CGAN_jittor

CGAN_jittor is an open-source implementation of CGAN based on jittor, and the repo provides a demo for the application of it, which enables quick hands-on with jittor

Introduction

The repo aims to train a Conditional GAN model that maps random noise and categories to the image of numbers, e.g. the number from 0 to 9. And you can use the model to generate any image of the input number string you like. For example, your telephone number.

The sample code is provided in the directory ./warm_up_competition, including the download of MNIST dataset, the definition of model and the training function.

The output of an example input number string is generated as below:

Installation

You can install CGAN_jittor by Git

Git

# clone the repo
git clone git@code.gitlink.org.cn:hbx20/CGAN_jittor.git
# enter the directory
cd CGAN_jittor
# install the required packages
sudo python3 -m pip install -r requirements.txt
# enter the directory
cd warm_up_competition
# run the demo CGAN model
python3 CGAN.py

Structure

The structure of the CGAN_jittor is listed below:

.
├── assets
├── data
├── models
└── warm_up_comp
    ├── CGAN.py
    └── result.png

assets: contains the images used in the readme of models implemented by jittor
data: contains the scripts for downloading datasets
model: contains all the models implemented by jittor
warm_up_comp: contains the demo code and an example generated image

Usage

Besides from the CGAN model, all other models are provided in the directory “models”, which also contains a README.md for the usage of these models. You can try any of them to check if it works well.