calicio/StarCraftMotion_deprecated

StarCraftMotion是一个基于《星际争霸II》回放构建的多智能体运动预测基准数据集。每个样本是一个固定长度的场景窗口（145帧，16 FPS，约9秒），包含所有单位状态、动态地图层和每个玩家的经济时间序列。发布的数据集分割是**对抗性的**：场景被二次采样以过度加权交互密集的窗口（相互可见性和玩家间转换），使其成为部分可观测性下多智能体预测的压力测试。数据集包含469,187个场景，分为训练集（362,075）、验证集（45,121）和测试集（61,991）。数据来源于64,327个回放级别的HDF5文件（Blizzard `3.16.1-Pack_1-fix`）。数据集旨在用于多智能体轨迹/运动预测、部分可观测性下的基准测试以及《星际争霸II》单位运动的表示学习。

StarCraftMotion is a multi-agent motion-forecasting benchmark built from StarCraft II replays. Each example is a fixed-length scenario window (`145` frames at `16 FPS`, ~9 seconds) containing all unit states, dynamic map layers, and per-player economy time series. The released split is **adversarial**: scenarios are subsampled to overweight interaction-heavy windows (mutual-visibility and inter-player transitions), making it a stress test for multi-agent prediction under partial observability. The dataset includes 469,187 scenarios, split into training (362,075), validation (45,121), and test (61,991) sets. It is sourced from 64,327 replay-level HDF5 files (Blizzard `3.16.1-Pack_1-fix`). The dataset is intended for multi-agent trajectory / motion forecasting, benchmarks for fog-of-war handling, and representation learning over StarCraft II unit motion.