Interactive rodent behavior annotation in video using active learning

Manual annotation of rodent behaviors in video is time-consuming. By learning a classifier,we can automate the labeling process. Still, this strategy requires a sufficient number oflabeled examples. Moreover, we need to train new classifiers when there is a change in theset of behaviors that we consider or in the manifestation of these behaviors in video. Con-sequently, there is a need for an efficient way to annotate rodent behaviors. In this paper weintroduce a framework for interactive behavior annotation in video based on active learn-ing. By putting a human in the loop, we alternate between learning and labeling. We applythe framework to three rodent behavior datasets and show that we can train accurate behav-ior classifiers with a strongly reduced number of labeled samples. We confirm the efficacyof the tool in a user study demonstrating that interactive annotation facilitates efficient,high-quality behavior measurements in practice.