Data-driven fault detection has been regarded as a 3-D image segmentation task. The models trained from synthetic data are difficult to generalize in some surveys. Recently, training 3-D fault ...