Practical overview
When it fits
Use PCN when you need research code for direct partial-to-dense point-cloud completion and can work with its documented local software environment and datasets.
Who it is for
- 3D computer-vision researchers studying point-cloud or shape completion.
- Developers evaluating completion models on ShapeNet or KITTI data.
- Researchers training point-cloud completion models with ShapeNet data.
Documented uses
- Complete missing geometry in a partial 3D point cloud.
- Run a pretrained completion demo on the included example inputs.
- Evaluate completion workflows using ShapeNet or KITTI data.
- Train the model using ShapeNet training and validation data.
- Run the documented KITTI registration experiment after completing point clouds.
Requirements & limits
The code is publicly available on GitHub under the MIT License.
- The documented stack is TensorFlow 1.12, CUDA 9.0 and Python 3.5, with Ubuntu 16.04 reported as the tested environment.
- Open3D, built point-cloud distance operations and downloaded trained models are part of the documented setup.
- ShapeNet training data takes 49 GB for all eight categories or 9 GB for the car category, according to the README.
- KITTI registration is documented as a separate step that depends on first completing the KITTI point clouds.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run the pretrained point-cloud demo
Prepare the documented PCN environment and start its demo on the example input.
Before you start
- Python 3.5, TensorFlow 1.12 and CUDA 9.0; the README reports testing on Ubuntu 16.04.
- Install repository dependencies and Open3D for point-cloud I/O.
- Build the point-cloud distance operations in pc_distance, with makefile paths set correctly.
- Download the trained models.
Steps
Install the Python dependencies using the repository requirements file.
Install Open3D and build the point-cloud distance operations.
Download the trained models.
Run python3 demo.py; use --input_path to switch among examples in demo_data.
Expected outcome
The repository demo runs with a selected example input to show PCN point-cloud completion.
Guide limitations
- The README does not specify additional input formats or deployment as a hosted service.
Evaluate completion on ShapeNet test data
Run the repository's ShapeNet completion evaluation with the required test splits.
Before you start
- The documented PCN software setup: Python 3.5, TensorFlow 1.12, CUDA 9.0, Open3D, built point-cloud distance operations and trained models.
- ShapeNet test data in the shapenet folder, including test, test_novel, test.list and test_novel.list.
Steps
Download the required ShapeNet test data and trained models from the linked project data location.
Run python3 test_shapenet.py.
Use --model_type to choose a model architecture; consult python3 test_shapenet.py -h for available options.
Expected outcome
The repository's ShapeNet test workflow evaluates point-cloud completion using the selected model architecture.
Guide limitations
- The instructions cover evaluation on the named test splits; they do not describe a general-purpose data import workflow.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- wentaoyuan/pcn · Official homepage title · retained in its source language