E-BARD: Basketball Player and Referee Detection from 2024-25 NBA Games

E-BARD extends the original Basketball Action Recognition Dataset with granular annotations for multi-stage computer vision pipelines. The dataset was extracted from 60 full-game recordings from the 2024-2025 NBA season, covering all 30 teams. It was introduced by GabrieleGiudici in a paper addressing the need for player detection, team attribution, and jersey number recognition.

Use Cases

• Train object detection models to locate players and referees based on the described annotations.
• Develop team attribution models based on jersey colors or other visual features mentioned in the description.
• Build jersey number recognition systems using the dataset's granular annotations.
• Create multi-task learning pipelines for simultaneous player detection and attribute recognition as referenced in the paper.

Strengths

• Extracted from 60 full-game recordings, providing substantial source material.
• Covers all 30 NBA teams from the 2024-2025 season, offering broad team representation.
• Addresses a specific gap for multi-stage pipelines with granular annotations.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect temporal and league bias inherent to the 2024-2025 NBA season.

Provenance

Source

huggingface

Collection Method

Extracted from full-game recordings.

Time Range

2024-2025 NBA season

Freshness

Last updated 2026-03-09 10:21:43; freshness should be verified.

Geography

us