🏀👁️ BARD: A Basketball Action Recognition Dataset for Multi-Label Classification

🔍 Abstract

We present the BARD dataset (Basketball Action Recognition Dataset). It is designed to advance video action recognition in basketball through high-quality annotations and enriched contextual data. BARD improves upon existing datasets by including player jersey numbers, team colors and a novel output format supporting multi-label classification. To ensure annotation quality, we conducted a human validation study on a subsample of the annotations, with expert reviewers assessing the labeling quality and reporting the evaluation results, thereby providing human validated independent benchmarks. Moreover, in addition to standard caption-based action recognition metrics, we introduce Basketball Caption Evaluation Framework (BaCEF), a new application-oriented evaluation framework. Finally, to demonstrate the quality and challenging nature of the dataset, as well as the utility of our evaluation framework and its potential applications, we evaluate both proprietary models (e.g., Gemini 2.5 Pro) and open-source models (Qwen2.5-VL-7B-Instruct, Qwen2.5-VL-3B-Instruct), including BQwen2.5-VL-3B, a BARD fine-tuned variant of Qwen2.5-VL-3B-Instruct, across our defined benchmarks.

📦 Dataset

This Hugging Face repository contains the BARD dataset released as part of the BARD project.

The complete project, including the dataset generation pipeline, preprocessing scripts, annotation generation, validation, evaluation code, and additional resources, is available in the official GitHub repository:

https://github.com/GabrieleGiudic/BARD

The dataset contains 14,676 final video clips, obtained after filtering and consolidation from an initial collection of 24,692 clips. The videos correspond to 60 sampled games involving 30 selected NBA teams from the 2024–2025 season.

📁 Repository Structure

The repository is organized primarily by game. Each game has its own directory containing the corresponding video clips:

BARD/
├── bkn-vs-det-0022400861/
├── bos-vs-was-0022401217/
├── chi-vs-bos-0022400363/
├── ...
├── nyk-vs-orl-0022401227/
├── ...
└── captions/
    └── caption.json

The captions/caption.json file contains the captions associated with the video clips and can be used for video-language and caption-based action recognition experiments.

Additional annotation and processing files used to construct the dataset are available in the original BARD project repository.

📚 Citation

If you use BARD in your research, please cite:

@article{giudici2026bard,
  title={BARD: A Basketball Action Recognition Dataset for multi-label classification},
  author={Giudici, Gabriele and Maurino, Andrea and Zuccolotto, Paola},
  journal={Computer Vision and Image Understanding},
  pages={104713},
  year={2026},
  publisher={Elsevier}
}

🧑‍⚖️ License

This project is licensed under the Creative Commons Attribution 4.0 International License.

For further information about the dataset, its construction, annotations, processing pipeline, and usage, please refer to the official BARD repository:

https://github.com/GabrieleGiudic/BARD