Fine-Grained object recognition is the task of distinguishing between highly similar objects such as cars or bird species. Although it is an important problem in computer vision, current recognition models are unlikely to accurately classify objects in the wild. These fully supervised models need additional annotated images to classify objects in every new scenario, a task that is infeasible. However, sources such as e-commerce websites and field guides provide annotated images for many classes. For example, last year, I spoke about our work using characteristics of cars recognized in Google Street View images to predict neighborhood demographics. In that work, we leveraged many labeled images from e-commerce websites, along with a smaller number of labeled Google street view images, to train a car classifier. In this talk, I'll discuss fine-grained domain adaptation as a step towards overcoming the dataset shift between easily acquired annotated images (such as those found in e-commerce sites) and the real world.