Generating Anime Faces using Vanilla Variational Autoencoder.
Dataset Link : https://www.kaggle.com/splcher/animefacedataset
NOTE : The generated Images can be improved on further training but it may not be possible to remove the blur/noise factor from images generated by vanilla VAE,also all the training images resized to a lower resolution (60x60x3) which adds up the factor, using GAN networks or another VAE architecture can solve this issue. Using "tanh" as activation in the final layer of encoder instead of "relu" also increases the reconstruction accuracy.
Images generated by the model after training for 400 epochs :
