Accurate interpretations of subsurface salts are vital to oil and gas exploration. However, manually interpreting them from seismic depth images is labor-intensive. Consequently, the use of deep-learning tools, such as a convolutional neural network, for automatic salt interpretation recently became popular. Because of poor generalization capabilities, interpreting salt boundaries using these tools is difficult when labeled data are available from one geologic region, and we like to make predictions for other nearby regions with varied geologic features. At the same time, due to vast amount of the data involved and the associated computational complexities needed for training, such generalization is necessary for solving practical salt interpretation problems. We have adopted a semisupervised training method, which allows the predicted model to iteratively improve as more and more information is distilled from the unlabeled data into the model. In addition, by performing a mixup between labeled and unlabeled data during training, we encouraged the predicted models to linearly behave across training samples, thereby improving the generalization capability of the method. For each iteration, we used the model obtained from the previous iteration to generate pseudolabels for the unlabeled data. This automated consecutive data distillation allowed our model prediction to improve with iteration, without any need for human intervention. To demonstrate the effectiveness and efficiency, we applied the method on 2D images extracted from a real 3D seismic data volume. By comparing our predictions and fully supervised baseline predictions with those that were manually interpreted and that we consider as “ground truth,” we found that the prediction quality our new method surpassed the baseline prediction. Therefore, we concluded that our new method is a viable tool for automated salt delineation from seismic depth images.