This site contains the source code and dataset to reproduce the defense experiments of the paper: That Doesn’t Go There: Attacks on Shared State in Multi-User Augmented Reality Applications.

Access the code and dataset -> Click on the link below:

https://drive.google.com/drive/folders/11aqGmpVjqdWQ74cBpW3mwy1ouguiZWcj?usp=sharing 

Disclosure to Meta -> Click on the link below:

https://drive.google.com/file/d/10F_Qtv-4YnnaNUDC6Rxszg_NdxSmT6r5/view?usp=sharing 

 Dataset

• The dataset is collected from Microsoft Hololens 2.

• The color images are saved as 640x480 24-bit RGB images in PNG format.

• The depth images are stored as 640x480 16-bit monochrome images in PNG format.

• The images were pre-processed step using camera intrinsics to ensure a 1:1 correspondence between pixels in the color and depth images.

• 15 different scenes are collected, and 300 pairs of color and depth images are used for each scene.

• Real Scene 

• The color and depth images are taken in the real scene. (a) The color images. (b) The depth images.

• Fake Scene

• The color and depth images are taken in the fake scene. (To simulate malicious queries, we collected images in front of screens displaying images of the actual scene.) (a) The color images. (b) The depth images.

Prerequisites

• Please ensure you have the following environments set up with all the dependencies to be able to reproduce the experiments.

PyTorch >= 2.0

Jupyter Notebook

CNN-based Defense

• Please first unzip dataset.zip, then run the following script for defense evaluation results.

• cnn_rgbd.ipynb