INSTALLATION.md

Installation Guide 🔧

1. Create and activate conda environment

conda create -n UniPre3D python=3.11
conda activate UniPre3D

2. Install PyTorch and dependencies

# Install PyTorch with CUDA support
pip install torch==2.2.2 torchvision==0.17.2

# Install project dependencies
pip install -r requirements.txt

# Install flash-attn for efficient attention mechanisms
pip install flash-attn --no-build-isolation

3. Install C++ extensions

# Install PointNet++ modules
cd openpoints/cpp/pointnet2_batch
python setup.py install
cd ../

# Install Chamfer Distance and emd modules
cd chamfer_dist
python setup.py install --user
cd ../emd
python setup.py install --user
cd ../../../

4. Install Mamba3D dependencies

# Install PointNet2 operations library
pip install "git+https://github.com/erikwijmans/Pointnet2_PyTorch.git#egg=pointnet2_ops&subdirectory=pointnet2_ops_lib"

# Install GPU KNN
pip install --upgrade https://github.com/unlimblue/KNN_CUDA/releases/download/0.2/KNN_CUDA-0.2-py3-none-any.whl

# Install Mamba SSM dependencies
pip install causal-conv1d==1.2.2.post1
pip install mamba-ssm==1.2.2

causal-conv1d and mamba-ssm are required for the Mamba3D model, you should select the version that matches your CUDA and pytorch version.

5. Install Gaussian Splatting Renderer

The Gaussian Splatting renderer is required for rendering Gaussian Point clouds to images.

# Clone the repository
git clone https://github.com/graphdeco-inria/gaussian-splatting.git --recursive
cd gaussian-splatting

# Install the renderer
pip install submodules/diff-gaussian-rasterization

6. Download pre-trained image feature extractor

Please download the pre-trained image feature extractor diffusion_pytorch_model.bin from here and put it in the weights folder.

Troubleshooting

• If you encounter issues installing PointNet2 operations, please refer to this solution for manual installation steps.
• For Gaussian Splatting, ensure your system meets the hardware requirements.