CS2-10K: A Large-Scale Egocentric Counter-Strike 2 Dataset

CS2-10K is a large-scale dataset containing over 600,000 first-person Counter-Strike 2 gameplay videos with dense per-frame annotations of player inputs and 3D trajectories, designed to train interactive world models. It spans 10,000+ hours of footage paired with synchronized keyboard, mouse, and positional data from professional match replays, and Reka also released an open-source rendering pipeline to extend the dataset.