U-shape Transformer for Underwater Image Enhancement

The light absorption and scattering of underwater impurities lead to poor imaging quality. existing data-driven based image enhancement (UIE) techniques suffer from the lack a large-scale dataset containing various scenes high-fidelity reference images. Besides, inconsistent attenuation in different color channels space areas is not fully considered for boosted enhancement. In this work, we constructed (LSUI) including 4279 pairs, reported an U-shape Transformer network where transformer model first time introduced UIE task. integrated with channel-wise multi-scale feature fusion (CMSFFT) module spatial-wise global modeling (SGFMT) specially designed task, which reinforce network’s attention more serious attenuation. Meanwhile, order further improve contrast saturation, novel loss function combining RGB, LAB LCH spaces following human vision principle. extensive experiments on available datasets validate state-of-the-art performance technique than 2dB superiority. demo code are https://bianlab.github.io/codes.html .