Anime-Face-Generation

In this various models will be used to generate Anime Faces starting with Genrative Models.
As of now DCGAN is done, will upload W-GAN soon.

Dataset

The Dataset for this was taken from Anime Face Dataset

DCGANs

The notebook was run on Kaggle
Exact Architecure has been used in the model as mentioned in the paper [1.] , adding an layer to discriminator did improve the quality (visually) but any other hyperparameter tuning worsened the quality.

Architecure

Results

Interpolation between two Random Instances

Loss Curve during Training

WGANs with generator same as DCGAN (In progess: A run with more Epochs is required)

This was also run on Kaggle
Exact Architecure has been used in the model as mentioned in the paper [3.] [1.] , with changed Loss function and removal of sigmoid in Discriminator(Critic)

Loss Function

Results

Interpolation between two Random Instances

Loss Curve during Training

References

1. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
2. DCGAN for classroom images
3. Wasserstein GAN