Mechanical Part Point Cloud Completion Dataset

This dataset consists of 2,000 mechanical part samples collected using high-precision laser scanning. It is designed for point cloud completion tasks and can also support multi-modal learning if applicable.Dataset StructureThe dataset is organized into three subsets with the following split:<b>Training</b>: 85%<b>Validation</b>: 5%<b>Testing</b>: 10%Each subset contains three folders:<pre><pre>machine_part_point_cloud_dataset/<br>├── train/<br>│ ├── complete/ # Ground truth complete point clouds (.npy)<br>│ ├── incomplete/ # Incomplete point clouds (.npy)<br>│ └── rgb/ # Contour/rgb images (.npg)<br>├── val/<br>│ ├── complete/<br>│ ├── incomplete/<br>│ └── rgb/<br>└── test/<br> ├── complete/<br> ├── incomplete/<br> └── rgb/<br></pre></pre>Data Format<code>complete/</code>: Contains the complete point cloud of each sample in <code>.npy</code> format. Each file stores a NumPy array representing the full geometry of a mechanical part.<code>incomplete/</code>: Contains partial point clouds obtained from real-world occlusion and viewpoint limitations. Also in <code>.npy</code> format.<code>rgb/</code>: Contains <code>.npg</code> format files representing the 2D contour or appearance information of each part. These can serve as auxiliary input for multi-modal learning frameworks.ApplicationsThis dataset is suitable for tasks such as:3D point cloud completionMulti-modal 3D reconstructionLearning shape priors for mechanical partsPlease ensure your model correctly handles <code>.npy</code> and optionally <code>.npg</code> formats if using multi-modal inputs.

本数据集包含2000个通过高精度激光扫描采集的机械零件样本，专为点云补全（point cloud completion）任务设计，同时也可支持多模态学习（multi-modal learning）任务。

### 数据集结构
本数据集划分为三个子集，划分比例如下：
- 训练集（Training）：85%
- 验证集（Validation）：5%
- 测试集（Testing）：10%

每个子集均包含三个文件夹，目录结构如下：

machine_part_point_cloud_dataset/
├── train/
│ ├── complete/ # 完整点云真值（.npy格式）
│ ├── incomplete/ # 不完整点云（.npy格式）
│ └── rgb/ # 轮廓/RGB图像（.npg格式）
├── val/
│ ├── complete/
│ ├── incomplete/
│ └── rgb/
└── test/
├── complete/
├── incomplete/
└── rgb/

### 数据格式
- `complete/`：存储每个样本的完整点云，格式为.npy。每个文件均为一个NumPy数组，表征机械零件的完整几何形态。
- `incomplete/`：存储由真实场景下的遮挡与视角限制得到的局部点云，同样采用.npy格式。
- `rgb/`：存储格式为.npg的文件，用于表征每个零件的二维轮廓或外观信息，可作为多模态学习框架的辅助输入。

### 应用场景
本数据集适用于以下任务：
1. 三维点云补全
2. 多模态三维重建
3. 学习机械零件的形状先验知识

请确保模型能够正确处理.npy格式数据，若使用多模态输入则需额外支持.npg格式。