UnShadowNet: Illumination Critic Guided Contrastive Learning for Shadow Removal
Subhrajyoti Dasgupta, Arindam Das, Senthil Kumar Yogamani, Sudip Das, Ciarán Eising, Andrei Bursuc, Ujjwal Bhattacharya · IEEE Access · 2023
Shadows are frequently encountered natural phenomena that significantly hinder the performance of computer vision perception systems in practical settings, e.g., autonomous driving. A solution to this would be to eliminate shadow regions from the images before the processing of the perception system. Yet, training such a solution requires pairs of aligned shadowed and non-shadowed images which are difficult to obtain. We introduce a novel weakly supervised shadow removal frameworkUnShadowNettrained using contrastive learning. It is composed of aDeShadowernetwork responsible for the removal of the extracted shadow under the guidance of anIlluminationnetwork which is trained adversarially by the illumination critic and aRefinementnetwork to further remove artefacts. We show thatUnShadowNetcan be easily extended to a fully-supervised set-up to exploit the ground-truth when available.UnShadowNetoutperforms existing state-of-the-art approaches on three publicly available shadow datasets (ISTD, adjusted ISTD, SRD) in both the weakly and fully supervised setups.