google/MapTrace

MapTrace是一个包含200万个样本的合成数据集，用于地图上的路径追踪任务训练。数据集分为两部分：maptrace_parquet包含在宣传册、公园指南或购物中心等中发现的更复杂、风格化的地图路径；floormap_parquet包含在办公楼、公寓综合体或校园地图中常见的更简单、结构化的楼层地图。每部分都包括地图图像的字节、定义目标路径的坐标列表字符串、要求模型找到指定路径的自然语言问题提示以及用于生成合成图像的地图文本描述。

MapTrace is a synthetic dataset containing 2M samples for path tracing tasks on maps. The dataset is split into two parts: maptrace_parquet, which contains paths on more complex, stylized maps such as those found in brochures, park directories, or shopping malls; and floormap_parquet, which contains paths on simpler, structured floor maps typical of office buildings, apartment complexes, or campus maps. Each part includes the bytes of the map image, a string representation of a list of coordinates defining the target path, a natural language question prompt asking the model to find the specified path, and a natural language description of the map image used to create the synthetic image.