-master
$ ./face_train.py
  • Train the mnist numbers (you can modify the epochs and other args) 
    # Berfore the train, create the g_loss.txt and d_loss.txt
$ cd DCGAN-pytorh-master
$ ./mnist_train.py

Results

Anime Avatar

  • Generate the Anime Avatar using fixed random noise (totally 50 epochs)

    comic

Mnist numbers

  • Generate the mnist numbers using fixed random noise (totally 50 epochs)

    comic