Pre-release

This release is a pre-release and may not be stable for production use.

OpenSportsLib

OpenSportsLib is a modular Python library for sports video understanding.

It provides a unified framework to train, evaluate, and run inference for key temporal understanding tasks in sports video, including:

Action classification
Action localization / spotting
Visual Question Answering (VQA)

Retrieval and action description/captioning are roadmap areas. They do not yet have first-class task wrappers or training workflows in this package.

OpenSportsLib is designed for researchers, ML engineers, and sports analytics teams who want reproducible and extensible workflows for sports video AI.

Why OpenSportsLib?

Unified workflow for training and inference
Modular design for adding new tasks, datasets, and models
Config driven experiments for reproducibility
Optional SpoTTA test-time adaptation for E2ESpot inference
Support for multiple modalities and sports workflows
Research friendly while still usable in applied settings

Quick links

Documentation: https://opensportslab.github.io/opensportslib/
OSL JSON format: https://opensportslab.github.io/opensportslib/data/osl-json-format/
Inference server: server/README.md
PyPI: https://pypi.org/project/opensportslib/
Issues: https://github.com/OpenSportsLab/opensportslib/issues

Installation

Requires Python 3.12+.
Supports CUDA 12.6 / 12.8 / 13.0 (with CPU fallback).
PyTorch Geometric uses a dedicated PyTorch 2.12.1 compatibility profile.

Create conda env

conda create -n osl python=3.12 pip -y
conda activate osl

Stable release

pip install opensportslib

Pre release

pip install --pre opensportslib

Source development version

pip install -e .

Setup Environment (PyTorch, CUDA aware & Optional Dependencies)

# Install PyTorch (CPU/GPU auto-detected)
opensportslib setup

# Optional: install PyTorch Geometric support. This replaces the installed
# Torch stack with the PyG-compatible PyTorch 2.12.1 profile.
opensportslib setup --pyg

# Optional: install for DALI support
opensportslib setup --dali

# Optional: install the X-VARS-compatible VQA dependency profile
opensportslib setup --vqa_xvars

# Optional: install the Qwen-compatible VQA dependency profile
opensportslib setup --vqa_qwen

Note:
Run opensportslib setup to automatically configure dependencies.
If issues occur, manually install compatible versions of torch, torchvision, and related libraries according to your CUDA version or system compatibility.

For VQA, use exactly one backend-specific dependency profile:

--vqa_xvars installs the X-VARS-compatible Hugging Face stack from XVARS_DEPENDENCY_PINS
--vqa_qwen installs the Qwen-compatible Hugging Face stack from QWEN_DEPENDENCY_PINS

The vqa_qwen config supports Qwen/Qwen2.5-7B-Instruct and Qwen/Qwen3.5-9B-Base.

Data and pretrained models

OpenSportsLib uses external annotation files, datasets, and pretrained checkpoints.

Public assets are hosted under the OpenSportsLab Hugging Face organization:

https://huggingface.co/OpenSportsLab

Use it as the main entry point to find:

datasets
annotation files
extracted features
pretrained models and checkpoints

See the Model Zoo for available pretrained models, reported scores, datasets, and loading snippets.

Dataset format

OpenSportsLib uses the OSL JSON v2.0 annotation format for multimodal datasets and predictions. See the OSL JSON format guide for its schema, examples, and conversion notes.

Quickstart

Classification

from opensportslib.apis import ClassificationModel

my_model = ClassificationModel(
    config="/path/to/classification.yaml",
    weights=None,  # optional: path or Hugging Face model ID
)

my_model.train(
    train_set="/path/to/train_annotations.json",
    valid_set="/path/to/valid_annotations.json",
)

predictions = my_model.infer(test_set="/path/to/test_annotations.json")
my_model.save_predictions(
    output_path="/path/to/predictions.json",
    predictions=predictions,
)

For localization, VQA, and additional end-to-end examples, see the API guide, quickstart scripts, and VQA guide.

Hugging Face Dataset Transfer

OpenSportsLib provides APIs and scripts for downloading and uploading OSL datasets with Hugging Face.

For SN-GAR classification and action-spotting configurations, setup, caching, and training commands, see the SN-GAR examples.

