Aerial LiDAR Classifier
Classify airborne and mobile mapping LiDAR point clouds with deep learning (LitePT-L, SegFormer 3D).
Classifies airborne and mobile mapping LiDAR point clouds
(LAS / LAZ / COPC) with deep-learning models. For mobile mapping,
LitePT-L MLS is a point transformer using 5 cm voxels.
It labels ground, low and high vegetation, buildings,
pole-like objects, vehicles, fences and barriers, wires and other
objects, and each class code can be edited. For airborne LiDAR,
LitePT-L at 10 cm (default on NVIDIA GPUs) and the
UrbanFiltering 3D SegFormer from NRCan's TreeAIBox (GPU or CPU) write
standard ASPRS codes. The LitePT-L models need an NVIDIA CUDA GPU.
External dependencies: PyTorch, laspy, lazrs, timm, scipy and, on
NVIDIA GPUs, spconv. A one-click setup installs them into a separate
per-user folder on first use (no admin rights, nothing to
configure), and the model weights download automatically. Files in
feet are converted from their CRS; tiling and streaming handle files
larger than RAM. Model weights are CC BY-NC 4.0. Documentation:
https://github.com/akharroubi/AerialLidarClassifier
The plugin author teaches a live cohort, LiDAR Point Clouds
Processing in QGIS: https://maven.com/geomatics/qgis3d?utm_source=plugins.qgis.org&utm_medium=listing&utm_campaign=qgis3d