Invert and perturb GAN images for test-time ensembling

GAN Ensembling

Ensembling with Deep Generative Views.
Lucy Chai, Jun-Yan Zhu, Eli Shechtman, Phillip Isola, Richard Zhang
CVPR 2021

teaser-1

Prerequisites

  • Linux
  • Python 3
  • NVIDIA GPU + CUDA CuDNN

Table of Contents:

  1. Colab – run a limited demo version without local installation
  2. Setup – download required resources
  3. Quickstart – short demonstration code snippet
  4. Notebooks – jupyter notebooks for visualization
  5. Pipeline – details on full pipeline

teaser
We project an input image into the latent space of a pre-trained GAN and perturb it slightly to obtain modifications of the input image. These alternative views from the GAN are ensembled at test-time, together with the original image, in a downstream classification task.

results-2
To synthesize deep generative views, we first align (Aligned Input)

 

 

 

To finish reading, please visit source site