Encapsulating Knowledge in One Prompt

Qi Li Runpeng Yu Xinchao Wang^†

LV-Lab, National University of Singapore  ^†corresponding author

Installation & Preparation

1. Clone the repo and prepare the virtual environment.

git clone https://github.com/LiQiiiii/Encapsulating-Knowledge-In-One-Prompt.git

cd Encapsulating-Knowledge-In-One-Prompt

conda create -n kiop python=3.10.0

conda activate kiop

pip install -r requirements.txt

2. Prepare the dataset and models. You can use your own models and dataset. For quick start, we provide several models and datasets, which can be downloaded directly from google drive:

gdown https://drive.google.com/uc?id=19o2EItRw-LOJUdjDf-mOz0zh0QalF8wj

gdown https://drive.google.com/uc?id=18XDK2fdhCQuwGm4sJntfSvESpbZEv1bY

unzip KiOP_models.zip

unzip KiOP_data.zip

Training & Evaluation

We provide several scripts in ./scripts. For example, for running KiOP-B, you may use the KiOP_B.sh as follows. You can adjust the hyperparameters in the shell file to customize your setup:

sh ./scripts/KiOP_B.sh

Citation

If you finding our work interesting or helpful to you, please cite as follows:

@inproceedings{li2024encapsulating,
  title={Encapsulating Knowledge in One Prompt},
  author={Li, Qi and Yu, Runpeng and Wang, Xinchao},
  booktitle={European Conference on Computer Vision},
  pages={215--232},
  year={2024},
  organization={Springer}
}

Acknowledgements

This implementation is built on top of the code from ILM-VP and CMI. We would like to express our gratitude to the authors of these repositories.