This is a Python port of the Depth Clustering algorithm originally written in C++. The original GitHub repository can be found at:
While we strive to follow the original implementation closely, there are some differences in our version. Please note that the current implementation lacks some features of the original code. We will consider porting those if there are any requests.
If you are comfortable with C++, we recommend using the original implementation.
Prerequisites:
- Python >= 3.8
- numpy >= 1.23.1
- numba >= 0.56.2
- opencv-python >= 4.8.0
You can install this depth_clustering library with the command below:
$ pip install git+https://github.com/uchiyamalab/depth_clustering_py.git@main
Please refer to the /notebooks/examples.ipynb and /notebooks/ground-remover.ipynb.
The author worked at a new media art lab and learned about Depth Clustering while working on 3D LiDAR projects. Unfortunately, we needed to run the algorithm on multiple student computers with different environments (including M1 Mac, Windows, and Raspberry Pi), which required much effort to prepare the C++ build environments. As a solution, we ported the algorithm to Python. While Python code is generally much slower than C++ code, we found that using Numba, a just-in-time (JIT) compiler based on LLVM, made the code relatively fast.
Please cite the original author's papers if you use this code.
MIT License
Copyright (C) 2023 T. Kamatani
Copyright (C) 2020 I. Bogoslavskyi, C. Stachniss
Ported to Python by Takahiro Kamatani in Uchiyama Lab, Institute of Art and Design, University of Tsukuba.