whitebox-adversarial-toolbox 0.1.1

WHite-box Adversarial Toolbox (WHAT) - Python Library for Deep Learning Security

WHite-box Adversarial Toolbox (WHAT)

A Python Library for Deep Learning Security that focuses on Real-time White-box Attacks.

Installation

pip install whitebox-adversarial-toolbox

Usage (CLI)

Usage: what [OPTIONS] COMMAND [ARGS]...

  The CLI tool for WHitebox-box Adversarial Toolbox (what).

Options:
  --help  Show this message and exit.

Commands:
  attack   Manage Attacks
  example  Manage Examples
  model    Manage Deep Learning Models

Useful commands:

# List supported models
$ what model list

# List supported Attacks
$ what attack list

# List available examples
$ what example list

Available models:

[x] 1 : YOLOv3      (    Darknet    )   Object Detection        YOLOv3 pretrained on MS COCO dataset.
[x] 2 : YOLOv3      (   Mobilenet   )   Object Detection        YOLOv3 pretrained on MS COCO dataset.
[x] 3 : YOLOv3 Tiny (    Darknet    )   Object Detection        YOLOv3 Tiny pretrained on MS COCO dataset.
[x] 4 : YOLOv3 Tiny (   MobileNet   )   Object Detection        YOLOv3 Tiny pretrained on MS COCO dataset.
[x] 5 : YOLOv4      (    Darknet    )   Object Detection        YOLOv4 pretrained on MS COCO dataset.
[x] 6 : YOLOv4 Tiny (    Darknet    )   Object Detection        YOLOv4 Tiny pretrained on MS COCO dataset.
[x] 7 : SSD         ( MobileNet  v1 )   Object Detection        SSD pretrained on VOC-2012 dataset.
[x] 8 : SSD         ( MobileNet  v2 )   Object Detection        SSD pretrained on VOC-2012 dataset.
[x] 9 : FasterRCNN  (     VGG16     )   Object Detection        Faster-RCNN pretrained on VOC-2012 dataset.

A Man-in-the-Middle Hardware Attack

The Universal Adversarial Perturbation (UAP) can be deployed using a Man-in-the-Middle Hardware Attack.

[ Talk ] [ Video ] [ Paper ] [ Code ]

The Man-in-the-Middle Attack consists of two steps:

• Step 1: Generating the perturbation.
• Step 2: Deploying the perturbation.