The recent achievements of Deep Learning rely on the test data being similar\nin distribution to the training data. In an ideal case, Deep Learning models\nwould achieve Out-of-Distribution (OoD) Generalization, i.e. reliably make\npredictions on out-of-distribution data. Yet in practice, models usually fail\nto generalize well when facing a shift in distribution. Several methods were\nthereby designed to improve the robustness of the features learned by a model\nthrough Regularization- or Domain-Prediction-based schemes. Segmenting medical\nimages such as MRIs of the hippocampus is essential for the diagnosis and\ntreatment of neuropsychiatric disorders. But these brain images often suffer\nfrom distribution shift due to the patient's age and various pathologies\naffecting the shape of the organ. In this work, we evaluate OoD Generalization\nsolutions for the problem of hippocampus segmentation in MR data using both\nfully- and semi-supervised training. We find that no method performs reliably\nin all experiments. Only the V-REx loss stands out as it remains easy to tune,\nwhile it outperforms a standard U-Net in most cases.\n