/ReachProcess

ETL pipeline for raw experimental data. Uses deep learning and statistical analysis to create high-dimensional time-series data from raw experimental data.

Primary LanguagePythonMIT LicenseMIT

A library to automate the splitting, prediction, and compilation of rat reaching kinematics from video, microcontroller, and other experimental data sources.

Features

  • utilizes state-of-the-art technique in markerless pose estimation to extract predicted keypoints from video data
  • leverages predicted keypoint positions and the Direct Linear Transform of the camera state-space to estimate 3-D positions
  • calculates kinematic variables for each keypoint, such as speed
  • renders predicted 3-D variables, kinematics along-side video data in video format
  • saves all data within the Neurodata Without Borders ecosystem (see https://www.nwb.org/)

Requires

  • to extract keypoints, a GPU is recommended to speed the process up
  • scripts are provided to run ReachProcess in HPC-performance mode, see the Scripts folder

Example Output from ReachProcess

alt text