Skip to content

Semantic segmentation deep learning for aerial, high density Lidar point clouds.

License

Notifications You must be signed in to change notification settings

MichelDaab/lidar-deep-segmentation_CICD

 
 

Repository files navigation

Lidar Deep Segmentation

PyTorch Lightning Config: Hydra

Documentation Build



Aerial Lidar HD Semantic Segmentation with Deep Learning

Context

The French Lidar HD project ambitions to map France in 3D using 10 pulse/m² aerial Lidar. The data will be openly available, including a semantic segmentation with a minimal number of classes: ground, vegetation, buildings, vehicles, bridges, others.

Considering the scale of this task, deep learning is leveraged to speed-up the production and give tools for quality control

Content

Lidar-Deep-Segmentation is a deep learning library designed with a focused scope: the multiclass semantic segmentation of large scale, high density aerial Lidar points cloud.

The library implements the training of 3D Segmentation neural networks, with optimized data-processing and evaluation logics at fit time. Inference on unseen, large scale point cloud is also supported. It allows for the evaluation of single-class IoU on the full point cloud, which results in reliable model evaluation.

Although the library can be easily extended with new neural network architectures or new data signatures, the library makes some opiniated choices in terms of neural network architecture, data processing logics, and inference logic.

Additionnaly, two data signatures are supported:

  • French Lidar HD, produced by the French geographical Institute. The data is colorized with both RGB and Infrared. Therefore, data processing will include Infrared channel as well as NDVI.
  • Swiss Lidar from SwissSurface3D (en), a similar initiative from the Swiss geographical institute SwissTopo. The data comes from the SwissSurface3D Lidar database and is not colorized, so we have to join it with SwissImage10 orthoimages database. The procedure is described in this standalone repository.

Lidar-Deep-Segmentation is built upon PyTorch. It keeps the standard data format from Pytorch-Geometric. Its structure was bootstraped from this code template, which heavily relies on Hydra and Pytorch-Lightning to enable flexible and rapid iterations of deep learning experiments.

-> For installation and usage, please refer to full Documentation.

About

Semantic segmentation deep learning for aerial, high density Lidar point clouds.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 69.0%
  • Jupyter Notebook 30.1%
  • Shell 0.9%