Shadow Free

Shadow removal research aims to computationally eliminate shadows from images, improving the accuracy and reliability of computer vision systems. Current efforts focus on developing sophisticated deep learning models, including diffusion models and transformer-based architectures, often leveraging both global and local image information to achieve high-quality shadow-free reconstructions, sometimes incorporating auxiliary tasks like shadow mask prediction. These advancements are significant because effective shadow removal enhances the performance of various applications, such as object detection, remote sensing, and 3D modeling, by mitigating the adverse effects of shadows on image analysis. Recent work also explores weakly supervised and self-supervised learning approaches to address the challenge of obtaining paired shadow/shadow-free image data for training.