Skip to content

Latest commit

 

History

History
58 lines (30 loc) · 1.86 KB

README.md

File metadata and controls

58 lines (30 loc) · 1.86 KB

Stealth T-shirt

1. Introduction

Purpose

To raise people's awareness of techniques that can deceive deep learning models, we designed this "Stealth T-shirt" with the adversarial pattern to fool an object detector. The T-shirt is capable of hiding a person who wears it from an open-source object detector.

What is this?

This a T-shirt with adversarial pattern printed on the chest capable of fooling an object detector. The adversarial pattern is generated via known gradients of the model.

How does it work?

By wearing it and showing the adversarial pattern in front of a camera and its object detector behind it, the person who wears it disappears, whereas the person who doesn't wear the T-shirt is still under object detector detection.

Thanks

The attacking method from : http://gixnetwork.org/wp-content/uploads/2018/12/Building-Towards-Invisible-Cloak_compressed.pdf

The object detection code & paper is originally from : http://pjreddie.com/darknet/yolo/

The code we used here is a tensorflow implementation from : https://github.com/gliese581gg/YOLO_tensorflow

2. Install

(1) git clone https://github.com/baidu/AdvBox

(2) Download weight from

Baidu : https://pan.baidu.com/s/13bhF5lTtADdi2DLHE-Qvfg (extraction code: 8dsh)

Google : https://drive.google.com/file/d/0B2JbaJSrWLpza0FtQlc3ejhMTTA/view?usp=sharing

3. Usage

(1) Entering ./applications/StealthTshirt

(2) Run python StealthTshirtDemo.py

(3) Try to hind yourself from object detector. Find the best angle or posture to lower the detection confidence.

4. Requirements

  • TensorFlow
  • Opencv3

5. Changelog

2019/05/24 : First upload!

2019/05/31 : Upload sticker and test gif.