Skip to content

This repository is an implementation of the paper ModelGiF: Gradient Fields for Model Functional Distance (ICCV2023)

Notifications You must be signed in to change notification settings

zju-vipa/modelgif

Repository files navigation

[ICCV 2023] ModelGiF

This repository is an implementation of the paper ModelGiF: Gradient Fields for Model Functional Distance. We provide theoretical insights into the proposed ModelGiF for model functional distance, and validate the effectiveness of the proposed ModelGiF with a suite of testbeds, including task relatedness estimation, intellectual property protection, and model unlearning verification. For more details, please read the paper.

An illustrative diagram of the overall pipeline of obtaining ModelGiF curves.

Dependencies

Our code is implemented and tested on PyTorch. Following packages are used:

torch==1.12.1+cu116
numpy==1.23.1
captum==0.5.0
opencv-python==4.6.0.66
pandas
scikit-learn
scikit-image
xlsxwriter
lime
saliency
h5py
tqdm
pillow

You can install pytorch with following command:

pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116

Then, install the rest of dependencies:

pip install -r requirements.txt

Experiments

Task Relatedness Estimation
All the experimental details about task relatedness estimation are provided in the task_relatedness folder.
Affinity Matrix Figure
Intellectual Property Protection Model Unlearning Verification
This folder contains the experimental details about intellectual property protection with ModelGiF. All the experimental details about model unlearning verification are provided in the model_unlearning folder.
Comparison between the proposed ModelGiF and existing methods for IP proctection. Cosine distances between the reference classifier Cref and unrelated classifier Cunrelated, the directly unlearned classifier Cdirect and the approximately unlearned classifier Capprox.

Citation

@InProceedings{Song_2023_ICCV,
    author    = {Song, Jie and Xu, Zhengqi and Wu, Sai and Chen, Gang and Song, Mingli},
    title     = {ModelGiF: Gradient Fields for Model Functional Distance},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023}
}

Contact

If you have any question, please feel free to contact

Jie Song, sjie@zju.edu.cn;

Zhengqi Xu, xuzhengqi@zju.edu.cn.

About

This repository is an implementation of the paper ModelGiF: Gradient Fields for Model Functional Distance (ICCV2023)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published