A visualization of results on the HERA dataset. The figure is divided into an upper and a lower part, corresponding to Only-Sup and the proposed method, respectively. In each subfigure, the x-axis denotes frequency channels and the y-axis denotes subintegrations. The first column on the left shows the input time–frequency visibility data and the corresponding ground truth. The four columns on the right show, in order, the segmentation masks predicted by different models (top row) and their difference maps with respect to the ground truth (bottom row). The difference maps are computed as the absolute difference between the predicted mask and the ground truth. Red indicates false positives, and green indicates false negatives. The less area red + green covers, the better the performance is.