Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies

In social behavior research using rodents, there is a growing demand for evaluating more natural interactive behaviors under freely moving conditions. Accurate pose tracking of multiple animals is essential for this purpose. However, current markerless multi-animal pose tracking tools face a significant challenge: tracking accuracy tends to decline under conditions of occlusion and crowding. This problem becomes especially pronounced when the animals are visually indistinguishable from one another. To overcome this issue, we developed virtual marker tracking (vmTracking), a method that improves the accuracy of multi-animal pose tracking under such challenging conditions by maintaining individual identity across frames using virtual markers. vmTracking can also be applied to existing markerless multi-animal video data by incorporating additional processing steps that add individual identity labels into standard tracking workflows. Here, we describe both the method for assigning virtual markers and the protocol for tracking animals in the resulting labeled videos. High-accuracy multi-animal tracking enabled by vmTracking provides a reliable foundation for subsequent quantitative analyses of social interactions under semi-natural conditions.