Pointcloud Quality Assessment

This is the official repository for our proposed PCQA method, which can be integrated into industrial 3D automatic scanning, providing detailed quality scores for acquired point clouds.

Paper link: here

Principle

A subset of the scanned points is mapped onto a 2D distribution based on designated distance and angular values. The quality of the point cloud is then estimated from this 2D distribution. The calculation of the quality score for the points is presented as follows:

Run a demo

We recommend using Anaconda to create the environment. You can build the environment by:

conda create -n "PCQA" python=3.8 ipython
conda activate PCQA

pip install -r requirements.txt
# you can install torch based on your CUDA version
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

We would like to thank DeepFit. Some parts of our implementation are based on their excellent work.

1. Run

   You can jump into the ./src folder, and run

   python n-jet_fitting.py

   where the default order ($jet_order_fit) for n-jet is 3, and K size ($points_per_patch) is 256. A fitting result file is outputed in ./src/data folder named "sample_order3_normal_beta.txt".

2. Calculate the Quality Score

   python pcqa_demo.py -tn 14 -sn 8

   where -tn denotes thread_number and -sn is skip_number. The smaller the skip_number, the slower the algorithm runs.

3. [optional] Calculate the Quality Score in Jupyter Notebook

   Install the kernel for Jupyter notebook

   ipython kernel install --name "PCQA" --user
   jupyter notebook

   You can open QualityScoreEstimation.ipynb to have a better understanding.

4. Visualization

   A quality score file can be found in folder ./src/data/sample_quality_score_ours.txt. You can use code in ./src/quality2color.py to covert it to points with colors.

   Additionally, the software CloudCompare can be used to visualize it, as shown below

   

Citation & Contact

If you find this work helpful, please consider citing us

@article{LI2025102863,
title = {A robust assessment method of point cloud quality for enhancing 3D robotic scanning},
journal = {Robotics and Computer-Integrated Manufacturing},
volume = {92},
pages = {102863},
year = {2025},
issn = {0736-5845},
author = {Leihui Li and Xuping Zhang},
}

For any questions about this project, feel free to contact me at leihui@mpe.au.dk or open an issue in this repository.