Compact multidomain single-image dehazing network based on gUNet and test-time training
Liwei Zhang, Haiyi Bian · 2023
Image dehazing is a field that focuses on enhancing the visibility of hazy images. It is important in various applications such as improving image quality, enhancing surveillance capabilities, and aiding in remote sensing and navigation systems under haze weather. Recently, image dehazing has notably advanced by the development of deep learning. Although compact single-domain single image dehazing network (gUNet) and a technique towards multi-domain single image dehazing (test-time training) have been proposed, compact multi-domain single image dehazing network is still not been implemented. We combine gUNet and test-time training to build such a compact network. A qualitative comparison of dehazed images predicted by our network and gUNet demonstrate the remarkable improvement achieved by our proposal in multidomain single-image dehazing. The peak signal-to-noise ratio and structural similarity index confirm the higher performance of our network.