Paper Overview
Field: Computer Vision (CV) Authors: Kerui Chen, Jinglu Wang, Xiaoyi Zhang, Yan Lu Published: 2026-07-13 arXiv: 2607.11844
Abstract
Recent Multimodal Large Language Models (MLLMs) achieve strong performance on single-view video understanding benchmarks. However, sports videos involve dense occlusion, rapid motion, and complex interactions that are difficult to resolve from a single viewpoint. In practice, sports events are recorded from multiple camera angles, providing complementary evidence used by referees. Yet, no existing benchmark evaluates MLLMs on multi-view sports video understanding.
To address this gap, the authors introduce SportMV-Bench, a comprehensive benchmark built from official match recordings, created through a dedicated pipeline combining LLM-based generation, MLLM-based verification, and human filtering to ensure quality and consistency. The benchmark contains 787 multi-view video bundles and 2,592 question-answering pairs, spanning three task categories:
1. PAR — Perception And Recognition 2. REI — Rule-aware Event Interpretation 3. ADR — Adjudication Decision Reasoning
Key Findings
- Current MLLMs fail to effectively leverage multi-view information.
- The bottleneck lies in fine-grained visual perception and viewpoint selection, rather than logical reasoning or domain knowledge.
- Proactive viewpoint selection
- Perception tool execution
- Evidence-grounded reasoning
Proposed Solution: SportMV-Agent
The authors propose SportMV-Agent, an agentic framework that performs:
---
Paper link: https://arxiv.org/abs/2607.11844
*Auto-collected on 2026-07-15*