Installation and Environment

Shared Environment Setup

The state-of-the-art completion methods provided in the Third-Party directory can be run within a shared Python environment with CUDA support. The following setup has been tested using an Anaconda platform installed on EndeavourOS:

conda create --name svd python=3.11 -y
conda activate svd

pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/cu128
conda install -c nvidia cuda-toolkit=12.8.0
pip3 install -r requirements.txt

To verify that CUDA support is functioning properly, you can run the following script provided in the Third-Party directory:

python torchtest.py
nvcc --version

Extensions and pretrained models

In order to use a point cloud completion framework, a pretrained model must be obtained and additional extensions must be installed. For detailed instructions, please visit one of the original repositories:

• Building-PCC
• PoinTr
• SeedFormer
• SVDFormer

PCC-HAUS Requirements

PCC-HAUS should be used in the same environment as the preferred point cloud completion framework in order to execute its inference commands. However, its remaining functionalities can be used in any environment that meets the following requirements:

• python >= 3.11
• numpy
• open3d
• scipy