Installation

We recommend using Conda to manage the environment.

1. Create a new python environment.

We tested our code with python version=3.10. Other versions may be compatible.

conda create -n rris python=3.10
conda activate rris

2. Install pytorch

Our repo is built upon pytorch=1.12.1, torchvision=0.13.1, torchaudio=0.12.1, and cudatoolkit=11.6.0. Other versions may be compatible.

conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge

3. Install mmcv

In this repo, we use mmcv-full=1.7.1. 2.x versions are not compatible with this repo. Other installation methods can be found in mmcv official repo.

pip install -U openmim
mim install mmcv-full==1.7.1

Then evaluate the installation.

python .dev_scripts/check_installation.py

4. Install other dependencies

Install pycocotools locally from this repo.

cd ..
git clone git@github.com:cocodataset/cocoapi.git
cd cocoapi/PythonAPI
pip install -e .

Then install other dependencies from requirements.txt.

pip install -r requirements.txt