Cycle-Dehaze: Enhanced CycleGAN for Single Image Dehazing

In this paper, we present an end-to-end network, called Cycle-Dehaze, for single image dehazing problem, which does not require pairs of hazy and corresponding ground truth images for training. That is, we train the network by feeding clean and hazy images in an unpaired manner. Moreover, the proposed approach does not rely on estimation of the atmospheric scattering model parameters. Our method enhances CycleGAN formulation by combining cycle-consistency and perceptual losses in order to improve the quality of textural information recovery and generate visually better haze-free images. Typically, deep learning models for dehazing take low resolution images as input and produce low resolution outputs. However, in the NTIRE 2018 challenge on single image dehazing, high resolution images were provided. Therefore, we apply bicubic downscaling. After obtaining low-resolution outputs from the network, we utilize the Laplacian pyramid to upscale the output images to the original resolution. We conduct experiments on NYU-Depth, I-HAZE, and O-HAZE datasets. Extensive experiments demonstrate that the proposed approach improves CycleGAN method both quantitatively and qualitatively.

1. Introduction

Bad weather events such as fog, mist, and haze dramatically reduce the visibility of any scenery and constitute significant obstacles for computer vision applications, e.g. object detection, tracking, and segmentation. While images captured from hazy fields usually preserve most of their major context, they require some visibility enhancement as a pre-processing before feeding them into computer vision algorithms, which are mainly trained on the images captured at clear weather conditions. This pre-processing is generally called as image dehazing/defogging. Image de-hazing techniques aim to generate haze-free images purified from the bad weather events. Sample hazy and haze-free images from the NTIRE 2018 challenge on single image dehazing [4] are illustrated in Figure 1. Figure 1:

Hazy and clean examples from the NTIRE 2018 challenge on single image dehazing datasets: I-HAZE [6] & O-HAZE [7] datasets.