GAN-jittor

Brief Introduction

This is the code completed by team from Nanjing Uniersity(NJU) for Jittor AI Competition warm-up. The Conditional GAN (Conditional generative adversarial Nets) model is mainly trained on the digital image dataset MNIST by inputting a random vector Z and additional auxiliary information Y (such as category label),and generates an image of a specific phone number.

Environments

• Python >= 3.7
• jittor == 1.3.4.12
• imageio == 2.9.0
• imageio-ffmpeg == 0.4.3
• matplotlib == 3.3.0
• configargparse == 1.3
• tensorboard == 1.14.0
• tqdm == 4.46.0
• opencv-python == 4.2.0.34

Anaconda is recommended to create a conda environment by running

conda create -n gan-jittor python=3.7
conda activate gan-jittor
pip install -r requirements.txt

Training

• By running

  python CGAN.py

  to train the model, and the following results will be obtained.

Thanks

This project refers to the following materials and projects, thank the author for sharing!

• https://cg.cs.tsinghua.edu.cn/jittor/tutorial/2020-5-13-22-47-cgan/
• https://github.com/Jittor/gan-jittor/blob/master/models/cgan/cgan.py
• https://github.com/arturml/mnist-cgan/blob/master/mnist-cgan.ipynb