Recurrent 3D Convolutional Network for Rodent Behavior Recognition

Animal, specially rodent, studies are critical in understanding human health, disease and treatments. Behavior is an important observed outcome in many such studies. Thus, quantifying rodent behaviors is key. This is typically done by trained human observers, making the process very slow and subjective. This has has led to a growing interest in developing automated assessment tools. Existing approaches commonly rely on hand-crafted features which are often obtained through a tracking process. Motivated by state of the art results in image and video analysis using deep learning, we propose a deep architecture which is a combination of recurrent network and 3D convolutional network to learn long and short-term video representations. We test the proposed solution with the dataset collected by [1] and demonstrate that our framework can obtain accuracy on par with human assessment.