A GAN-based input-size flexibility model for single image dehazing

Image-to-image translation based on generative adversarial network (GAN) has achieved state-of-the-art performance in various image restoration applications. Single dehazing is a typical example, which aims to obtain the haze-free of haze one. This paper concentrates challenging task single dehazing. Based atmospheric scattering model, novel model designed directly generate image. The main challenge that two parameters, i.e., transmission map and light. When they are estimated respectively, errors will be accumulated compromise quality. Considering this reason sizes, input-size flexibility conditional (cGAN) proposed for dehazing, at both training test stages image-to-image with cGAN framework. A simple effective U-connection residual (UR-Net) combine generator adopt spatial pyramid pooling (SPP) design discriminator. Moreover, trained multi-loss function, consistency loss paper. Finally, multi-scale fusion built realize performance. models receive as input output Experimental results demonstrate effectiveness efficiency models.