README.md

---
pretty_name: SoccerHigh
license: cc-by-nc-sa-4.0
task_categories:
- video-classification
- feature-extraction
tags:
- soccer
- sports
- video
- video-summarization
- highlight-detection
- temporal-annotations
homepage: https://ipcv.github.io/SoccerHigh/
repository: https://github.com/IPCV/SoccerHigh
dataset_size: 22GB
---

# ⚽ SoccerHigh

This dataset provides **annotations and pre-extracted features** for the SoccerHigh benchmark introduced in:

**SoccerHigh: A Benchmark Dataset for Automatic Soccer Video Summarization**  
[![arXiv](https://img.shields.io/badge/arXiv-2501.01234-b31b1b.svg)](https://arxiv.org/abs/2509.01439)
[![ACM DL](https://img.shields.io/badge/ACM-DL-blue)](https://dl.acm.org/doi/pdf/10.1145/3728423.3759410)    
[Artur Díaz-Juan](https://scholar.google.com/citations?user=WlPmWzwAAAAJ&hl=ca), [Coloma Ballester](https://scholar.google.com/citations?user=fLNi-SoAAAAJ&hl=ca), [Gloria Haro](https://scholar.google.com/citations?user=edEh3UMAAAAJ&hl=ca)  
[ACM MMSports 2025](http://mmsports.multimedia-computing.de/mmsports2025/cfp.html)

## 📦 Contents

- Highlight summary annotations
- Train / validation / test splits
- Pre-extracted visual features (no raw videos)
  
All data are provided as `.npy` feature arrays, `.srt` temporal annotations, and `.json` metadata files.

## 🌍 Data Source
Originally hosted at:
https://github.com/IPCV/SoccerHigh

## ⚠️ Important Note

**Raw videos are NOT included.**  
Videos must be obtained separately from the SoccerNet dataset:
https://huggingface.co/datasets/SoccerNet/SoccerNet_raw_HQ

The provided features are **non-invertible** and intended solely for research purposes.

## 📂 Dataset Structure

The dataset is organized hierarchically:

```text
train.txt
validation.txt
test.txt
<league>/
├── <season>/
│   ├── <game>/
│   │   ├── 1_HQ_224p_VideoMAEv2_Giant_K710_1408.npy
│   │   ├── 1_HQ_224p_VideoMAEv2_SmallFromGiant_K710_384.npy
│   │   ├── 1_intervals.srt
│   │   ├── 2_HQ_224p_VideoMAEv2_Giant_K710_1408.npy
│   │   ├── 2_HQ_224p_VideoMAEv2_SmallFromGiant_K710_384.npy
│   │   ├── 2_intervals.srt
│   │   ├── Labels-summary.json
```

### 📝 Files per game

- **`1_HQ_224p_VideoMAEv2_Giant_K710_1408.npy`**  
  Frame features from the game's first half, extracted with the [VideoMAEv2-Giant](https://huggingface.co/OpenGVLab/VideoMAE2/tree/main/mae-g) backbone.

- **`1_HQ_224p_VideoMAEv2_SmallFromGiant_K710_384.npy`**  
  Frame features from the game's first half, extracted with the [VideoMAEv2-SmallFromGiant](https://huggingface.co/OpenGVLab/VideoMAE2/tree/main/distill) backbone.

- **`1_intervals.srt`**  
  Annotated temporal segments for the first half in `.srt` format.  

- **`2_HQ_224p_VideoMAEv2_Giant_K710_1408.npy`**  
  Frame features from the game's second half, extracted with the [VideoMAEv2-Giant](https://huggingface.co/OpenGVLab/VideoMAE2/tree/main/mae-g) backbone.

- **`2_HQ_224p_VideoMAEv2_SmallFromGiant_K710_384.npy`**  
  Frame features from the game's second half, extracted with the [VideoMAEv2-SmallFromGiant](https://huggingface.co/OpenGVLab/VideoMAE2/tree/main/distill) backbone.

- **`2_intervals.srt`**  
  Annotated temporal segments for the second half in `.srt` format.

- **`Labels-summary.json`**  
  Metadata describing the game (teams, date, score, video URLs, and annotations).

## 🚀 Usage

This dataset is distributed as structured files (features and annotations).
Users can download the data directly from Hugging Face and load it using custom data pipelines.

A Hugging Face `datasets` loading script is not provided at this time.

## 📖 Citation

```bibtex
@inproceedings{10.1145/3728423.3759410,
  author = {D\'{\i}az-Juan, Artur and Ballester, Coloma and Haro, Gloria},
  title = {SoccerHigh: A Benchmark Dataset for Automatic Soccer Video Summarization},
  year = {2025},
  isbn = {9798400711985},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  url = {https://doi.org/10.1145/3728423.3759410},
  doi = {10.1145/3728423.3759410},
  booktitle = {Proceedings of the 8th International ACM Workshop on Multimedia Content Analysis in Sports},
  pages = {121–130},
  numpages = {10},
  location = {Dublin, Ireland},
  series = {MMSports '25}
}
```

## ⚖️ Legal Notice

This dataset contains only annotations and non-invertible feature representations derived from videos available in the SoccerNet dataset.

Redistribution of raw videos is **not permitted**.
Any access to or use of the original videos must comply with the SoccerNet license and terms of use.