CourtSI-1M: 1 Million Spatial Intelligence Benchmarks for Sports VLMs

CourtSI-1M provides 1 million visual question-answering pairs for benchmarking spatial intelligence in sports, released by Charlie019 in March 2026. It serves as a testbed for vision-language models to interpret high-intensity human motion and dynamic object interactions.

Use Cases

• Benchmarking spatial reasoning in VLMs using visual question-answering pairs
• Training models to identify dynamic object interactions in sports footage
• Evaluating human motion understanding through spatial intelligence tasks

Strengths

• 1,000,000 record scale
• Apache 2.0 permissive license
• Specialized focus on high-intensity human motion

Limitations

• Geographic bias toward US-centric sports
• Domain-specific limitation to athletic environments

Provenance

Source

Charlie019 (Hugging Face)

Freshness

Last updated March 2026

Geography

United States