# Installation

## Environment
Tested in environment:
- Conda
- PyTorch 1.4
- CUDA 10.1
- Ubuntu 18.04
- Python 3.7

## Installation steps
0. Creating pytorch environment:
```
conda create -n pvrcnn python=3.7
conda activate pvrcnn
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch
```

1. Installing Pointnet2:
```
git clone https://github.com/sshaoshuai/Pointnet2.PyTorch.git
cd Pointnet2.PyTorch && python setup.py install
export PYTHONPATH=$PYTHONPATH:/path/to/Pointnet2.PyTorch/
```

2. Installing patched spconv:
```
git clone https://github.com/jhultman/spconv.git --recursive
cd spconv && python setup.py bdist_wheel
cd ./dist && pip install *.whl
```

3. Installing torchsearchsorted:
```
git clone https://github.com/aliutkus/torchsearchsorted.git
cd torchsearchsorted && pip install .
```

4. Installing misc python packages:
```
pip install -r requirements.txt
```

5. Installing pvrcnn (this package):
```
python setup.py develop
```

## Common Issues
If the custom ops are not compiled with GPU support, make sure your 
`CUDA_HOME` environment variable is set, and try running `setup.py` with `FORCE_CUDA=1`.