HockeyAI: Multi-Class Ice Hockey Object Detection with 1,000+ Images

SimulaMet-HOST published this dataset in 2025, providing between 1,000 and 10,000 images for multi-class ice hockey object detection. The collection facilitates the identification of game-specific elements like pucks and players in rink settings.

Use Cases

• Training object detection models to locate the 'puck' in high-speed sports footage
• Developing player tracking systems using multi-class labels
• Benchmarking computer vision performance on small-object detection in sports

Strengths

• 1,000 to 10,000 labeled records
• Multi-class annotations for specialized sports objects
• MIT licensed for open research

Limitations

• Small sample size (max 10,000 images) compared to general-purpose vision datasets
• Geographic bias toward US rinks as indicated by metadata tags

Provenance

Source

SimulaMet-HOST

Freshness

Last updated December 2025.

Geography

United States