jusunglee9931 / pytorch-1dgan-gan-dcgan

GAN 1dgan-gan-dcgan

1d GAN

Result

Epoch timelapse

Z Distritubtion

= Uniform(range = 8)

Data Distribution

= N(-3,1)

Static result

Z Distritubtion	Z Distritubtion	Z Distritubtion
= Uniform(range = 8)	= Uniform(range = 8)	= Uniform(range = 8)
Data Distribution	Data Distribution	Data Distribution
= N(0,1)	= N(0,2)	= N(0,0.5)

Enviroment

1. epoch : 1000, batch size : 8, learning rate : 0.01

Reference

1. https://github.com/hwalsuklee/tensorflow-GAN-1d-gaussian-ex
2. https://github.com/togheppi/vanilla_GAN
3. https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/01-basics/feedforward_neural_network

Vanilla Gan

Mnist result

Epoch timelapse

Generated image	Loss Graph

Enviroment

1. epoch : 100, batch size : 25, learning rate : 0.0002 with dropout

extra result

Generated image	Loss Graph

Enviroment

1. epoch : 100, batch size : 25, learning rate : 0.0002 with batch normalization

Reference

1. https://github.com/znxlwm/pytorch-MNIST-CelebA-GAN-DCGAN
2. https://github.com/togheppi/DCGAN
3. https://github.com/wiseodd/generative-models/tree/master/GAN/vanilla_gan

DCGAN

Mnist Result

Epoch timelapse

Generated image	Loss Graph

Enviroment

1. epoch : 8, batch size : 25, learning rate : 0.0002 ,activation fuction : ReLU For both (generator, discriminator) net , output activation fuction : Sigmoid

Reference

1. https://github.com/znxlwm/pytorch-MNIST-CelebA-GAN-DCGAN

celeba Result

Epoch timelapse

Generated image	Loss Graph

Generated image(cropped)	Loss Graph

Epoch 20	Epoch 40	Epoch 60	Epoch 80	Epoch 100

Enviroment

1. epoch : 60, batch size : 25, learning rate : 0.0002 ,activation fuction : ReLU For both (generator, discriminator) net , output activation fuction : Sigmoid