Abstract: We present a new perceptual rendering pipeline which takes into account visual masking due to contrast and spatial frequency. Our framework predicts inter-object, scene-level masking caused by partial occlusion and shadows. It is designed for interactive applications and runs efficiently on the GPU. This is achieved using a layer-based approach together with an efficient GPU-based computation of threshold maps. We build upon this prediction framework to introduce a perceptually-based level of detail control algorithm. We conducted a perceptual user study which indicates that our perceptual pipeline generates results which are consistent with what the user perceives. Our results demonstrate significant quality improvement for scenes with masking due to frequencies and contrast, such as masking due to trees or foliage, or due to high-frequency shadows.