Privacy-Preserving Image Features

(via Adversarial Affine Subspace Embedding)

🚧 This repository is still work in progress.

It contains a cleaned-up reimplementation of the following paper:

"Privacy-Preserving Image Features via Adversarial Affine Subspace Embeddings".
M. Dusmanu, J.L. Schönberger, S.N. Sinha, and M. Pollefeys. CVPR 2021.

[Paper on arXiv]

Requirements

OpenMP and Eigen must be installed prior to the setup of this repository. CUDA is also a recommended dependency for best performance.

1. Start by creating and activating the base python environment:

conda env create --file=env.yml; conda activate ppif

2. Install the pyppif bindings:

cd py_ppif; pip install .; cd ..

3. [Optional] Install the pyppifcuda bindings:

cd py_ppif_cuda; pip install .; cd ..

HPatches Sequences evaluation

Start by installing COLMAP (used for SIFT feature extraction) and set the env variable COLMAP_PATH to the COLMAP executable, e.g.:

export COLMAP_PATH=~/sources/colmap/build/src/exe/colmap

1. Download the dataset:

bash download_hpatches_sequences.sh

2. Extract SIFT / DoG + HardNet features:

python feature-utils/extract_sift.py --dataset_path data/hpatches-sequences-release/ --colmap_path $COLMAP_PATH
python feature-utils/extract_hardnet.py --dataset_path data/hpatches-sequences-release/

3. Run the evaluation:

python evaluate_hpatches_sequences.py --dataset_path data/hpatches-sequences-release/ --descriptor sift
python evaluate_hpatches_sequences.py --dataset_path data/hpatches-sequences-release/ --descriptor hardnet

Privacy attacks

Coming soon...

BibTeX

If you use this code in your project, please cite the following paper:

@InProceedings{Dusmanu2021Privacy,
    author = "Dusmanu, Mihai and Sch\"onberger, Johannes L. and Sinha, Sudipta N. and Pollefeys, Marc",
    title = "{P}rivacy-{P}reserving {I}mage {F}eatures via {A}dversarial {A}ffine {S}ubspace {E}mbeddings",
    booktitle = "Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition",
    year = "2021"
}