Installation

Pre-requisites

• Make sure CUDA is installed
• Install pytorch with support for the installed CUDA version

Requirements

pip install -r requirements.txt

Docker (Linux only)

Alternatively you can use Docker to run the code. Build an image with the provided Dockerfile:

docker build -t easy_digi_twin .

Make sure that Nvidia Container Toolkit is installed and used as the default runtime.

Preprocessing

Transform Recording Points

Modify transform_recording_points.py to set the following parameters:

dir_path = "/path/to/data/dir"
run_id = 0
output_path = "/path/to/recording_points.gpkg"
csv_path = "/path/to/gnss.csv"

Create Test Images

Modify create_test_images.py to set the following parameters:

dir_path = "/path/to/data/dir"
output_dir_path = "/path/to/test/output/dir"
recording_points_path = "/path/to/recording_points.gpkg"

Preprocess Data

Modify preprocess.py to at least set the following parameters:

dir_path = "/path/to/data/dir"
output_dir_path = "/path/to/output/dir"
recording_points_path = "/path/to/recording_points.gpkg"
vmu_mask_path = "/path/to/VMU_Mask.png"
hmu_mask_path = "/path/to/HMU_Mask.png"

Training

Modify train.py to at least set the following parameters:

save_dir = "/path/to/save/dir/"
scene_name = "scene_name"
dataset_path = "/path/to/dataset/"
run_name = "run_name"

The rest of the file should be modified to adjust the optimization process.

Georeference Scene

Modify georeference_model.py to set the following parameters:

scene_path = "/path/to/saved/scene"
reference_file_path = "/path/to/geo_reference.txt"

GUI

1. Run gui.py
2. Select General -> Load Scene from the menu
3. Select .ply or scene_model.json to load

LICENSE

All code in this repository is licensed under the GPL-3.0 License. Installed requirements are listed in requirements.txt and are licensed under their respective licenses.

NOTE: The installed dependency EasyDigiTwin-gaussian-rasterization is licensed under the Gaussian-Splatting License which prohibits commercial use. The SegFormer Model used for preprocessing is licensed under the NVIDIA Source Code License for SegFormer which also prohibits commercial use. Therefore, the code in this repository can not be run for commercial use without replacing these dependencies.