---
license: cc-by-4.0
---
# CAD Part Recommendation Dataset

This dataset is designed for **CAD part recommendation**.  

- **assemblies.zip**  
  Contains CAD assemblies across seven categories, with each category comprising at least 50 assemblies.  

- **assemblies_3-15.zip**  
  Includes assemblies from **assemblies.zip** whose number of constituent parts ranges from 3 to 15. Each assembly not only contains the assembly itself but also its **constituent parts** along with their corresponding **graph structures**.  

We represent each assembly as a **graph**, where **vertices** correspond to the constituent parts and **edges** indicate the contact relationships between parts, as illustrated in the figure below.

![70e404e4-c6d7-44b4-9d70-cb19662da757](https://cdn-uploads.huggingface.co/production/uploads/65eaf8bf684b5d61ab57f131/iN8E9wXlWNGTeyRxTT9N7.png)
- **part_library.zip**  
  Contains all the parts included in **assemblies_3-15.zip**.  

- **Partial Assemblies and Labels Generation**  
  We employ **Algorithm 2** to generate **partial assemblies** and their corresponding **labels** from the assemblies in **assemblies_3-15.zip**. These are used for the **CAD part recommendation** task.

![375bf7df-177c-4956-8f01-c727e37ce301](https://cdn-uploads.huggingface.co/production/uploads/65eaf8bf684b5d61ab57f131/UXx5CmtqiiiWJTsN57exA.png)
![26291f53-49d0-4667-b2e2-a1b54ea00352](https://cdn-uploads.huggingface.co/production/uploads/65eaf8bf684b5d61ab57f131/YO6OuTpWY2wplgByc0COi.png)

- **Dataset Splits**  
  The files **train_dataset.json**, **test_dataset.json**, and **val_dataset.json** store the **training**, **testing**, and **validation** sets, respectively.  

  Each of these files contains the **graph structures** corresponding to the **partial assemblies**.  

  Using the **graph structures** together with the assemblies in **assemblies_3-15.zip** and the parts in the**part_library.zip**, one can easily reconstruct the **CAD models** of the partial assemblies along with their corresponding **labels**.  

  Therefore, we **do not** include the CAD models of the partial assemblies in the dataset, as doing so would result in an excessively **large dataset**.