Imaging through dynamic scattering media is a challenging problem with broad applications. This study proposes a comprehensive learning-based technique for real-time, non-invasive, incoherent imaging of real-world objects through dense and dynamic scattering media, demonstrating its capability through experiments with turbid water and natural fog. The technique surpasses existing approaches in image quality, inference speed, and memory consumption.
