ADSKAILab/Zero-To-CAD-1m
收藏资源简介:
Zero-to-CAD 1M是一个大规模的数据集,包含1,000,000个可执行的CAD构造序列,这些序列是由大型语言模型(LLM)在反馈驱动的CAD环境中生成的。每个样本都是可执行的CadQuery Python代码,具有可解释性,并且经过几何验证。数据集涵盖了广泛的CAD操作,包括布尔运算、倒角、倒圆角、壳、放样、扫描和模式等。此外,数据集还提供了预计算的DINOv3视觉嵌入和FAISS IVF-PQ索引,用于最近邻搜索。数据集适用于训练CAD序列模型、图像到CAD重建、CAD程序理解和基准测试等用途。
Zero-to-CAD 1M is a large-scale dataset comprising 1,000,000 executable CAD construction sequences generated by large language models (LLMs) in a feedback-driven CAD environment. Each sample consists of executable CadQuery Python code that is interpretable and geometrically validated. The dataset covers a wide range of CAD operations, including Boolean operations, chamfering, filleting, shelling, lofting, sweeping, and pattern operations, among others. Additionally, the dataset provides pre-computed DINOv3 visual embeddings and FAISS IVF-PQ indexes for nearest neighbor search. This dataset is suitable for applications such as training CAD sequence models, image-to-CAD reconstruction, CAD program understanding, and benchmarking.




