African-Robotics-Unit/acinoset_viewer

A repository for a 3D ground truth data creation tool for the AcinoNet cheetah motion tracking paper

AcinoSet Viewer

Author: Daniel Joska

This is a repository for a ground truth data viewer/modifier for the AcinoSet cheetah dataset. The main repository for AcinoSet may be found here.

Installation

The required PyPi packages are listed in requirements.txt. You may pip install all dependencies by running:

pip install -r requirements.txt

Alternatively, you may build the supplied conda environment for an easy installation. This is done by entering the conda_env directory and running:

conda env create -f acinoset.yml

Usage

Enter the src directory by running:

cd .. / src

And simply run python3 gui.py to launch the GUI.

NOTE: to proceed, ensure that all 6 CAM videos are placed in the same folder. It is highly recommended to retain the original folder structure from the official AcinoSet dataset.

3D GT Data Creation

To begin, while in the Setup tab:

1. "Chose Video Folder:" load the folder that has DLC.h5 files, videos, and the fte.pickle file

2. "Chose SBA File:" select the associated n_cam_scene_sba.json file. If your data is structured as it is in the official dataset, this should be done automatically.

Navigate to the Create tab and click "Load Data" to create/check 3D GT data.

Check Video Synchronisation

If you are only using the tool to check video synchronisation, while in the Setup tab:

1. "Chose Video Folder:" select the folder that contains all 6 CAM videos in .mp4 format

Navigate to the Analyze tab and click "Load Data" to check synchronisation.