Stride-level analysis of mouse open field behavior using deep learning-based pose estimation

This release contains data files for our publication: "Stride-level analysis of mouse open field behavior using deep learning-based pose estimation". Also note that our source code is made available at our github page under the repositories deep-hrnet-mouse and gaitanalysis.

Files:

• deep-hres-net-2019-06-28.simg: a singularity image containing everything needed to run our neural network
• deep-hrnet-mouse.tgz: a tar'd, gzip'd version of the repository at https://github.com/KumarLabJax/deep-hrnet-mouse (commit 13014885861f87bf9fd5a99afdfe3153cca852a5)
• gaitanalysis.tgz: a tar'd, gzip'd version of the repository at https://github.com/KumarLabJax/gaitanalysis (commit 18a7f9b0192b77d94dc85ff413f0875b958c4938)
• infer-poseest-batch.sh: our inference script written to run on our cluster using the SLURM scheduler. This script will not work unchanged in other environments but it serves as a useful example of how to run the singularity image (deep-hres-net-2019-06-28.simg) on a cluster
• merged_pose_annos_2019-06-26.h5.gz: training data that was used to train the neural network that is embedded in deep-hres-net-2019-06-28.simg
• pose-est-conf.yaml: the configuration file used for training our pytorch model. This file is responsible for specifying most of the important hyperparameters during training
• pose-est-model.pth: the best model generated from our pytorch training. This file is embedded in our singularity image (deep-hres-net-2019-06-28.simg)
• strain-survey-gait-2019-11-12_summary.csv: a table of summary statistics for our gait metrics which were derived from our strain survey dataset
• vidplot-app.mp4: A video representation of gait extraction from pose estimation. The top panel shows a segment of video with a gait overlay. The bottom panel contains two plots that update with the video: an angular velocity plot and a hind paw speed plot. Green bouts are considered strides, left and right paws are orange and blue respectively.