Skip to content

Latest commit

 

History

History
110 lines (63 loc) · 4 KB

README.md

File metadata and controls

110 lines (63 loc) · 4 KB
Raw

SoccerNet: A Scalable Dataset for Action Spotting in Soccer Videos

[DEPRECATED] Please visit https://github.com/SilvioGiancola/SoccerNetv2-DevKit for an updated version of that repository

CVPR'18 Workshop on Computer Vision in Sports

Available at openaccess.thecvf.com

@InProceedings{Giancola_2018_CVPR_Workshops,
  author = {Giancola, Silvio and Amine, Mohieddine and Dghaily, Tarek and Ghanem, Bernard},
  title = {SoccerNet: A Scalable Dataset for Action Spotting in Soccer Videos},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  month = {June},
  year = {2018}
}

Project page: https://silviogiancola.github.io/SoccerNet/

Data available:

Clone this repository

git clone https://github.com/SilvioGiancola/SoccerNet-code.git

Create the conda environement (Python3)

conda env create -f src/environment.yml

source activate SoccerNet

Download the data

We recommand to use https://github.com/wkentaro/gdown to download large files from google drive.

pip install gdown (already in the conda environment)

Please use the following script to download automatically the data:

  • Frames Features:

./src/SoccerNet_CSV_Downloader.sh data/SoccerNet_V1.1_Features.csv

  • Labels:

./src/SoccerNet_CSV_Downloader.sh data/SoccerNet_V1.1_Labels.csv

  • Commentaries:

./src/SoccerNet_CSV_Downloader.sh data/SoccerNet_V1.1_Commentaries.csv

  • Videos (224p) (csv file available after filling this form):

./src/SoccerNet_CSV_Downloader.sh data/SoccerNet_V1.1_Videos.csv

  • Videos (HD) (csv file available after filling this form):

./src/SoccerNet_CSV_Downloader.sh data/SoccerNet_V1.1_Videos_HQ.csv

Read data

Read data for a single game

python src/ReadData.py "data/england_epl/2014-2015/2015-05-17 - 18-00 Manchester United 1 - 1 Arsenal"

Read commentaries for a single game

python src/ReadCommentaries.py data france_ligue-1 2016-2017 "Paris SG" "Marseille"

Loop and read over Train/Valid/Test

python src/ReadSplitData.py data src/listgame_Train_300.npy

Loop and read over all games

python src/ReadAllData.py data

Source code for data reproducibility

Features Extraction from videos

See src/feature_extraction for more details.

Action Classification

See src/Classification for more details.

Action Detection/Spotting

See src/Detection for more details.

Getting Started with Colab

It is possible to use Colab to work with SoccerNet on the Google Cloud. Colab provides a colaborative python environment in the cloud including unlimited storage as well as a free Tesla K80 GPU.

To us SoccerNet on Colab, please check this jupyter notebook.

(Acknowlegments: thanks to lamia13Alg for sharing her Colab notebook)