Skip to content

Latest commit

 

History

History
113 lines (89 loc) · 6.71 KB

README.md

File metadata and controls

113 lines (89 loc) · 6.71 KB
Raw

SMDD-Synthetic-Face-Morphing-Attack-Detection-Development-dataset

Official repository of the paper Privacy-friendly Synthetic Data for the Development of Face Morphing Attack Detectors

*** Accepted at CVPR workshops 2022 ***

grafik

Paper available under this LINK

grafik

NOTE: Please note that the SMDD data aims at being a development (training) data for MAD. For an authentic and comprehensive evaluation data we recomend using the MAD22 benchmark of the SYN-MAD-2022 competetion available at its official REPO as well as its extention with Diffusion Autoencoder based mrophs (MorDIFF) also available at its own REPO.

The training data split of the SMDD data can be downloaded from this LINK (please share your name, affiliation, and official email in the request form).

The source images for the morphing attacks of the training split can be dowloaded from this LINK (please share your name, affiliation, and official email in the request form). The morphing pairs follow the file names of the morphing attacks in the training split. Please note that any morphing attacks based on this source data and/or pairing must be (1) made public, (2) shared with the authors of this work, and (3) linked to this repository and share don this repository.

The testing data split of the SMDD data can be downloaded from: LINK (please share your name, affiliation, and official email in the request form)

The source images for the morphing attacks of the evaluation split can be dowloaded from this LINK (please share your name, affiliation, and official email in the request form). The morphing pairs follow the file names of the morphing attacks in the evaluation split. Please note that any morphing attacks based on this source data and/or pairing must be (1) made public, (2) shared with the authors of this work, and (3) linked to this repository and share don this repository.

The pretrained weight of MixFaceNet-MAD model on SMDD training data can be downloaded from this LINK (please share your name, affiliation, and official email in the request form).

Data preparation

Our face data is preprocessed by the face detection and cropping. The implementation can be found in image_preprocess.py file. Moreover, for further training and test, the corresponding CSV files should be generated. The format of the dataset CSV file in our case is:

image_path,label
/image_dir/image_file_1.png, bonafide
/image_dir/image_file_2.png, bonafide
/image_dir/image_file_3.png, attack
/image_dir/image_file_4.png, attack

Experiment

The main.py file can be used for training and test:

  1. When training and test: 
    python main.py \
  --train_csv_path 'train.csv' \
  --test_csv_path 'test.csv' \
  --model_path 'mixfacenet_SMDD.pth' \
  --is_train True \
  --is_test True \
  --output_dir 'output' \
  2. When test by using pretrained weight, first download the model and give the model path: 
    python main.py \
  --test_csv_path 'test.csv' \
  --model_path 'mixfacenet_SMDD.pth' \
  --is_train False \
  --is_test True \
  --output_dir 'output' \

More detailed information can be found in main.py.

Citation:

If you use SMDD dataset, please cite the following paper:

@article{SMDD,
  author    = {Naser Damer and
               C{\'{e}}sar Augusto Fontanillo L{\'{o}}pez and
               Meiling Fang and
               No{\'{e}}mie Spiller and
               Minh Vu Pham and
               Fadi Boutros},
  title     = {Privacy-friendly Synthetic Data for the Development of Face Morphing
               Attack Detectors},
  booktitle = {{IEEE/CVF} Conference on Computer Vision and Pattern Recognition Workshops,
               {CVPR} Workshops 2022, New Orleans, LA, USA, June 19-20, 2022},
  pages     = {1605--1616},
  publisher = {{IEEE}},
  year      = {2022},
  url       = {https://doi.org/10.1109/CVPRW56347.2022.00167},
  doi       = {10.1109/CVPRW56347.2022.00167},
}

If you use the MixFaceNet-MAD, please cite the paper above and the original MixFaceNet paper (repo, paper):

@inproceedings{mixfacenet,
  author    = {Fadi Boutros and
               Naser Damer and
               Meiling Fang and
               Florian Kirchbuchner and
               Arjan Kuijper},
  title     = {MixFaceNets: Extremely Efficient Face Recognition Networks},
  booktitle = {International {IEEE} Joint Conference on Biometrics, {IJCB} 2021,
               Shenzhen, China, August 4-7, 2021},
  pages     = {1--8},
  publisher = {{IEEE}},
  year      = {2021},
  url       = {https://doi.org/10.1109/IJCB52358.2021.9484374},
  doi       = {10.1109/IJCB52358.2021.9484374},
}

The "IJCB-SYN-MAD-2022: Competition on Face Morphing Attack Detection Based on Privacy-aware Synthetic Training Data" (PAPER) is based on the SMDD development data. The IJCB-SYN-MAD-2022 also released a real data evaluation bemchmark that can be found on its REPO.

License:

The dataset, the implementation, or trained models, use is restricted to research purpuses. The use of the dataset or the implementation/trained models for product development or product competetions (incl. NIST FRVT MORPH) is not allowed. This project is licensed under the terms of the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license. Copyright (c) 2020 Fraunhofer Institute for Computer Graphics Research IGD Darmstadt.