Adversarial training with ResNet-50

Researching influence of adversarial training on AI model performance

To Run

• Clone this repository
• Download data from Imagenet2012(https://www.image-net.org)
• In the root directory, run python -m venv .venv to create a virtual environment.
• Activate the virtual environment by running source .venv/bin/activate
• Run pip install -r requirements.txt -q
• In VS code, select the kernel as the venv
• In the root directory, create a datasets folder and add the downloaded data from Imagenet. Here is an example of the datasets folder and data it contains: https://drive.google.com/drive/folders/1Y56Ta5sxSWJ7MANHLwgmH3ch4Sq3sG8a?usp=share_link
• Run each of the notebooks in the order below:-
  • data_processing.ipynb - Preprocess the data, shows the dataset class distribution and generates adversarial data
  • resnet_50_base_metrics - Evaluates the base resnet-50 against adversarial and standard test dataset
  • adversarial_model_training - Use adversarial data to retrain the model, generating a robust model as the result
  • performance_evaluation - Evaluates performance of adversarially trained resnet-50 model.