LOSER - LOw-Light Structural Enhancement and Restoration

Images taken in low-light environments are subject to high loss of information due to the presence of increased noise, lack of color contrast, and potential for extreme color distortion, ultimately resulting in an unclear, low-visibility image. These low-light images pose a challenge to a variety of computer vision algorithms (autonomous driving, facial recognition, object detection, etc.) and their applications under low exposure or nighttime conditions, interfering with the accuracy and generalizability of their results. To mitigate impact on the integrity of such models, we target images of building interiors, attempt to reconstruct lost structural information, and generate normal lighting outputs from low-light inputs.