Adversarial examples - One pixel attack

Repository for the B.Sc. thesis about adversarial examples in particular One pixel attack, this attack was proposed in 2018 by Jiawei Su et al.

B.Sc. thesis paper

---

How to try the jupyter notebook

Downloading precompiled .h5 files

• First of all we need to create the folder structure and download needed files for the project.

  • Inside your root directory which contains the .ipynb file create a folder structure as Data/Saves, inside Saves introduce the next three files.
    • Model (10.5 MB)
    • Training data (1.01 GB)
    • Test data (666.1 MB)

• Execute requirements.txt as pip install -r requirements.txt

• Install protobuf version 3.6.0 using pip since tensorflow requires protobuf version 3.7.1 and won't let install protobuff==3.6.0 from requirements.txt. Newer versions (>=3.7.0) don't work properly with NVIDIA GPUs at least my computer configuration. This is not critical but it will probably return an error inside the first notebook box which checks for CPU/GPUs.

  pip install protobuf==3.6.0

• Launch jupyter jupyter notebook or jupyter notebook Traffic-signals-cnn.ipynb

Downloading images and training a new

• Inside your root directory which contains the .ipynb file create a folder structure as Data/Dataset/Training and Data/Dataset/Test

• Download dataset files

  • Dataset