LidarAug

A toolbox for LiDAR point cloud data, providing point cloud transformations, point cloud augmentation, realistic weather simulation and 2D & 3D AP evaluation, with an easy-to-use Python API. This module supports several essential tasks for the development of LiDAR-based perception methods in automated driving.

Installation

First clone and enter the repository:

git clone https://github.com/ekut-es/LidarAug && cd LidarAug

C++ library

The following dependencies are necessary to build and test the C++ library for development:

• libtorch
• google test
• boost
• OpenMP

It is also necessary to set the environment variable TORCH_PATH to point to where libtorch is installed on your system.

After that, just run make ctest to compile the library and run google test.

Note that the tests written for the backend include some controlled RNG tests which might fail on different platforms with different architectures such as the Apple MX chips. The tests were developed for Linux x86 using GCC.

Python module

The following dependencies are necessary to install the Python module:

• PyTorch/libtorch
• boost
• OpenMP
• cnpy
• pybind11

To use the Python module, just run make install after cloning and entering the repository.

To test the python functions/wrappers, install pytest (pip install pytest) and run make testpy.

The required Python version is 3.11.

Submodules

The lidar_aug Python module contains 5 submodules:

1. transformations:

transformations contains any C++ enums, structs and functions that have bindings and are used for transformations.

2. weather_simulations:

weather_simulations contains any C++ enums, structs and functions that have bindings and are used for weather simulations.

3. augmentations:

augmentations contains the Python wrappers for any C++ function (weather simulation or transformation).

4. evaluation:

evaluation contains (C++) function to evaluate the accuracy of bounding boxes. This can be done for 2D and 3D bounding boxes.

5. point_cloud:

point_cloud contains things that is specific to point clouds that is used across modules and functionally not specific to the task of one of those. Such as the IntensityRange enum that is used to set/determine the maximum intensity of the points in a point cloud.

Docker

Alternatively the module can be run inside a Docker container.

After installing Docker and cloning the repository, all you need to do is run make docker, which will start building the image and automatically run the tests during the build process.

NOTE: If you're running the docker image on ARM run make docker-arm instead.