nazikus / player-team-clusterization

Workshop: Person Classification in Sports

Workshop: Person Classification in Sports

This workshop aims to show:

• efficient techniques for image representation to keep maximum information about color distribution,
• which clusterizations methods are optimal for person group identification, why some methods work better than the others on images of football players,
• exotic metrics which allow to achieve better clusterization accuracy,
• types of Neural Networks with specific loss functions which are the most suitable for person image representation,
• how incremental learning can help to solve clusterization task on long video streams.

Pre-requirements

• python 3.6
• pip
• conda (Windows users)

Requirements

Install requirements:

pip install -r requirements.txt

Download data:

groups_to_cluster_from_tracker.tar.gz

team_color_dataset_splitted.tar.gz

cd player-team-clusterization
tar -xvzf team_color_dataset_splitted.tar.gz
tar -xvzf groups_to_cluster_from_tracker.tar.gz

---

Diving into details (for beginners):

Linux

• Install python3.6:

mkdir /tmp/Python36
cd /tmp/Python36

sudo wget https://www.python.org/ftp/python/3.6.6/Python-3.6.6.tgz
sudo tar xzf Python-3.6.6.tgz
cd /tmp/Python36/Python-3.6.6/
sudo ./configure
sudo make altinstall

• Return to folder with workshop sources:

python3.6 -m venv ./workshop
source workshop/bin/activate

• Add python3.6 kernel to Jupyter

pip install jupyter
python3.6 -m pip install ipykernel
python3.6 -m ipykernel install --user

Windows

• Install Anaconda

---

Contributors: Raid Arfua, Bogdan Zhurakovskyi

Speakers:

Raid Arfua (github: arfua, skype: raid_arfua)

Bogdan Zhurakovskyi (github: dzhurak, skype: zhurak)