Official source code for the paper "3D Part Segmentation via Geometric Aggregation of 2D Visual Features", WACV 2025. Work by Marco Garosi, Riccardo Tedoldi, Davide Boscaini, Massimiliano Mancini, Nicu Sebe, and Fabio Poiesi.
A detailed list of requirements can be found in requirements.txt. However, some packages require specific installation procedures, therefore the following steps should be followed in order to correctly install the environment.
We suggest using Mamba to create and manage the environment.
mamba create -n pytorch3d python=3.10 -y
mamba activate pytorch3d
pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121
pip install fvcore iopath
pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py310_cu121_pyt210/download.html
pip install torch_geometric
pip install pyg_lib torch_scatter torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-1.13.1+cu116.html --force-reinstall --no-cache-dir
pip install ftfy h5py huggingface-hub imageio matplotlib numba numpy open3d opencv-python packaging pandas pillow platformdirs plotly pyrender safetensors scikit-image scikit-learn scipy tqdm traitlets transformers trimesh umap-learn yacs torchmetrics lightningWe refer to the README.md file under tests/ for detailed instructions on how to run benchmarks. You will need to download the datasets from the respective websites (some might require registration or filling a request form):
- PartNet
- PartNetE
- ScanObject-NN
- FAUST (from SATR)
- ShapeNetPart is downloaded from torch geometric automatically
Lastly, the config.py file shall be edited according to the datasets' location on the file system, so that they can be properly loaded.