Depth Completion through Colorization

The algorithm has been tested on WSL2 Ubuntu 20.04 using Python 3.8.

Setup conda environment if needed.(With conda environment, the processing is slower.)

conda create --name colorization_depth python=3.8
conda activate colorization_depth

To get necessary dependencies, you may use

bash bash/install_dependency.sh

Test for solid-state LiDAR dataset

To setup the dataset, you may use

mkdir data
bash bash/setup_dataset.sh

For depth completion, change the parameters in ./bash/colorization_ssl.sh, you may use

bash bash/colorization_ssl.sh

Test for KITTI depth completion validation dataset

To setup the KITTI dataset, you may use

mkdir data
bash bash/setup_dataset_kitti.sh

For depth completion, you may use

bash bash/colorization_kitti.sh

To test for colorization with guidance image, change the path for your guidance image. The example use the prediction from colorization with equal weights as guidance, you may use

bash bash/colorization_depth_guidance.sh

The code for the algorithm and evaluation is adapted from open source github repository.

To cite:

@inproceedings{Li2025,
author = {Li, Ying Yin and Herschel, Jan and Abayev, Roman and Paulus, Dietrich and von Gladiss, Anselm},
booktitle = {3D-NordOst 2024},
pages = {31--39},
title = {{Depth Completion by Colorization for Solid-State LiDAR Sensors}},
url = {https://www.gfai.de/fileadmin/Downloads/Tagungsband/gfai-tagungsband-2024.pdf},
year = {2025}
}

@inproceedings{Li2023,
author = {Li, Ying Yin and Herschel, Jan and Theisen, Nick and Abayev, 
Roman and Paulus, Dietrich and von Gladiss, Anselm},
booktitle = {3D-NordOst 2023},
pages = {83--87},
title = {{Evaluation of Depth Completion Using Solid-State LiDAR Dataset}},
url = 
{https://www.gfai.de/fileadmin/Downloads/Tagungsband/gfaI-tagungsband-2023.pdf},
year = {2023}
}