installation.md

Getting Started with WildScenes

Prerequisites

WildScenes can be run in any Python environment, and we recommend using a package manager. In this guide we use mamba: https://mamba.readthedocs.io/en/latest/

Mamba is a replacement for conda and behaves the same except is faster especially in larger and more complex environments.

These installation instructions are written for CUDA version 12.1.

Quick Install

For easy installation of the dependencies for running WildScenes, we provide a dockerfile for building a docker image which contains all the required dependencies. To build this docker image run the following command from the docker folder in this repository:

docker build . -t wildscenes

This docker image can then be used for training and testing models using the mmdetection3d and mmsegmentation repositories. See docker/test_mm3d_docker.sh for an example on how to run training and testing scripts using a docker container.

Long Install

mamba create --name wildscenes python=3.10
mamba activate wildscenes

Installation of Dependencies

Step 1: Install CUDA

mamba install cuda -c nvidia

Step 1: Install Pytorch

Using CUDA 12.1:

mamba install pytorch torchvision pytorch-cuda -c pytorch -c nvidia

On CPU only platforms:

mamba install pytorch torchvision cpuonly -c pytorch

Step 2: Wildscenes utilizes mmsegmentation for training 2D networks on our dataset. To install mmsegmentation, the instructions are as follows.

pip install -U openmim
mim install mmengine
mim install "mmcv>=2.0.0rc4"
pip install "mmsegmentation>=1.0.0"

Step 3: Wildscenes utilizes mmdetection3d

mim install "mmdet>=3.0.0"
pip install "mmdet3d>=1.3.0"

Step 4: Install Torchsparse

sudo apt-get install libsparsehash-dev
pip install --upgrade git+https://github.com/mit-han-lab/torchsparse.git@v1.4.0

Step 5: Install other required pip packages. Many of these should already have been installed by default during earlier steps.

pip install opencv-python
pip install keyboard
pip install pynput
pip install tqdm
pip install pillow
pip install tensorboard
pip install matplotlib
pip install open3d==0.18.0
pip install pyntcloud
pip install wand
pip install ftfy
pip install regex

Step 6: Install other required mamba packages (or install using pip)

mamba install numpy -c conda-forge
mamba install pandas -c conda-forge
mamba install scikit-learn -c conda-forge
mamba install quaternion -c conda-forge