This paper explores object detection in the small data regime, where only a limited number of annotated bounding boxes are available due to data rarity and annotation expense. This is a common challenge today with machine learning being applied to many new tasks where obtaining training data is more challenging, e.g. in medical images with rare diseases that doctors sometimes only see once in their life-time and where professional medical annotation resources are expensive. In this work we explore this problem from a generative modeling perspective by learning to generate new images with associated bounding boxes, and using these for training an object detector. We show that simply training an existing generative model does not yield satisfactory performance due to it optimizing for image realism instead of object detection accuracy. To this end we develop a new model that jointly learns the generative model and a detector such that the generated images improve the performance of the detector. We show that this method significantly outperforms the state of the art, improving the average precision on the NIH Chest X-ray dataset by 50% and pedestrian detection by X%.