Landmark detection in medical images is important for many clinical applications. Learning-based landmark detection is successful at solving some problems but it usually requires a large number of annotated datasets for the training stage. In addition, traditional methodsusually fail for the landmark detection of fine objects. In this paper, we tackle the issue of automatic landmark annotation in 3D volumetricimages from a single example based on a one-shot learning method. It involves the iterative training of a shallow convolutional neural network combined with a 3D registration algorithm in order to perform automatic organ localization and landmark matching. We investigated both qualitatively and quantitatively the performance of the proposed approach on clinical temporal bone CT volumes. The results show that our oneshot learning scheme converges well and leads to a good accuracy of the landmark positions.