Priv_NF

Adam Taras 2022 Work done under Donald G. Dansereau as part of the Robotic Imaging Group, The University of Sydney

This is a git repo for source code for the second half of my thesis, with an aim of quantifying the privacy of our proposed methods.

Requirements

Python 3.9

matplotlib==3.5.3
numpy==1.23.2
opencv_python==4.6.0.66
pytorch_lightning==1.7.7
scipy==1.9.0
seaborn==0.12.1
torchvision==0.13.1+cu113

Structure

The folders for this project are:

• NF: (Normalising Flows) contains an adaptation of the UvA tutes for training and evaluating the fidelity of flow models.
• hashes: an implementation of random circle and random line extrema
• optimization_attack: a least squares based optimization attack on the hashes
• tests: unittsts on base functionality
• data_manipulation: misc tools for transforming LSUN bedroom data into a consistent form

Pre trained models

Pre-trained models can be found here