DeepLeague is an open-source computer vision project that applies deep learning to the League of Legends minimap, detecting what is happening in a match from the map alone. It has about 1,200 stars on GitHub and ships with a dataset of over 100,000 labeled minimap images.
Esports broadcasts and VODs contain a goldmine of strategic information on the minimap, but it is locked in pixels. DeepLeague trains object-detection models (built on a YOLO implementation in Keras) to find champions and other elements on the minimap frame by frame, turning raw video into structured data about player positions and movements. Alongside the code, the author released a dataset of more than 100,000 labeled minimap images to encourage further AI research in esports. The project is candidly a fast-built research hack rather than a polished library, but it remains a frequently cited starting point for game analytics with computer vision.
Champion and object positions extracted from the minimap enable downstream analytics that raw video cannot provide directly: heat maps of player movement, studies of early-game pathing and jungle routes, and automated detection of fights or objectives as they develop. The author presents the work openly as a learning project, and its labeled dataset has made it a recurring teaching example for applied object detection.
DeepLeague is free and open source, published on GitHub. The code and the labeled minimap dataset are available to download and experiment with at no cost.
DeepLeague is for computer vision students, esports analysts and hobbyists who want a fun, concrete dataset and codebase for learning object detection on real game footage.