SoccerNet datasets are now hosted on Hugging Face.
The official SoccerNet Hugging Face organization provides access to our raw videos, features, annotations, task-specific datasets, and yearly SoccerNet Challenge datasets.
🤗 Browse all datasets → SoccerNet datasets on Hugging Face
🎥 Raw videos → SoccerNet Raw HQ
📊 Features → SN-Features
🏷️ Labels → SN-Labels
Public datasets can be downloaded directly. Some datasets containing protected video content are gated and require a Hugging Face account and acceptance of the SoccerNet Non-Disclosure Agreement (NDA).
Sign in to your Hugging Face account.
Open the SoccerNet dataset you want to access.
Complete the access request directly on Hugging Face and accept the NDA conditions.
Once your request is approved, access is granted to your Hugging Face account.
A personalized copy of the NDA is sent to the email address associated with your Hugging Face account.
You can download SoccerNet data either directly from Hugging Face or through our SoccerNet Python package.
Install the tools:
`pip install -U SoccerNet huggingface_hub`
For gated datasets, authenticate once:
`hf auth login`
The SoccerNet Python package now downloads migrated datasets from HuggingFace by default while keeping the familiar SoccerNet download API.
We provide a pip package to easily download the data, including all videos, images and labels for our different tasks.
Feel free to check out our python package: SoccerNet and install it with simple python commands. Here are the steps to take before any download:
500 + 50 broadcast videos in .mkv at 25 fps and 720p or 224p resolution
The previous NDA (Non-Disclosure Agreement) form is deprecated, please request access throught HuggingFace.
Pre-trained features extracted from the broadcast videos at 2 frames per second
Images of some actions from the action spotting annotations and the same moment in their corresponding replays
Action spots on the 500 broadcast videos for the action spotting task
Camera shots on the 500 broadcast videos for the replay grounding task
Line segments on the action and replay images for the camera calibration tasks
Player correspondances on the action and replay images for the re-identification task