Video-based analysis of animal behaviour

Overview

This will be an introductory course on analysing animal behaviour from video data. The course will cover:

• Overview of animal tracking methods and terminology.

• Pose estimation and tracking with SLEAP. This includes hands-on practice with training a pose estimation model, evaluating its performance, and using it to predict pose tracks in new videos.

• Analysing pose tracks with movement. This includes loading the predicted pose tracks in Python, filtering and smoothing them, computing kinematic variables, and visualising the results.

• Extracting behavioural syllables with keypoint-moseq.

Note

Training and prediction with pose estimation models are GPU-intensive tasks. Since many students will not have access to a GPU on their own machine, we highly recommend that they also attend the follow-up course on Running pose estimation on the SWC HPC system. This will cover how to run pose estimation at scale, using the GPUs of the SWC HPC cluster.

Instructors

• Niko Sirmpilatze

• Chang Huan Lo

• Sofía Miñano

Prerequisites

Hardware Requirements

This is a hands-on course, so please bring your own laptop and charger. A mouse is recommended but not essential. A dedicated GPU is not required but will be helpful.

General Software Requirements

Note

If you are an incoming PhD student attending the full General Software Skills for Systems Neuroscience course and have already installed the general software requirements on Day 1, you may skip this section.

• An IDE for Python programming. We recommend one of the following:

  • Visual Studio Code with the Python extension

  • PyCharm

  • JupyterLab

• A working conda (or mamba) installation. If you don’t have it, install via Miniforge.

• A working Git installation.

Specific Software Requirements

Note

Only proceed with this section after fulfilling the general software requirements above.

You will need to pre-install two different conda environments for the practical exercises. Create them as follows:

1. SLEAP: Use the conda package method from the SLEAP installation guide. You may use either conda or mamba in the installation command. An NVIDIA GPU is not required for this course as you will only use the SLEAP GUI (launched using sleap-label).

2. Keypoint-MoSeq: Use the recommended conda installation method.

You should now have two new conda environments called sleap and keypoint_moseq. To view all your conda environments, run conda env list.

Sample Data

Download the sample data for this course from https://tinyurl.com/behav-analysis-course-data. Click “Download” to get the behav-analysis-course.zip archive, then unzip it.

Alternatively, if you are connected to the SWC network and have access to the SWC’s ceph filesystem, the dataset is available at /ceph/scratch/neuroinformatics-dropoff/behav-analysis-course. Ensure you copy the data to a convenient location on your laptop. The instructions to mount ceph on your laptop can be found on the SWC wiki.

Note

If you encounter any issues with these steps, please contact Niko Sirmpilatze in advance of the course.

Materials

• GitHub repository

• Slides

• Sample data

Useful links:

• SLEAP

• DeepLabCut

• movement

• keypoint-moseq

Recommended readings:

• Neuroscience Needs Behavior: Correcting a Reductionist Bias

• Quantifying behavior to understand the brain

• Open-source tools for behavioral video analysis: Setup, methods, and best practices