CIMAR-LA Crash Course: Introduction to DeepLabCut

Summary crash course

The crash course introduces participants to DeepLabCut (DLC), an open-source tool for markerless pose estimation, and guides them through its complete process, from installation and video processing to model training, performance evaluation, and behavioural data analysis.
The course will cover the fundamentals of artificial intelligence and DeepLabCut itself, software installation and setup, project creation and workflow, model training and performance evaluation, as well as the applications and challenges of DeepLabCut in fish ethology.
The goal is to provide participants with the knowledge and practical skills needed to develop, evaluate, and interpret DeepLabCut tracking projects for fish behavioural research.

Course Instructor

Andrea Meloni is an Italian researcher who earned his MSc in Marine Biology from the University of Padua, conducting his thesis research on fish reproduction at the Hellenic Centre for Marine Research (HCMR) in Heraklion. He went on to obtain his PhD in Veterinary Science, focusing on a holistic approach to study the Fish Welfare. His doctoral research balanced the study of stress physiology with the development of AI-based methodologies to acquire behavioral metadata for ethological investigations. Through this work, he established novel protocols for AI tracking across various fish species and setups, gaining extensive, hands-on expertise in implementing applied DeepLabCut methods and practical data analysis.

General Introduction
a. Brief introduction to artificial intelligence (AI)
b. Introduction to DeepLabCut (DLC) and its applications
c. Applications and potential of DLC in fish ethology
Installing DeepLabCut
a. Minimum hardware and software requirements
b. Where and how to download and install DLC: a step-by-step guide
Understanding DeepLabCut
a. Software structure and general workflow
b. Essential terminology and key concepts
The DeepLabCut Workflow: A Step-by-Step Guide
a. Creating a project
b. Extracting frames
c. Labelling keypoints
d. Training a model
e. Evaluating model performance and analysing predictions
Applying DeepLabCut to Fish Ethology
a. Assessing project feasibility before starting
b. Analysing videos to select the most appropriate approach
c. Identifying critical points and challenges in fish pose and tracking
d. Selecting and extracting frames for fish-related applications
e. Choosing keypoints to label based on the research question and how to label them
f. Training a model: key parameters and how to adjust them
g. Evaluating model performance and analysing the results
Assessing the goodness of the results
a. Obvious error
b. Non-obvious error
Working with DeepLabCut Outputs
a. Extracting and organising tracking data
b. Visualising and validating tracking results
c. Using tracking data for behavioural analysis

This course is aimed at MSc & PhD students and post-doctoral researchers in the fields of ethology, welfare and animal behaviour.

Participants are expected to have at least some experience or contact with behavioural studies.

As this is a CIMAR-LA-accredited course, priority is given to members of CCMAR and CIIMAR – places are limited to 15 participants and registration must be completed via the online form with automatic payment. The fees are set out in the table below.