Skip to content
jiamingzhang94Public

About

[TIFS 2025] MF-CLIP: Leveraging CLIP as Surrogate Models for No-box Adversarial Attacks

Topics

Resources

Stars

5 stars

Watchers

1 watching

Forks

Repository files navigation

Overview

This repository contains the implementation of the proposed MF-CLIP and the baseline methods. The implementation is based on a popular open-source tool, CoOp, which can be found at CoOp GitHub Repository. Follow the steps below to reproduce the results of MF-CLIP.

Instructions

Please refer to DATASET.MD for dataset preparation instructions. After setting up the dataset, run the following scripts in order to reproduce MF-CLIP:

  1. Finetune CLIP with Margin-based Loss

    ./scripts/finetune.sh
  2. Train Generator

    ./scripts/unet.sh
  3. Train Target Models

    ./scripts/train.sh
  4. Evaluate Results

    ./scripts/eval.sh
  5. Optional: Obtain Baseline Method Results

    ./scripts/baseline.sh

About

[TIFS 2025] MF-CLIP: Leveraging CLIP as Surrogate Models for No-box Adversarial Attacks

Topics

Resources

Stars

5 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages