A modular framework for experimenting with score-level fusion across multiple data sources. This project supports flexible combinations of modalities, normalization strategies, and fusion rules, with built-in evaluation using ROC curves and verification metrics (e.g., TMR@FMR).
The system is designed for reproducible experimentation across heterogeneous datasets, enabling analysis of how different data sources contribute to overall performance.
Note: A cleaner, more recent object orientated implementation is available in the V2022-MajorRefactor branch.
The first version of this application is released. For a full description of this app, how to use it and how to adapt it, consult the user's manual (https://github.com/melissadale/ScoreFusionApp/blob/master/Score_Fusion_App_Documentation.pdf) or one of the wikis listed in the Documentation section below.
There are various ways to acquire the Score Fusion App. Exe executables are given in the Release directory, if the Exe does not execute, we recommend dowloading from source. A sh script is provided in the Utilities directory that describes the necessary steps.
- Executable: https://github.com/melissadale/ScoreFusionApp/tree/master/Releases.
- From Source: https://github.com/melissadale/ScoreFusionApp/blob/master/Utilities/ScoreFusion.sh
The application provides an end-to-end pipeline for loading score data, applying fusion strategies, and analyzing performance across verification and identification tasks.
The first tab allows the user to upload the score files, specify the training testing split, normalization techniques, and how to handle missing data:
The second tab provides information about the scores in the dataset, including the number of genuine/imposter scores, detected modalities, and the samples sizes in the train, test, and entire score dataset. Additionally, it provides a PDF and histograms of the genuine and imposter scores within each modality:
The third tab allows the user to select which fusion rules to apply, specify if verification and/or identification tasks should be anayized, and give a name to the experiment:
The fourth tab displays the results of the fusion experiments to the user:
- Input File Formats This describes the formats and files allowed by the application.
- Errors, debugging, and feature requests