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个经高精度激光扫描采集的机械零件样本，专为点云补全任务设计，同时可在适用场景下支持多模态学习。数据集结构 本数据集划分为三个子集，划分比例如下：<b>训练集</b>：85%<b>验证集</b>：5%<b>测试集</b>：10%。各子集均包含三个文件夹，目录结构如下：<pre>机械零件点云数据集/<br>├── train/<br>│ ├── complete/ # 真值完整点云（.npy格式）<br>│ ├── incomplete/ # 残缺点云（.npy格式）<br>│ └── rgb/ # 轮廓/RGB图像（.npg格式）<br>├── val/<br>│ ├── complete/<br>│ ├── incomplete/<br>│ └── rgb/<br>└── test/<br> ├── complete/<br> ├── incomplete/<br> └── rgb/<br></pre>数据格式 <code>complete/</code>：存储各样本的完整点云，格式为.npy，每个文件均为表征机械零件完整几何形态的NumPy数组。<code>incomplete/</code>：存储由真实场景遮挡与视角限制得到的残缺点云，格式同样为.npy。<code>rgb/</code>：存储格式为.npg的文件，用于表征各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：3D点云补全、多模态三维重建、机械零件形状先验学习。请确保你的模型能够正确处理.npy格式数据，若使用多模态输入则需额外支持.npg格式。