Loyola University of Chicago

Deep learning models have shown promise for image-quality rating tasks, yet the visual cues they rely on are often difficult to interpret. Grad-CAM and related saliency methods are commonly used to visualize attribution, but these heatmaps often yield unclear...