Optimal viewpoint prediction is an essential task in many computer graphics applications. Unfortunately, common viewpointqualities suffer from two major drawbacks: dependency on clean surface meshes, which are not always available, and the lack ofclosed-form expressions, which requires a costly search involving rendering. To overcome these limitations we propose to sepa-rate viewpoint selection from rendering through an end-to-end learning approach, whereby we reduce the in¿uence of the meshquality by predicting viewpoints from unstructured point clouds instead of polygonal meshes. While this makes our approachinsensitive to the mesh discretization during evaluation, it only becomes possible when resolving label ambiguities that arise inthis context. Therefore, we additionally propose to incorporate the label generation into the training procedure, making the labeldecision adaptive to the current network predictions. We show how our proposed approach allows for learning viewpoint pre-dictions for models from different object categories and for different viewpoint qualities. Additionally, we show that predictiontimes are reduced from several minutes to a fraction of a second, as compared to state-of-the-art (SOTA) viewpoint quality eval-uation. Code and training data is available at https://github.com/schellmi42/viewpoint_learning, whichis to our knowledge the biggest viewpoint quality dataset available.