Software and applications

Rail3D: Multi-Context Point Cloud Dataset and New Approach for Railways Semantic Understanding

This is the first multi-context point cloud dataset called Rail3D. This dataset covers three countries: Hungary, France, and Belgium, with a total length of almost 5.8 kilometers and approximately 288 million points. It covers the 9 most relevant classes for railway applications: Ground, Vegetation, Rail, Poles, Wires, Signalling, Fences, Installations, and Buildings. We conducted a generic classification with 9 universal classes (Ground, Vegetation, Rail, Poles, Wires, Signaling, Fence, Installation and Building) and evaluated the performance of three state-of-the-art models: KPConv (Kernel Point Convolution), LightGBM, and Random Forest. The best performing model, a fine-tuned KPConv, achieved a mean Intersection over Union (mIoU) of 86%.

Link to Rail3D dataset

Automatic instance segmentation of LiDAR data using Segment-Anything Model (SAM)

This package is specifically designed for unsupervised instance segmentation of LiDAR data. It brings together the power of the Segment-Anything Model (SAM) developed by Meta Research and the segment-geospatial package from Open Geospatial Solutions to automatize instance segmentation of 3D point cloud data.

Link to segment-lidar code

Optim3D: Efficient and scalable generation of large-scale 3D building

Optim3D is a powerful tool for efficient and scalable generation of highly detailed and large-scale 3D building models. The modeling process is based on GeoFlow. The tool focuses mainly on preparing data for efficient reconstruction through indexing, tiling and parallel computing, which significantly reduces the processing time and resources required to generate large-scale 3D building models.

Link to Optim3D code

A light and compact CityJSON management tool

This tool is built as a high-level MERN application. It therefore allows developing with a single language in both server and client side. It does not rely on an operating system layer.

3D city models are structured following the CityJSON encoding, a JSON-based encoding for storing 3D city models, also called digital maquettes or digital twins. Thanks to MongooseJS, inserted models are validated in regard of CityJSON 1.0.1 specifications. These specifications define the city model schema used to structure the database (several modifications were however made to suit to document-oriented databases).

Link to Measur3D code