SoccerBench: 10,000 Multimodal QA Pairs for Soccer Understanding

SoccerBench contains approximately 10,000 standardized multimodal multiple-choice question-answering pairs across 14 distinct soccer understanding tasks. Developed by Homie0609 and released in 2025, the benchmark utilizes automated pipelines combined with manual verification to evaluate multi-agent systems.

Use Cases

• Evaluating multi-agent systems on soccer-specific video understanding tasks
• Training multimodal models to answer multiple-choice questions about soccer imagery
• Benchmarking temporal reasoning in soccer match videos using the 14 task categories

Strengths

• 10,000 standardized QA pairs
• 14 distinct soccer understanding tasks
• Manual verification of automated pipeline outputs

Limitations

• Restricted to multiple-choice format which may not capture open-ended reasoning
• Non-commercial license (CC BY-NC-SA 4.0) limits industrial application

Provenance

Source

Homie0609 (Arxiv: 2505.03735)

Collection Method

Automated pipelines followed by manual verification

Freshness

Last updated November 2025.