Linkage-Attack

Code accompanying Linkage Attack on Skeleton-based Motion Visualization CIKM'23

Attacking Models

Contains the attacking models. Split into the Random Forest and the SGN-based attacking models.

Defense Models

Contains the classical Motion Retargeting anonymizer model. The other anonymizer models utilized are naive adaptations from the following repos: UNET/ResNet, Deep MR

Linkage Attack

Contains the SGN-based (Main) and RF (Baseline) versions of the Linkage Attack proposed in this paper.

Skeleton Info

Contains code used to parse and process the .skeleton files provided by the NTU paper here. Raw data is available here

Setup

All contributions were written inside Jupyter Notebooks but have been exported to standalone Python files. Details on installing Jupyter can be found Here.

Specifics on how to run each file are located inside the README in each relevant directory.

Data

All pickle files containing the preprocessed data and saved models are located inside the README file inside each relevant directory. Additionally, the parent folder can be viewed Here.

Acknowledgements

This work was supported in part by UNC Charlotte startup fund and NSF grant 1840080.

This code relies on multiple previous works as listed below:

NTU RGB+D (Code) (Paper (60)) (Paper (120))

SGN (Code) (Paper)

Skeleton-anonymization (Code) (Paper)

2D MR (Code) (Paper)

Deep MR (Code) (Paper)