dream-cubed/DreamCubedNatural

Dream-Cubed Natural 是一个程序生成的 Minecraft Java Edition v1.12.2 地形数据集，以32x32x32体素块的形式表示。该数据集旨在用于可控3D生成、生物群系条件体素建模、修复、扩展和合成环境生成的研究。此仓库仅包含Dream-Cubed的自然/程序生成部分，包括从Minecraft世界中提取的原始块以及用于训练的处理后类条件数据集。数据集涵盖多种生物群系标签：beaches、birch_forest、cave、desert、extreme_hills、forest、ice、jungle、ocean、plains、river、savanna、swampland、taiga和village。原始块存储为整数Minecraft块ID的NumPy数组，而处理后的数据集包括映射文件、训练和验证分割、内存映射体素数组以及标签数组。数据收集通过自定义gRPC提取API从本地运行的Minecraft服务器完成，仅包含程序生成内容，不涉及玩家创建的地图或标识符。数据集受CC BY-NC 4.0许可证和Minecraft EULA约束，反映了v1.12.2版本的生成特性，但存在块间连续性有限和生物群系分布偏差等限制。Dream-Cubed Natural is a dataset of procedurally generated Minecraft Java Edition v1.12.2 terrain represented as 32x32x32 voxel chunks. It is intended for research on controllable 3D generation, biome-conditioned voxel modeling, inpainting, outpainting, and synthetic environment generation. This repository contains only the natural/procedural portion of Dream-Cubed. It includes raw chunks extracted from Minecraft worlds as well as processed class-conditional datasets used for training. The raw natural data covers biome labels: beaches, birch_forest, cave, desert, extreme_hills, forest, ice, jungle, ocean, plains, river, savanna, swampland, taiga, and village. Raw chunks are stored as NumPy arrays of integer Minecraft block IDs, while processed datasets include mapping files, training/validation splits, memory-mapped voxel arrays, and label arrays. Data was collected via a custom gRPC extraction API from locally run Minecraft servers, containing only procedurally generated content without player-created maps or identifiers. The dataset is released under CC BY-NC 4.0 license and subject to the Minecraft EULA, reflecting v1.12.2 generation but with limitations such as limited inter-chunk continuity and biome distribution biases.