ICCV2019最佳论文Marr奖
作者:Tamar Rott Shaham,Technion(以色列理工)


Intro

一张自然图像训练,测试时可以生成类似该图像的合成图像。

Method

利用金字塔式的多个GAN,每层GAN均引入噪声,对应不同的分辨率。如下图
[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image
生成器的具体细节:
[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image

Results

[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image

Application

Super-Resolution

[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image

[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image

Paint-to-Image

[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image

Harmonization

[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image

Editing

[ICCV2019] SinGAN: Learning a Generative Model from a Single Natural Image

Single Image Animation

https://youtu.be/xk8bWLZk4DU

相关文章: