Technical Note: DeepLabCut-Display: open-source desktop application for visualizing and analyzing two-dimensional gait parameter data in livestock

Abstract

AbstractGait assessments are a key part of determining the wellbeing of livestock. Techniques for gait assessment have traditionally involved human-eye inspections or reflective markers, but markerless computer vision methods have been developed in recent years. Despite many computer vision tools providing high-quality pose estimations in an efficient manner, they lack post-processing functionality. A review of model performance and calculation of gait parameters is a necessary step to fully harness the capability of this new technology. Thus, this study developed DeepLabCut-Display, an open-source desktop software application. DeepLabCut-Display allows a user to upload the video and coordinate data associated with the output of DeepLabCut, a prominent pose-estimation software tool. A user can review the video and coordinate data in parallel, filter points by a likelihood threshold, and automatically calculate gait parameters. Specific video frames, filtered data, and gait parameters can be exported from the application for further usage. The source code is publicly hosted on GitHub alongside installation and usage instructions. DeepLabCut-Display, the product of interdisciplinary and collaborative design between software developers and animal scientists, will alleviate a critical bottleneck in processing of data for locomotor analysis in livestock. Summary StatementO_LIDeepLabCut-Display is a utility to dynamically visualize raw marker coordinates, and to automatically produce gait parameters for locomotion analysis of horses and other livestock. C_LI Lay SummaryArtificial intelligence systems that can predict and track the positions of objects are now being applied in many fields, including animal science. Veterinarians and animal scientists use these systems to create pose estimations, a digital label of anatomical landmarks overlaid on a video of an animal in motion. They are used to quantify the subjects motion and detect anomalies that may be indicative of disease or injury. Pose estimation systems are efficient and accurate, but they lack features like data visualization and post-processing analysis that are necessary to make determinations about the animals motion. This study developed DeepLabCut-Display, a software application that can visualize the data from a pose estimation system and provides a set of tools for further analysis. After a user is done with analysis, they can save the results to their computer. The application was made by a collaboration between software developers and animal scientists, highlighting how interdisciplinary teams are effective at producing useful software.