VirtuaLearn3D++: An open-source AI-based framework for 3D point clouds
In Catallactical, we are happy to announce the alpha release of the open-source VirtuaLearn3D++ framework. This artificial intelligence-based framework automates a wide variety of tasks on 3D point clouds, including point-wise classification, clustering, instance segmentation, geometric feature extraction, cubication, transferring information from the 3D point cloud to the 2D raster, and more.
Feel free to download the code from its GitLab repository and look at its documentation to understand how it works, and perhaps try one of its many examples. Note that the VirtuaLearn3D++ software is delivered with an MIT License, which means you can freely use it for academic and industrial purposes, with the only obligation of recognizing the original authors of the software.
If you are interested in further details, you can have a look at recent papers using the software to approach state-of-the-art problems in 3D point clouds. For example, VirtuaLearn3D++ has been used in a recent work to explore ultra-large-scale semantic segmentation of ALS 3D point clouds: 10.1109/ACCESS.2025.3647154
In this work, the framework has been used to classify 36,369 million points along 29,557 square kilometers for the first time using deep learning models. The study region is whole Galicia from the PNOA-II dataset. The results and materials are available in a Zenodo repository. You can have a look at the image below to see what the technology can do.
