Demo
R E S O U R C E S
OFFICIAL DOCUMENTATION

VL3D++ technical documentation

The documentation is the main source to understand the framework, its components, pipelines, examples and paradigms (Deep Learning 3D and Active Learning).

VIEW OFFICIAL DOCUMENTATION
Introduction to the framework
Pipelines configurable via JSON
Components and modules
Reproducible examples by domain (geographic, infrastructure, forestry, medical)
Evaluation, metrics, uncertainty and visualisation
REPOSITORY AND EXECUTION

Source code and usage guide (GitLab)

The repository contains the full framework, JSON configuration, installation, tests, examples and GPU considerations.

VIEW REPOSITORY AND EXECUTION
Installation (Ubuntu 24 + Python 3.12 as main environment)
Alternatives for Windows/macOS/Linux via requirements or conda
Test execution (python vl3d.py --test)
Pipeline execution (python vl3d.py --pipeline ..json)
Notes on git-lfs for test data
GPU recommendations (TensorFlow/Keras + CUDA/cuDNN)
SCIENTIFIC PUBLICATIONS

Peer-reviewed articles and methodological foundations that support the viability and technology of the VL3D++ framework

Deep Learning for Ultra-Large-Scale Semantic Segmentation of Geographic 3D Point Clouds With Missing Labels
Deep learning with simulated laser scanning data for 3D point cloud classification
Methodology for Identifying Optimal Pedestrian Paths in an Urban Environment

Datasets and open data used in use cases

Hessigheim

Point clouds for geographic and topographic classification.

SUBSCRIBE HESSIGHEIM

IntrA Dataset

3D medical point clouds for aneurysm detection.

ORIGINAL PUBLICATION

Teeth3DS Dataset

Segmentation of teeth and jaw in clinical settings.

PROJECT PAGE

Pielach 2024

Topo-bathymetric LiDAR data of river ecosystems (Active Learning).

PIELACH 2024

DALES Dataset

Large-scale classification of urban/geographic scenarios.

DALES DATASET

PNOA2

Classified ALS point clouds covering the entire territory.

PNOA2
REPRODUCIBLE EXAMPLES (BY DOMAIN)

See the framework working in different scenarios

The documentation includes runnable examples showing the complete pipeline:
raw cloud → prediction → errors/uncertainty → metrics.

VIEW EXAMPLES
KEY CONCEPTS TO UNDERSTAND THE APPROACH

Deep Learning 3D

VL3D++ works directly on 3D geometry: local neighbourhoods, receptive fields, local prediction and global reconstruction.

LEARN MORE

Active Learning

The framework enables cycles where the model identifies areas of uncertainty and the expert only corrects where it adds value.

LEARN MORE
RECOMMENDED EDUCATIONAL RESOURCES

To understand neural networks visually

If you need a well-made introductory explanation to understand the concept of a neural network (without diving into papers), this resource is excellent:

WHAT IS A NEURAL NETWORK?

Licences and attribution

The documentation of VirtualLearn3D++ is published under the CC BY 4.0 licence (attribution required).
The VL3D++ framework is published under the MIT licence.
On the project website, any external figure or resource must keep its corresponding attribution.