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

2026-06-25 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Gaming & Interactive Media · Depth: Expert, quick

Summary

A new video-based dataset, Dota2-Vis, and a baseline pipeline have been introduced for visibility analysis in professional Dota 2 matches. Addressing the limitation of traditional MOBA analytics that often overlook actual team vision, this work provides a comprehensive resource. The dataset comprises all 144 matches from The International 2025, totaling 288 Full HD videos recorded from both team perspectives, alongside 2,477 manually annotated minimap images. Researchers evaluated multiple object detectors for player-icon detection, with YOLO11l (large) achieving the best performance in identifying player icons even in cluttered minimap scenes. The resulting visibility curves offer novel player, hero, role, and team-level patterns, enriching conventional MOBA analytics by revealing behavioral differences not obtainable from structured data alone. The dataset and code are publicly available.

Key takeaway

For Computer Vision Engineers developing game analytics, this work offers a critical new resource. You should explore the Dota2-Vis dataset and its YOLO11l baseline to enhance your understanding of in-game visibility. This approach allows you to move beyond structured data, revealing player and team behavioral patterns previously difficult to discern. Consider adapting these computer vision techniques to other MOBA titles or complex game UIs to gain deeper strategic insights.

Key insights

Video-based computer vision can reveal nuanced visibility patterns in MOBA games, complementing traditional structured data analytics.

Principles

• Video-based analysis enhances MOBA analytics beyond structured data.
• Object detection reliably extracts player visibility from game minimaps.

Method

Record both team perspectives of matches, manually annotate minimap images, then apply object detection (e.g., YOLO11l) to identify player icons and derive visibility curves.

In practice

• Utilize Dota2-Vis dataset for MOBA visibility research.
• Apply object detection to game UI elements for novel analytics.

Topics

• Computer Vision
• MOBA Analytics
• Dota 2
• Dota2-Vis Dataset
• Object Detection
• Visibility Analysis

Code references

• RicardoRCarvalho/dota2-vis

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• Learning Perceptual Representations for Gaming NR-VQA with Multi-Task FR Signals — MTL-VQA uses multi-task learning with FR signals to enable label-efficient NR-VQA for gaming videos.
• HYolo: An Intelligent IoT-Based Object Detection System Using Hypergraph Learning — HYolo integrates hypergraph learning into YOLO to model complex high-order relationships, improving object detection in IoT.
• XPASS-Vis: A Dataset for Cross-Domain Personalized Image Aesthetic Assessment — Personalized aesthetic preferences are meaningfully transferable across visual domains, despite individual subjectivity.
• SVHighlights: Towards Extremely Long Sport Video Highlight Detection — The SVHighlights benchmark and TF-SELECTOR method enable effective highlight detection in extremely long sports videos.
• 346 Blog Posts To Learn About Computer Vision — Computer Vision leverages AI to interpret visual data, enabling diverse applications from image generation to real-world object detection.
• Grid2Matrix: Revealing Digital Agnosia in Vision-Language Models — VLMs exhibit "Digital Agnosia," failing to express fine-grained visual details despite their vision encoders retaining the information.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.