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:
Training: 85%Validation: 5%Testing: 10%Each subset contains three folders:
machine_part_point_cloud_dataset/
├── train/
│ ├── complete/ # Ground truth complete point clouds (.npy)
│ ├── incomplete/ # Incomplete point clouds (.npy)
│ └── rgb/ # Contour/rgb images (.npg)
├── val/
│ ├── complete/
│ ├── incomplete/
│ └── rgb/
└── test/
├── complete/
├── incomplete/
└── rgb/
Data Formatcomplete/: Contains the complete point cloud of each sample in .npy format. Each file stores a NumPy array representing the full geometry of a mechanical part.incomplete/: Contains partial point clouds obtained from real-world occlusion and viewpoint limitations. Also in .npy format.rgb/: Contains .npg 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 .npy and optionally .npg formats if using multi-modal inputs.