Python API

from opensportslib.tools import (
    download_dataset_split_from_hf,
    download_dataset_sample_inputs_from_hf,
    upload_dataset_inputs_from_json_to_hf,
    upload_dataset_as_parquet_to_hf,
)

Scripts

python tools/download/download_osl_hf.py --repo-id <org/repo> --revision main --split test --format parquet --output-dir downloaded_data --annotations-only
python tools/download/upload_osl_hf.py --repo-id <org/repo> --json-path <local_dataset.json> --split test --revision main

Examples and documentation

Use the README for the fast start, then go deeper through:

Full documentation: https://opensportslab.github.io/opensportslib/
OSL JSON format: docs/data/osl-json-format.md
High-level API guide: opensportslib/apis/README.md
Configuration guide: https://opensportslab.github.io/opensportslib/config/configuration-guide/
Example configs: examples/configs/
Quickstart scripts: examples/quickstart/
Contribution guide: CONTRIBUTING.md
Developer guide: DEVELOPERS.md

Development setup

For contributors who want to work from source:

git clone https://github.com/OpenSportsLab/opensportslib.git
cd opensportslib
pip install -e .

Conda option

If you prefer conda:

conda create -n osl python=3.12 pip
conda activate osl
pip install -e .

Setup Environment (PyTorch, CUDA aware & Optional Dependencies)

# Install PyTorch (CPU/GPU auto-detected)
opensportslib setup

# Optional: install PyTorch Geometric support. This replaces the installed
# Torch stack with the PyG-compatible PyTorch 2.12.1 profile.
opensportslib setup --pyg

# Optional: install for DALI support
opensportslib setup --dali

# Optional: install the X-VARS-compatible VQA dependency profile
opensportslib setup --vqa_xvars

# Optional: install the Qwen-compatible VQA dependency profile
opensportslib setup --vqa_qwen

Contributing

We welcome contributions. Pull requests must target dev, and each GitHub-linked commit author must accept the Individual Contributor License Agreement when prompted by the CLA check.

See CONTRIBUTING.md for the contribution workflow and DEVELOPERS.md for architecture and extension guidance.

License

OpenSportsLib is available under dual licensing.

Open source license

AGPL 3.0 for research, academic, and community use.

Commercial license

For proprietary or commercial deployment, please refer to LICENSE-COMMERCIAL.

Citation

If you use OpenSportsLib in your research, please cite the project.

@misc{opensportslib,
  title={OpenSportsLib},
  author={OpenSportsLab},
  year={2026},
  howpublished={\url{https://github.com/OpenSportsLab/opensportslib}}
}

Acknowledgments

OpenSportsLib is developed within the broader OpenSportsLab effort for sports video understanding. Core contributors affiliated with KAUST include:

Jeet Vora — Remote Research Engineer
Dr. Merey Ramazanova — Post-Doc
Dr. Silvio Giancola — Research Scientist

Metadata

Release files for opensportslib 0.3.1.dev25

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for opensportslib 0.3.1.dev25
File	Size	Uploaded
opensportslib-0.3.1.dev25.tar.gz	519.9 kB	Sep 30, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for opensportslib 0.3.1.dev25
File	Interpreter	ABI	Platform	Reset
opensportslib-0.3.1.dev25-py3-none-any.whl 631.5 kB Sep 30, 2026	Python 3	none	any	Details

Total release size: 1.2 MB

Release files / opensportslib-0.3.1.dev25.tar.gz

Download URL	opensportslib-0.3.1.dev25.tar.gz
Size	519.9 kB
Tags	Source
SHA-256 checksum How to use checksums	`1ad7e3b1914af142537a30cfae3d0821cd2d48ceb5d50584ecff8e40d73b56ee`
BLAKE2b-256 checksum How to use checksums	`25cb676c9bb557411e09d64268e052a81433cba2c772a8b929db3b81257e4220`
Upload date	Sep 30, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.14`

Release files / opensportslib-0.3.1.dev25-py3-none-any.whl

Download URL	opensportslib-0.3.1.dev25-py3-none-any.whl
Size	631.5 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`210b069203e6d2cc99baf9ff6d5c9d41891d6a1b647f0270860070b7f751b0c4`
BLAKE2b-256 checksum How to use checksums	`b03c975dc6a4f1fb66266e2bd741b99b4cc4d28064c0d44c006d29eb0bbfd582`
Upload date	Sep 30, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.14`