Create a conda environment and install dependencies:
git clone https://github.com/ZiyuGuo99/Point-Bind_Point-LLM.git
cd Point-Bind_Point-LLM
conda create -n pointbind python=3.8
conda activate pointbind
# Install the according versions of torch and torchvision
conda install pytorch torchvision cudatoolkit
pip install -r requirements.txt
Install GPU-related packages:
# PointNet++
pip install "git+https://github.com/erikwijmans/Pointnet2_PyTorch.git#egg=pointnet2_ops&subdirectory=pointnet2_ops_lib"
# GPU kNN
pip install --upgrade https://github.com/unlimblue/KNN_CUDA/releases/download/0.2/KNN_CUDA-0.2-py3-none-any.whl
We provide the pre-trained weights of Point-Bind with I2P-MAE and Point-BERT as the 3D encoders. Normally, Point-Bind with I2P-MAE performs better. Please create a /ckpts
folder and organize the downloaded files in the following structure
Point-Bind_Point-LLM/
├── ckpts
├── pointbind_i2pmae.pt
└── pointbind_pointbert.pt