GMP_TTA

Enhancing Semantic Segmentation of LiDAR Point Clouds through Global Maps

This is a code repository of my Thesis work.

You can find all .py files and notebooks here.

How to

I hade following workdir structure:

.
└── workdir/
    ├── dataset/
    │   └── sequences/
    │       └── ...
    ├── gits/
    │   └── ...
    ├── utils/
    │   ├── one.py
    │   └── two.py
    ├── .py files
    └── .ipynb files
    └── Pipfile

Then I created pipenv env and got provided Pipfile .

pip install --user pipenv
pipenv --python 3.10
pipenv install

Files in utils folder - are modified files from Semantic Kitti API repo.

Then you can created parts of map with following command:

pipenv run python map_creators.py 08 0 1000
pipenv run python map_creators.py 08 500 1500
...

And with depth-weighted samplaing, one can crate GMP:

pipenv run python velodyne2_creator.py 08 0 1000 2
pipenv run python velodyne2_creator.py 08 500 1500 2
...

And with uniform: And with depth-weighted samplaing, one can crate GMP:

pipenv run python velodyne2_creator.py 08 0 1000 3
pipenv run python velodyne2_creator.py 08 500 1500 3
...

Notebooks were primarly used as playgrounds.

And with visualize.py script one can view open3d windows with point clouds.