Sample characterization often represents samples as points distributed in some multidimensional feature space. The visualization of such data distribution, especially when most samples do not have any annotation, is important to assess the quality of the image descriptor, to identify the representative samples from each class and the most informative samples for active learning, and to understand the behavior of the apprentice classifier during active learning. We have explored visual analytics techniques for data projection, such that the specialist can understand and intervene in the active learning process, going beyond the simple label supervision. We aim at either improving or developing active learning techniques.