CelebA_GANS

Code for course Deep Learning for Image Analysis

Daniela Stern- Gabsi

github- dgabsi/CelebA_GANS(705_GANS)

(updates were made from danielaneuralx which is my working github but its all mine)

This project explores a few Generative Adversarial Networks(GAN) techniques . This project is based on the celebA dataset, which is a large scale dataset of human faces. I will explore DCGAN which was the first convolutional GAN, the Wassertstein GAN with gradient penalty, and the Self Attention GAN (SAGAN) which is based on the attention mechanism. In addition, I will implement the SRGAN for improve image resolution. I will measure the performance using the Frechet Inception distance (FID)

I have also used TensorBoard for following the training.

##Very Important: Due to the size of data, I have put the data files in OneDrive. In One Drive celeba directory please take the file img_align_celeba.zip and put under the directory : celeba_data-> celeba and unzip it. this will create directory img_align_celeba under celeba_data-> celeba https://cityuni-my.sharepoint.com/personal/daniela_stern-gabsi_city_ac_uk/_layouts/15/onedrive.aspx?id=%2Fpersonal%2Fdaniela%5Fstern%2Dgabsi%5Fcity%5Fac%5Fuk%2FDocuments%2F705data%5Fceleba%2Fceleba

Main notebook: GAN_main_notebook.ipynb

#Please make sure you have the following structure Project structure:

• celeba_data (Directory for data)
  • celeba (In this directory you have to put the zip file from MyDrive and unzip it)
    • img_align_celeba (This directory includes all images. It will be created after you unzip img_align_celeba.zip
    • list_eval_partition.txt
• experiments (Directory for Tensorboard logs)
• saved(Directory for saved models, pickles and visualization results charts)
• images (directory for showing images in jupyter)
• dcgan.py
• sagan.py
• main_gan.py
• training_gan.py
• utils.py
• inception_score.py
• celebA_dataset.py (Dataset)
• GAN_main_notebook.ipynb (Main notebook that should be used)

packages needed :

• torch 1.8.1
• torchvision 0.9.1
• datetime
• time
• matplotlib 3.3.4
• numpy 1.20.1
• pandas 1.2.3
• scikit-learn 0.24.1
• tensorboard 2.4.1
• pickle
• pillow 8.2.0
• scipy