Conditional Wasserstein Generative Adversarial Networks for Image Dehazing

Aug 1, 2018 · 1 min read
On the left is an image affected by fog, on the right is an image cleared by our algorithm.

Project Supervisor: Professor Sudipta Mukhopadhyay

Achieved state-of-the-art results by training a conditional Wasserstein GAN using the pix2pix model for single image dehazing, with perceptual loss, MSE loss, L1 loss, and texture loss, on the D-Hazy and O-Haze fog datasets, using Pytorch as the programming library.

Joshua Peter Ebenezer
Authors
Staff Research Engineer at Samsung Research
My research interests include computational photography, image and video quality assessment, and deep learning.