Stacked images with a dynamic range cutoff used to investigate the structural deviations in the Top Set (from SMILI as a representative). First, each Top Set image is normalized with its peak intensity, and the pixels with a flux larger than a certain threshold (e.g., 0.1) are converted to unity (unless 0). Then all the images across the Top Set are stacked and normalized by the maximum number of Top Set images so that the fraction of reliable features over the entire Top Set is emphasized. The ground-truth structure of each synthetic data model is shown in gray contours.