Abstract
Recent advances in soccer understanding have demonstrated rapid progress, yetexisting research predominantly focuses on isolated or narrow tasks. To bridgethis gap, we propose a comprehensive framework for holistic soccerunderstanding. Concretely, we make the following contributions in this paper:(i) we construct SoccerWiki, the first large-scale multimodal soccer knowledgebase, integrating rich domain knowledge about players, teams, referees, andvenues to enable knowledge-driven reasoning; (ii) we present SoccerBench, thelargest and most comprehensive soccer-specific benchmark, featuring around 10Kmultimodal (text, image, video) multi-choice QA pairs across 13 distinct tasks;(iii) we introduce SoccerAgent, a novel multi-agent system that decomposescomplex soccer questions via collaborative reasoning, leveraging domainexpertise from SoccerWiki and achieving robust performance; (iv) extensiveevaluations and comparisons with representative MLLMs on SoccerBench highlightthe superiority of our agentic system.