Skip to content

install.md

Latest commit

 

History

History
48 lines (30 loc) · 1.5 KB

install.md

File metadata and controls

48 lines (30 loc) · 1.5 KB
Raw

Install

Foundation models

For foundation model including DINO, CLIP, MAE the code will handle the download of the models, as for segment-anything model (SAM), you will need to download it mannually from SAM offical repo.

Place it under ./data/SAM/, or you can change the dir defination in the code.

Python environment

We recommend using conda for environment setup:

conda create -n forla python=3.10.12
conda activate forla

Then install PyTorch which is compatible with your cuda setting. In our experiments, we use PyTorch 2.6.0 and CUDA 12.6:

conda install pytorch==2.6.0 torchvision==12.6 torchaudio==2.6.0 cudatoolkit=12.6 -c pytorch -c conda-forge
pip install pytorch-lightning==2.5.1 torchmetrics==1.7.1

This will automatically install packages necessary for the project. Additional packages are listed as follows:

pip install pycocotools scikit-image lpips chardet omegaconf
pip install pytorch-fid einops transformers

Optional:

We use visdom for visualization:

pip install visdom

Possible Issues

  • In case you encounter any environmental issues (e.g., package version compatibility), you can check out the env file exported from our server requirements.txt. You can use this file to install specific version of a package.