GAN网络的发展及应用

Thanks for participation of Dr Chang Xu and assistant professor Ran He from The University of Sydney and National Laboratory of Pattern Recognition separately.

Unconditional Generative Models

Generative Adversarial Network [Goodfellow et al, 2014]
Variational Autoencoder [Kingma et al, 2014]

Autoregressive Models (PixelRNN, PixelCNN) [Oord et al, 2016]
Reversible Flow [Dinh et al, 2014] 

Unconditional Dual Generative Models

CoGAN [Liu et al, 2016] MERL
DVG [Fu et al, 2019] NLPR

Conditional Generative Models

Class-conditional

 

Conditional Image-to-image Translation

 

Conditional Image/Video-to-video Generation

Applications

`face generation and translation   

             inpainting/expression/rotation/aging/attribute

`style transfer

`super resolution

`text to image

`audio/video generation

`pose-based human image generation

A Siarohin, E Sangineto, Deformable GANs for Pose-based Human Image Generation CVPR2018.

`Human Motion Transfer

C Chan, S Ginosar, T Zhou, AA Efros, Everybody Dance Now, Siggraph 2018.

`Adversarial Domain Adaptation

Unsupervised domain adaptation by backpropagationICML 2015

Multiple source domain adaptation. [NeurIPS 2018]

Structured domain adaptation (e.g. segmentation). [CVPR 2018]

Conditional domain discriminator. [NeurIPS 2018]

Domain classifier →Task-specific classifier. [CVPR 2018]

Feature augmentation via another GANs. [CVPR 2018]

`Adversarial Examples

Goodfellow et al., 2015; Carlini and Wagner, 2017; Liu et al., 2017

`Visualization of GANs