Welcome to the official website of the Revelio framework!
Revelio framework is explicitly designed to simplify the development and the comparison of MAD algorithms.
The key features of the proposed Revelio framework are the following:
- Modularity: Revelio is a framework based on different modules explicitly designed to offer a shared development platform, simplifying the integration of new components and methods. It is designed for training and evaluation of new Single-image Morphing Attack Detection (SMAD) methods.
- Reproducibility: the experimental evaluation is carried out only on public datasets or morphed images that can be downloaded or generated through publicly released source datasets and morphing algorithms. Given the same data, an experiment is fully reproducible using the same configuration file!
- Flexibility: the proposed framework is customizable in a simple and straightforward manner through a human-readable configuration file. New modules and functionalities can be added through an effective plugin system.
Resources:
-
- Official Documentation
- Source code (GitHub page)
Results:
A list of the main results obtained through the framework:
-
- Best S-MAD method on the SOTAMD dataset [FVC-onGoing]
- Extensive and reproducible cross-dataset evaluation of Single-image Morphing Attack Detectors (S-MAD).
In particular, the evaluation has investigated the use of the following elements:- Face Detectors (MTCNN, Viola&Jones and DLib)
- CNN architectures (ResNet-50, Inception Resnet V1 and ViT)
- Data augmentation techniques (including JPEG compression and P&S simulation procedure)
- Forensic features (PRNU, Wavelets and Fourier)
- Different combinations of training datasets
- Introduction of the Weighted Average Error across Datasets (WAED) metric for the MAD task to summarize and simplify
the comparison of diverse approaches across different testing datasets. This metric can be used even with only a single dataset, weighting the traditional error-based metrics used in the MAD task (EER, BPCER@APCER).
Reference:
Borghi G., Di Domenico N., Franco A., Ferrara M., Maltoni D. “Revelio: a Modular and Effective Framework for Reproducible Training and Evaluation of Morphing Attack Detectors” (IEEE Access 2023) [IEEE Xplore Full Paper in open access].
Bibtex reference:
@article{borghi2023revelio, title={Revelio: a Modular and Effective Framework for Reproducible Training and Evaluation of Morphing Attack Detectors}, author={Borghi, Guido and Di Domenico, Nicol{\`o} and Franco, Annalisa and Ferrara, Matteo and Maltoni, Davide}, journal={IEEE Access}, year={2023}, publisher={IEEE} }
This project is part of the European H2020 Project “image Manipulation Attack Resolving Solutions” (iMARS) [website].