boboyes/leagueoflegends-minimap-detection

YOLOv11 - League of Legends Minimap Detection

🦁 Model Zoo & Performance

This repository contains a collection of YOLOv11 models fine-tuned on high-quality replay data. We provide multiple model sizes to suit different hardware constraints, from real-time edge devices to high-performance GPUs.

These results come from the real replay data tests including full occlusions, which are impossible to detect without additional metadata(e.g. prev frames, death logs) which significantly degrade the numbers.

Model	mAP@50	mAP@50-95	Precision	Recall
yolo11n	0.773	0.640	0.931	0.704
yolo11s	0.814	0.680	0.943	0.756
yolo11m	0.851	0.728	0.915	0.812
yolo11l	0.852	0.738	0.947	0.802
yolo11x	0.859	0.737	0.939	0.817

Model Description

These models are fine-tuned versions of Ultralytics YOLOv11, trained specifically to detect League of Legends champions on the minimap.

• Task: Object Detection (Minimap Icons)
• Training Data: Trained on boboyes/leagueoflegends-minimap + finetuned on real replay data(check the github repo)
• Architecture: YOLOv11 (n/s/m/l/x variants)

Intended Use

• Esports Analysis: Automated tracking of champion movements for post-game analytics.
• Content Creation: Auto-cam tools for observing specific lanes or fights.
• Research: Multi-object tracking (MOT) in crowded, low-resolution environments.

⚠️ Anti-Cheat Warning: This model is intended for research and post-game analysis ONLY. Using this for real-time gameplay advantages (scripting/overlay hacks) is strictly against Riot Games' Terms of Service and will result in a permanent account ban.

💻 How to Use

Installation

pip install ultralytics huggingface_hub