Soccer Action Spotting

This repository contains the code for video action spotting for soccer videos.

The repository includes:

• Source code for audio, video, and graph model
• Training/testing code for SoccerNet-V2
• Jupyter notebooks for visualizing results

The code is documented and designed to be easy to extend. If you use it in your research, please consider citing this repository (bibtex below).

Citation

The paper is available as an arXiv report. If you use this code, please cite the following paper:

Alejandro Cartas, Coloma Ballester, and Gloria Haro. "A Graph-Based Method for Soccer Action Spotting Using Unsupervised Player Classification" ACM MMSports Workshop (2022).

@inproceedings{cartas2022GraphActionSpotting,
author = {Cartas, Alejandro and Ballester, Coloma and Haro, Gloria},
title = {A Graph-Based Method for Soccer Action Spotting Using Unsupervised Player Classification},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
booktitle = {Proceedings of the 5th International ACM Workshop on Multimedia Content Analysis in Sports},
series = {MMSports '22}
pages = {93–102},
location = {Lisboa, Portugal},
year = {2022},
}

Requirements

Python 3.7, PyTorch 1.8.1, and other common packages listed in environment.yml. (A Conda environment can be created from it following the instructions below.)

SoccerNet-V2 Requirements:

To train or test on SoccerNet-V2, you'll also need:

• SoccerNet-V2 Dataset
• VGG-11 Audio features extracted by Bastien Vanderplaetse and Stéphane Dupont. NOTE: these can be extracted by following the instructions inside the features directory.

Setup

1. Clone this repository:

git clone https://github.com/IPCV/soccer_action_spotting

2. Create and initialize a conda environment:

cd soccer_action_spotting
conda env create --name soccer --file environment.yml
conda activate soccer

3. Link the SoccerNetV2 dataset inside the datasets directory:

mkdir datasets && cd $_
ln -s /path/to/soccernet soccernet

4. Follow the instructions of the README.md file in the features directory to create the CSV files.

Training/Testing

After activating the conda environment described in the setup section, type the configuration to train:

python main.py config/configuration.json

or for only testing an already trained model:

python main.py --test_only config/configuration.json

For instance, for training the audio baseline:

python main.py config/comparison/audio_stream_pooling.json

Full documentation of the training/testing code main.py:

usage: main.py [-h] [-d DATASET_PATH]
               [--split_train TRAIN_SPLITS [TRAIN_SPLITS ...]]
               [--split_valid VALIDATION_SPLITS [VALIDATION_SPLITS ...]]
               [--split_test TEST_SPLITS [TEST_SPLITS ...]]
               [--max_num_workers MAX_NUM_WORKERS]
               [--evaluation_frequency EVALUATION_FREQUENCY]
               [--weights_dir WEIGHTS_DIR] [--weights WEIGHTS] [--GPU GPU]
               [--log_config LOG_CONFIG] [--test_only]
               [--test_batch_size TEST_BATCH_SIZE] [--tiny TINY] [--no_cache]
               [--overwrite_cache]
               conf

Multi-modal action spotting implementation

positional arguments:
  conf                  JSON model configuration filepath

optional arguments:
  -h, --help            show this help message and exit
  -d DATASET_PATH, --dataset_path DATASET_PATH
                        Path for SoccerNet dataset (default: data/soccernet)
  --split_train TRAIN_SPLITS [TRAIN_SPLITS ...]
                        list of splits for training (default: [train])
  --split_valid VALIDATION_SPLITS [VALIDATION_SPLITS ...]
                        list of splits for validation (default: [valid])
  --split_test TEST_SPLITS [TEST_SPLITS ...]
                        list of split for testing (default: [test])
  --max_num_workers MAX_NUM_WORKERS
                        number of worker to load data (default: 4)
  --evaluation_frequency EVALUATION_FREQUENCY
                        Evaluation frequency in number of epochs (default: 10)
  --weights_dir WEIGHTS_DIR
                        Path for weights saving directory (default: weights)
  --weights WEIGHTS     Weights to load (default: None)
  --GPU GPU             ID of the GPU to use (default: -1)
  --log_config LOG_CONFIG
                        Logging configuration file (default:
                        config/log_config.yml)
  --test_only           Perform testing only (default: False)
  --test_batch_size TEST_BATCH_SIZE
                        Batch size for testing (default: 256)
  --tiny TINY           Consider smaller amount of games (default: None)
  --no_cache            Don't use cached data (default: False)
  --overwrite_cache     Overwrite cached data (default: False)

Data

The annotated data corresponds to the second half of the match england_epl/2014-2015/2015-02-21 - 18-00 Chelsea 1 - 1 Burnley. The data can be downloaded from here.

Acknowledgements

The authors acknowledge support by MICINN/FEDER UE project, ref. PID2021-127643NB-I00, H2020-MSCA-RISE-2017 project, ref. 777826 NoMADS, and ReAViPeRo network, ref. RED2018-102511-T.