We consider the problem of automatic camera selection in the context of in situ visualization. This problem is important because high-performance computing trends are increasingly mandating in situ processing, and this processing paradigm frequently has no human-in-the-loop — new research is needed to automate the decisions that have previously been made by human beings. We begin by briefly evaluating what makes an image good, i.e., informative, pleasing, etc. The majority of this work is in surveying existing techniques for camera selection that have been considered in a non-in situ setting, organizing them around geometric- and data-driven techniques. We then survey considering the in situ context that an automatic camera selection algorithm should run, specifically in the infrastructures that house such algorithms and in the driving use cases from application codes. Finally, we conclude the survey by considering data sets from representative simulation codes and evaluating the efficacy of various existing camera selection techniques.