Low-rank subspaces have long been a powerful tool in data modeling and analysis. In particular, they have proven very useful in computer vision: Subspace models are of great interest in computer vision for background subtraction [52], object tracking [5, 19], and to represent a single scene under varying illuminations [9, 20]. Other applications in communications [40], source localization and target tracking in radar and sonar [33], and medical imaging [3] all leverage subspace models in order to recover the signal of interest and reject noise. In these classical signal processing problems, a handful of high-quality sensors are co-located such that data can be reliably collected.