Enclosure Tracking

Scripts to perform tracking and pose estimation in the outside enclosures.

This repository contains a series of scripts designed to facilitate posture tracking using DeepLabCut. Each script corresponds to a specific step in the workflow, from project creation to model evaluation.

Installation

To install the necessary package that contains these scripts, you can use pip to install directly from the GitHub repository:

pip install git+https://github.com/t4d-gmbh/EnclosureTracking.git@main

Once installed you can run any of the scripts explained below directly in the command line.

Scripts Overview

1. init_pretrained

This script creates a new DeepLabCut project using a pretrained model and prepares it for labeling.

Key Steps:

• Create a project with a pretrained model.
• Convert the project into a multi-animal

Usage:

init_pretrained --user <username> --working_dir <path> --project_name <project_name> --model <model_name> --path_to_videos <path_to_videos>

2. add_videos

This script adds new video files to an existing DeepLabCut project.

Key Steps:

• Retrieve the configuration path for the specified project.
• Add the specified video files to the existing project.

Usage:

add_videos --user <username> --working_dir <path> --project_name <project_name> --videos_to_add <video1> <video2> ...

3. finetune_pretrained

This script performs training (a fine-tuning) of the pretrained model.

Key Steps:

• Retrieve the configuration path for the specified project.
• Converts labeled data to the required format.
• Checks the labels for correctness.
• Creates a training dataset by mapping markers.
• Trains the network using the specified parameters.

Usage:

finetune_pretrained --user <username> --working_dir <path> --project_name <project_name> --model <model_name> --batch_size <nbr>

4. evaluate_pretrained

This script evaluates a trained DeepLabCut model.

Key Steps:

• Retrieve the configuration path for the specified project.
• Evaluate the trained model using the specified configuration.

Usage:

evaluate_pretrained --user <username> --working_dir <path> --project_name <project_name>

5. tracking_pretrained

This script performs tracking of individuals in a video.

Key Steps:

• Retrieve the configuration path for the specified project.
• Analyze the specified videos to track individuals.
• Create annotated videos to check the performance of the tracking.

Usage:

tracking_pretrained --user <username> --working_dir <path> --project_name <project_name> --batc_size <nbr> --videos_to_analyze <video1> <video2> ...

Requirements

• DeepLabCut (DLC 3.0)
• Python 3.11

License

This project is licensed under the MIT License - see the LICENSE file for details.