Image-Generation-by-Variational-Autoencoder

This project focused on the Image Reconstruction using a Variational Autoencoder (VAE).

Dataset

All images in this project comes from Anime Face Dataset
Download the images from 👉 https://www.kaggle.com/splcher/animefacedataset

Execution & Overall Structure of system

1. Image Preprocessing :

   • Resize Images to [32,32] pixels (Cubic interpolation)
   • Reshape 2D image to 1D array [1024 set of RGB] = [3072]
     

   python3 Img_Preprocess.py
   

2. Training a VAE network with Torch package
   

   python3 VAE.py
   

3. Reconstruct the trained model with random latent variable Z

   python3 Reconstruct_Img.py
   

   • Reconstruction Results
     

   • Visualization fo VAE's Transitivity