This repository implements the DeblurGAN paper for blind motion deblurring using conditional adversarial networks. The model takes blurry images as input and produces corresponding sharp estimates. It uses a Conditional Wasserstein GAN with Gradient Penalty combined with perceptual loss based on VGG-19 activations. The same architecture also applies to other image-to-image translation tasks including super resolution, colorization, inpainting, and dehazing.