PolyU is a large-scale dataset containing real-world noisy images, including corresponding ground truth images. The dataset captures 40 scenes, including indoor normal lighting scenes and dark lighting scenes, and outdoor normal lighting scenes. Each scene is captured 500 times in a row.