SITCON2021-DL-Adversarial-Attack-Tutorial

An simple tutorial for generating adversarial examples against any DL model

Quick start

Step 1: Download imagenette2 dataset

To simplify the experiment, we only use a subset of ImageNet here. The imagenette2 dataset contains ten categories from the original ImageNet.

./getData.sh

Step 2: Fine-tune the pretrained model

Please fine-tune the model to fit the subset with ten categories.

python3 train.py

Step 3: Generate and visualize adversarial examples

Please open the Generate.ipynb and run it to generate adversarial examples