INSTALL.md

Step-by-Step Installation

This codebase is tested with python=3.10, torch==1.11.0 and torchvision==0.12.0, CUDA 11.3.

Step 1: Create Enviroment

conda create -n annotator -y python=3.10
conda activate annotator

Step 2: Install Packages

conda install pytorch==1.11.0 torchvision==0.12.0 cudatoolkit=11.3 -c pytorch -c conda-forge

conda install google-sparsehash -c bioconda

pip install pyyaml easydict numba wandb setproctitle prettytable sharedarray tqdm

Step 3: Install Necessary Libraries

3.1 - TorchSparse

Note: The following steps are required in order to use the voxel and fusion backbones in this codebase.

• Make a directory named torchsparse_dir

cd package/
mkdir torchsparse_dir/

• Unzip the .zip files in package/

unzip sparsehash.zip
unzip torchsparse.zip

• Setup sparsehash (Note that ${ROOT} should be your home path to the Annotator folder)

cd sparsehash/
./configure --prefix=/${ROOT}/Annotator/package/torchsparse_dir/sphash/

make

make install

• Compile torchsparse

cd ..
pip install ./torchsparse

• It takes a while to build wheels. After successfully building torchsparse, you should see the following:

Processing ./torchsparse
  Preparing metadata (setup.py) ... done
Building wheels for collected packages: torchsparse
  Building wheel for torchsparse (setup.py) ... done
  Created wheel for torchsparse: filename=torchsparse-2.0.0b0-cp310-cp310-linux_x86_64.whl size=8113060 sha256=aa5442e7d7b4537b7b18580ba5bd32c1fcb4930c3e0e46c811d4d40275e22610
  Stored in directory: /tmp/pip-ephem-wheel-cache-74ng7icc/wheels/51/5d/42/779ce27f2607ea50a81bd455bbb914023ea7f45a54e2174e0f
Successfully built torchsparse
Installing collected packages: torchsparse
Successfully installed torchsparse-2.0.0b0

3.2 - PyTorch Scatter

conda install pytorch-scatter -c pyg

3.3 - nuScenes devkit

Note: This toolkit is required in order to run experiments on the nuScenes dataset.

pip install nuscenes-devkit