| --- |
| title: README |
| emoji: 💻 |
| colorFrom: indigo |
| colorTo: gray |
| sdk: static |
| pinned: false |
| --- |
| |
| # ⚽ SoccerNet |
|
|
| ### Soccer Video Understanding Benchmark Suite |
|
|
| **Datasets · Benchmarks · Challenges · Research** |
|
|
| [🌐 Website](https://www.soccer-net.org) · |
| [🤗 Datasets](https://huggingface.co/SoccerNet/datasets) · |
| [💻 GitHub](https://github.com/SoccerNet) · |
| [💬 Discord](https://discord.gg/cPbqf2mAwF) |
|
|
|
|
| --- |
|
|
| ## ⚽ What is SoccerNet? |
|
|
| **SoccerNet** is a large-scale benchmark suite for **soccer video understanding**. |
|
|
| Started in 2018 with action spotting in broadcast soccer games, SoccerNet has progressively expanded toward a broad range of problems in video understanding, field understanding, player understanding, and game understanding. |
|
|
| The data currently includes: |
|
|
| - **550 complete broadcast games** |
| - **12 single-camera games** |
| - **300K+ temporal annotations** |
| - **1M+ spatial annotations** |
|
|
| Beyond the datasets, SoccerNet organizes yearly international challenges where researchers and practitioners benchmark their methods on common tasks and evaluation protocols. |
|
|
| 🌐 **Website:** [soccer-net.org](https://www.soccer-net.org) |
|
|
| --- |
|
|
| ## 🔐 Dataset access and NDA |
|
|
| Some SoccerNet datasets contain protected video and require NDA-based access. For datasets marked **NDA-gated**, follow these steps: |
|
|
| 1. Sign in to your [Hugging Face account](https://huggingface.co/login). |
| 2. Open the dataset repository you want to access. |
| 3. Complete the requested information, read the linked SoccerNet Non-Disclosure Agreement, and electronically sign it using the acknowledgment checkbox. |
| 4. Submit your access request and wait for manual approval by the SoccerNet team. |
| 5. Once approved, download the dataset using the same Hugging Face account. |
|
|
| **Access is granted separately for each repository.** Approval for `SoccerNet_raw_HQ` does not automatically grant access to `SN-BAS-2024`, `SN-BAS-2025`, or `SN-PCBAS-2026`. Existing approvals for these repositories remain valid. |
|
|
| ### Download after approval |
|
|
| Install the Hugging Face command-line tool and sign in with the account that received access: |
|
|
| ```bash |
| pip install --upgrade huggingface_hub |
| hf auth login |
| ``` |
|
|
| If prompted for an access token, use one from the approved account with read access to the dataset. For example, download `SN-BAS-2025` with: |
|
|
| ```bash |
| hf download SoccerNet/SN-BAS-2025 --repo-type dataset --local-dir SoccerNet/SN-BAS-2025 |
| ``` |
|
|
| Replace the repository ID and local directory with the dataset you want to download. Request access on that repository's page before downloading it. |
|
|
| --- |
|
|
| ## 🤗 Datasets on Hugging Face |
|
|
| The official SoccerNet Hugging Face organization currently hosts the following datasets. |
|
|
| ### Core and research datasets |
|
|
| | Dataset | Description | |
| |---|---| |
| | **[SoccerNet_raw_HQ](https://huggingface.co/datasets/SoccerNet/SoccerNet_raw_HQ)** | Gated repository for raw broadcast video (train/valid/test) — HQ, 720p, 224p, frames, and clips | |
| | **[SoccerNet_raw_HQ_Challenge](https://huggingface.co/datasets/SoccerNet/SoccerNet_raw_HQ_Challenge)** | Gated repository for raw broadcast video, challenge split — HQ, 720p, 224p, LQ | |
| | **[SoccerNet-Tracking-RAW-Video](https://huggingface.co/datasets/SoccerNet/SoccerNet-Tracking-RAW-Video)** | Gated repository for raw single-camera footage used in the tracking benchmark | |
| | **[SN-Features](https://huggingface.co/datasets/SoccerNet/SN-Features)** | Pre-extracted per-game features (frame embeddings, ResNET, player bounding boxes, field calibration), shared across editions | |
| | **[SN-Labels](https://huggingface.co/datasets/SoccerNet/SN-Labels)** | Per-game labels (action spotting, dense captioning, camera shots), shared across editions | |
| | **[ActionAnticipation](https://huggingface.co/datasets/SoccerNet/ActionAnticipation)** | Action anticipation dataset used for the SoccerNet 2026 Action Anticipation Challenge | |
| | **[SN-echoes](https://huggingface.co/datasets/SoccerNet/SN-echoes)** | SoccerNet-Echoes audio commentary dataset | |
