We present the BARD dataset. It is designed to advance Basketball Action Recognition task 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. Finally, we benchmark BARD using the Gemini model, demonstrating its effectiveness for structured, multi-label action recognition tasks.
Property | Value | Description |
---|---|---|
Season | 2024–2025 | Most updated season |
Teams | 30 | Selected NBA teams |
Games | 60 | Total number of games sampled |
Initial clips | 24,692 | Raw video segments collected |
Final clips | 14,676 | After filtering and consolidation |
Resolution | 720p | High-definition video |
Labels | Structured JSON | Multi-label format |
Action recognition | Coarse and Event-level | Play-by-play annotation |
New fields | Player numbers, team colors | Anonymous identification metadata |
Clip:
Green Tip Layup Shot (21 PTS)
[ { "player": "00", "action": "2PT Shot", "result": false, "assisted": false, "other_player": null, "color": "blue" }, { "player": "23", "action": "Rebound", "result": null, "assisted": null, "other_player": null, "color": "blue" }, { "player": "23", "action": "2PT Shot", "result": false, "assisted": false, "other_player": null, "color": "blue" }, { "player": "23", "action": "Rebound", "result": null, "assisted": null, "other_player": null, "color": "blue" }, { "player": "23", "action": "2PT Shot", "result": true, "assisted": false, "other_player": null, "color": "blue" } ]
We have open-sourced the code for downloading the data and reproducing the results presented in our paper.
If you use BARD in your work, please cite the associated paper.