This codebase is tested on Ubuntu 20.04.2 LTS with python 3.8. Follow the below steps to create environment and install dependencies.
- Setup conda environment (recommended).
# Create a conda environment
conda create -y -n pomp python=3.8
# Activate the environment
conda activate pomp
# Install torch and torchvision
# Please refer to https://pytorch.org/ if you need a different cuda version
pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html- Clone POMP code repository and install requirements
# Clone clip-openness code base
git clone https://github.com/amazon-science/prompt-pretraining.git
cd prompt-pretraining/
# Install requirements
pip install -r requirements.txt
# Update setuptools package
pip install setuptools==59.5.0- Install dassl library.
cd third_party/Dassl.pytorch/
# Install dependencies
pip install -r requirements.txt
# Install this library (no need to re-build if the source code is modified)
python setup.py develop
cd ../..To apply the pre-trained POMP prompt to semantic segmentation and object detection, please refer to ZSSeg and Detic codebases, respectively.