| | **[BannerReplacement](https://huggingface.co/datasets/SoccerNet/BannerReplacement)** | SoccerNet banner replacement dataset | |
|
|
| ### SoccerNet 2027 |
|
|
| The 2027 challenge includes four tasks. See the [challenge website](https://www.soccer-net.org/challenges/2027) for details. The submission deadline is **April 25, 2027, 23:59 Anywhere on Earth (UTC−12)**. |
|
|
| | Dataset | Task | Challenge repository | |
| |---|---|---| |
| | **[OpenSportsLab/SoccerNet-GAR](https://huggingface.co/datasets/OpenSportsLab/SoccerNet-GAR)** | Group Activity Recognition; hosted by OpenSportsLab | [sn-gar-2027](https://github.com/SoccerNet/sn-gar-2027) | |
| | **[SN-VQA-2027](https://huggingface.co/datasets/SoccerNet/SN-VQA-2027)** | Streaming Visual Question Answering | [sn-vqa-2027](https://github.com/SoccerNet/sn-vqa-2027) | |
| | **[SN-GSR-2027](https://huggingface.co/datasets/SoccerNet/SN-GSR-2027)** | Long Form Game State Reconstruction | [sn-gsr-2027](https://github.com/SoccerNet/sn-gsr-2027) | |
| | **[SN-NVS-2027](https://huggingface.co/datasets/SoccerNet/SN-NVS-2027)** | Novel View Synthesis | [sn-nvs-2027](https://github.com/SoccerNet/sn-nvs-2027) | |
|
|
| ### SoccerNet 2026 |
|
|
| | Dataset | Task | |
| |---|---| |
| | **[SN-PCBAS-2026](https://huggingface.co/datasets/SoccerNet/SN-PCBAS-2026)** | Player-Centric Ball Action Spotting — NDA-gated; password-free ZIPs | |
| | **[SN-NVS-2026](https://huggingface.co/datasets/SoccerNet/SN-NVS-2026)** | Novel View Synthesis | |
| | **[SN-VQA-2026](https://huggingface.co/datasets/SoccerNet/SN-VQA-2026)** | Visual Question Answering | |
|
|
| ### SoccerNet 2025 |
|
|
| | Dataset | Task | |
| |---|---| |
| | **[SN-BAS-2025](https://huggingface.co/datasets/SoccerNet/SN-BAS-2025)** | Ball Action Spotting — NDA-gated; password-free ZIPs | |
| | **[SN-Depth-2025](https://huggingface.co/datasets/SoccerNet/SN-Depth-2025)** | Monocular Depth Estimation | |
| | **[SN-MVFouls-2025](https://huggingface.co/datasets/SoccerNet/SN-MVFouls-2025)** | Multi-View Foul Recognition | |
| | **[SN-GSR-2025](https://huggingface.co/datasets/SoccerNet/SN-GSR-2025)** | Game State Reconstruction | |
|
|
| ### SoccerNet 2024 |
|
|
| | Dataset | Task | |
| |---|---| |
| | **[SN-BAS-2024](https://huggingface.co/datasets/SoccerNet/SN-BAS-2024)** | Ball Action Spotting — NDA-gated; password-free ZIPs | |
| | **[SN-MVFouls-2024](https://huggingface.co/datasets/SoccerNet/SN-MVFouls-2024)** | Multi-View Foul Recognition | |
| | **[SN-GSR-2024](https://huggingface.co/datasets/SoccerNet/SN-GSR-2024)** | Game State Reconstruction | |
|
|
| ### SoccerNet 2023 |
|
|
| | Dataset | Task | |
| |---|---| |
| | **[SN-Calibration-2023](https://huggingface.co/datasets/SoccerNet/SN-Calibration-2023)** | Camera Calibration | |
| | **[SN-ReID-2023](https://huggingface.co/datasets/SoccerNet/SN-ReID-2023)** | Player Re-Identification | |
| | **[SN-Tracking-2023](https://huggingface.co/datasets/SoccerNet/SN-Tracking-2023)** | Player Tracking | |
| | **[SN-Jersey-2023](https://huggingface.co/datasets/SoccerNet/SN-Jersey-2023)** | Jersey Number Recognition | |
| | **[SN-BAS-2023](https://huggingface.co/datasets/SoccerNet/SN-BAS-2023)** | Ball Action Spotting (gated) | |
|
|
| Action Spotting and Dense Video Captioning for the 2023–2024 editions are served from the shared **[SN-Features](https://huggingface.co/datasets/SoccerNet/SN-Features)** / **[SN-Labels](https://huggingface.co/datasets/SoccerNet/SN-Labels)** repositories above rather than a per-edition repository. |
|
|
| ### Legacy (pre-2023) |
|
|
| | Dataset | Task | |
| |---|---| |
| | **[SN-Calibration](https://huggingface.co/datasets/SoccerNet/SN-Calibration)** | Camera Calibration | |
| | **[SN-Depth](https://huggingface.co/datasets/SoccerNet/SN-Depth)** | Monocular Depth Estimation (football and basketball) | |
|
|
| ➡️ **[Browse all SoccerNet datasets](https://huggingface.co/SoccerNet/datasets)** |
|
|
| --- |
|
|
| ## 🎯 Benchmark Tasks |
|
|
| SoccerNet covers tasks ranging from low-level field and player localization to high-level understanding of complete soccer games. |
|
|
| ### 📺 Broadcast Video Understanding |
|
|
| - **Action Spotting** — localize actions in time |
| - **Replay Grounding** — retrieve a replayed action in a broadcast |
| - **Camera Shot Segmentation** — segment camera shots and localize their boundaries |
