CourtSI-Bench: Spatial Intelligence Benchmark for Sports VLMs

CourtSI-Bench is a spatial intelligence benchmark for vision-language models (VLMs) focusing on high-intensity human motion and dynamic object interactions in sports. Released by Charlie019 in March 2026, the collection contains between 1,000 and 10,000 records. It serves as a specialized testbed for understanding physical and cognitive capabilities in athletic contexts.

Use Cases

• Evaluating VLM spatial reasoning using visual question-answering pairs
• Benchmarking model performance on high-intensity motion sequences in sports imagery
• Testing dynamic object interaction understanding through text-based queries

Strengths

• Contains 1,000 to 10,000 curated records for benchmarking
• Apache 2.0 license allows for broad research application
• Specialized focus on high-intensity human motion and dynamic object interactions

Limitations

• Small sample size (under 10,000 records) compared to general-purpose VLM training sets
• Domain-specific focus on sports may limit generalizability to non-athletic spatial contexts

Provenance

Source

Charlie019

Collection Method

Annotated benchmark for vision-language models

Freshness

Last updated March 2026.