Supervised Shadow Removal

Supervised shadow removal aims to computationally restore images obscured by shadows, a crucial task in various fields like remote sensing and computer vision. Recent research emphasizes developing self-supervised and weakly-supervised methods to overcome the limitations of fully supervised approaches, which require large, meticulously paired datasets. These methods often leverage generative adversarial networks (GANs), diffusion models, and techniques like image inpainting, sometimes incorporating multi-modal data (e.g., visible and infrared imagery) to improve accuracy and reduce reliance on ground truth labels. Advances in this area improve image quality in diverse applications, enhancing the reliability of image analysis across various domains.