0x3/Zero-To-CAD-1m

Zero-to-CAD 1M是一个大规模数据集，包含1,000,000个可执行的、可解释的CAD构造序列，完全无需真实世界数据合成生成。该数据集由大型语言模型（LLM）在反馈驱动的CAD环境内操作生成，覆盖广泛的CAD操作词汇，包括布尔运算、圆角、倒角、抽壳、放样、扫描和阵列等，超越了传统的草图-拉伸工作流。每个样本都是有效的CadQuery Python代码，可运行并生成几何体，具有命名参数、逻辑构造顺序和人类可读代码，并经过多阶段几何验证（拓扑、连续性、复杂性、导出）。数据生成过程被构建为智能体搜索问题，LLM在CadQuery执行环境中循环使用工具进行迭代生成、执行和修复，直到验证成功。数据集包含训练集（979,633个样本）、验证集（10,000个样本）和测试集（10,000个样本），每个样本包含唯一标识符、CadQuery源代码、面数、操作列表、8个渲染视图、STL和STEP文件等字段。该数据集适用于训练CAD序列模型、从图像到CAD的重建、CAD程序理解和生成CAD模型的基准测试。

Zero-to-CAD 1M is a large-scale dataset of 1,000,000 executable, interpretable CAD construction sequences synthesized entirely without real-world data. It is generated by an LLM operating inside a feedback-driven CAD environment, covering a broad vocabulary of CAD operations including booleans, fillets, chamfers, shells, lofts, sweeps, and patterns, going beyond traditional sketch-and-extrude workflows. Each sample is executable CadQuery Python code that runs and produces geometry, with named parameters, logical construction order, human-readable code, and multi-stage geometric validation (topology, contiguity, complexity, export). The generation process is framed as an agentic search problem, where an LLM iteratively generates, executes, and repairs code in a loop with a CadQuery execution environment until validation succeeds. The dataset includes train (979,633 samples), validation (10,000 samples), and test (10,000 samples) splits, with each sample containing fields such as unique identifier, CadQuery source code, face count, operation list, 8 rendered views, STL, and STEP files. It is intended for training CAD sequence models, image-to-CAD reconstruction, CAD program understanding, and benchmarking generative CAD models.