Computer Vision for MOBA Analytics: A Dataset and Baseline for Visibility Analysis in Dota 2

Traditional MOBA analytics rely heavily on structured data, such as event logs and APIs, which often fail to capture the critical strategic dimension of visibility. In Dota 2, knowing what an opponent can see is essential for map control and decision-making, yet this is obscured by complex game mechanics like fog of war, terrain elevation, and vision-blocking structures. The authors address this gap by proposing a computer vision pipeline that analyzes gameplay footage to determine when players are visible to the opposing team.

To support this, they created Dota2-Vis, a dataset consisting of 288 full-match videos from The International 2025 (recorded from both team perspectives) and 2,477 manually annotated minimap images. By training a YOLO11l object detector on these minimap images, the researchers developed a baseline to track player icons and estimate visibility curves over time.

The YOLO11l model proved highly effective, achieving an F-score of 0.939 and demonstrating robustness even in visually dense scenes. By aggregating these detections, the authors generated temporal visibility curves that provide new insights into player and team behavior. For instance, they demonstrated that different professional players exhibit distinct visibility profiles while playing the same hero, reflecting different strategic approaches (e.g., aggressive pressure vs. conservative resource gathering). Additionally, team-level analysis shows distinct visibility trends correlated with match outcomes, offering a quantitative tool for post-match review and scouting.

This work bridges the gap between raw video data and high-level game analytics. By moving beyond structured logs, researchers and coaches can now analyze the "fog of war" dynamics that define professional MOBA strategy. The dataset and code are publicly available, providing a foundation for future studies on player intent, map control, and the strategic value of information in esports.