Deep Generative Image Models using a Laplacian Pyramid of Adversarial Network


WHY?

GAN had troble modeling the entire image.

Note

Laplacian Pyramid Framework is used for restoring compressed image. When a image is compressed to smaller size, it would lose information of high-resolution image so that simply enlarging the image would not be enough to restore original data. Laplacian pyramid framework save the differences between enlarged low resolution image and high resolution image at each stage.

WHAT?

Laplacian Generative Adversarial Network (LAPGAN) generate high resolution image by generating residuals with a series of generators. image

\tilde{I}_k = u(\tilde{I}_{k+1}) + \tilde{h}_k = u(\tilde{I}_{k+1}) + G_k(z_k, u(\tilde{I}_{k+1}))

image Generators at each stage take coarse image from previous stage and z as input and try to predict residuals. On the other hand, discriminators try to distinguish generated residuals and real residuals. Generators and discriminators can be conditioned with label vectors.

So?

LAPGAN produced better quality image of CIFAR10, STL and LSUN than original GAN.

Critic

Historical model that is used frequently as base line. Liked the Laplacian pyramid approach.

Denton, Emily L., Soumith Chintala, and Rob Fergus. “Deep generative image models using a laplacian pyramid of adversarial networks.” Advances in neural information processing systems. 2015.



© 2017. by isme2n

Powered by aiden