| - **Ball Action Spotting** — localize ball actions in time |
| - **Dense Video Captioning** — spot salient moments and generate commentaries |
| - **Multi-View Foul Recognition** — recognize fouls and their severity |
|
|
| ### 🏟️ Field Understanding |
|
|
| - **Field Localization** — localize soccer pitch markings and goal posts |
| - **Camera Calibration** — estimate camera calibration |
| - **Monocular Depth Estimation** — estimate depth from a monocular camera |
|
|
| ### 👤 Player Understanding |
|
|
| - **Player Tracking** — track multiple players through video sequences |
| - **Player Re-Identification** — re-identify players across views |
| - **Jersey Number Recognition** — identify player jersey numbers from tracklets |
|
|
| ### 🧠 Game Understanding |
|
|
| - **Game State Reconstruction** — reconstruct the game state by tracking the ball and players in a top-view representation |
|
|
| ➡️ **[Explore all SoccerNet tasks](https://www.soccer-net.org/tasks)** |
|
|
| --- |
|
|
| ## 🏆 SoccerNet Challenges |
|
|
| The **SoccerNet Challenges** have been organized yearly since 2021 and provide standardized datasets, development kits, evaluation protocols, and leaderboards for soccer understanding research. |
|
|
| The [2027 edition](https://www.soccer-net.org/challenges/2027) features Group Activity Recognition, Streaming Visual Question Answering, Long Form Game State Reconstruction, and Novel View Synthesis. The submission deadline is **April 25, 2027, 23:59 Anywhere on Earth**. |
|
|
| The **2026 edition**, the sixth SoccerNet Challenges edition, included: |
|
|
| - **Spiideo SoccerNet SynLoc** |
| - **Ball Action Anticipation** |
| - **Visual Question Answering** |
| - **Player-Centric Ball Action Spotting** |
| - **Novel View Synthesis** |
| - **FIFA Innovation Challenge — Skeletal Tracking Light** *(guest task)* |
|
|
| Challenge resources are available through the SoccerNet website, Hugging Face datasets, GitHub development kits, and Codabench evaluation servers. |
|
|
| ➡️ **[Explore the SoccerNet Challenges](https://www.soccer-net.org/challenges)** |
|
|
| --- |
|
|
| ## 💻 Open-Source Code |
|
|
| Benchmark implementations, development kits, evaluation code, and research projects are maintained in the official **SoccerNet GitHub organization**. |
|
|
| Some of the main repositories include: |
|
|
| - [**sn-gamestate**](https://github.com/SoccerNet/sn-gamestate) — Game State Reconstruction |
| - [**sn-calibration**](https://github.com/SoccerNet/sn-calibration) — Camera Calibration |
| - [**sn-tracking**](https://github.com/SoccerNet/sn-tracking) — Player Tracking |
| - [**sn-spotting**](https://github.com/SoccerNet/sn-spotting) — Action Spotting |
| - [**sn-reid**](https://github.com/SoccerNet/sn-reid) — Player Re-Identification |
| - [**sn-mvfoul**](https://github.com/SoccerNet/sn-mvfoul) — Multi-View Foul Recognition |
| - [**sn-depth**](https://github.com/SoccerNet/sn-depth) — Monocular Depth Estimation |
| - [**sn-echoes**](https://github.com/SoccerNet/sn-echoes) — SoccerNet-Echoes |
| - [**sn-jersey**](https://github.com/SoccerNet/sn-jersey) — Jersey Number Recognition |
| - [**sn-nvs**](https://github.com/SoccerNet/sn-nvs) — Novel View Synthesis |
|
|
| ➡️ **[Browse the SoccerNet GitHub organization](https://github.com/SoccerNet)** |
|
|
| --- |
|
|
| ## 📚 Research |
|
|
| SoccerNet research spans soccer video understanding, player and field analysis, multimodal understanding, and sports AI. |
|
|
| The project has grown from the original **SoccerNet action spotting benchmark released in 2018** to a suite of datasets, tasks, benchmark methods, yearly challenges, and community contributions. |
|
|
| ➡️ **[Browse SoccerNet publications](https://www.soccer-net.org/publications)** |
|
|
| --- |
|
|
| ## 🤝 Join the SoccerNet Community |
|
|
| SoccerNet is an open research community bringing together researchers and practitioners working on **computer vision, video understanding, multimodal AI, sports analytics, and soccer**. |
|
|
| <div align="center"> |
|
|
| [🌐 Website](https://www.soccer-net.org) · |
| [🤗 Hugging Face](https://huggingface.co/SoccerNet) · |
| [💻 GitHub](https://github.com/SoccerNet) · |
| [💬 Discord](https://discord.gg/cPbqf2mAwF) |
|
|
| </div> |
|
|
