CapeStone: Genrate Faces with Generative Adversarial Network

The goal of this project is to generate fake faces and make it look like real ones that never existed before by feeding the Network with some Noise. So I want to build Genreative advirtisal network and try to find if it possible ? That it can generate Fake FACES THAT LOOK LIKE REAL from Noise

THE DATASET

• Human Faces https://www.kaggle.com/c/deepfake-detection-challenge/discussion/122786#700963
• Anime Faces https://www.kaggle.com/soumikrakshit/anime-faces
• Simpson Faces https://www.kaggle.com/kostastokis/simpsons-faces

Bulding GAN

• replace pooling layer with Conv2DTranspose
• use BatchNormalization in both g and disc expect for gen outpot layer and dis input layer
• genrator activtion should be LeakyReLU or ReLU expect output layer we use tanh
• all layers in disc shplud use LeakyReLU or ReLU

check this paper

Results

<iframe width="457" height="210" style="position:absolute;top:0;left:0;width:100%;height:100%;" frameBorder="0" src="https://imgflip.com/embed/3ww9q0"></iframe>

via Imgflip