Robust Multimedia FOrensic Solutions To Enhance ModeRn MEdia TRustworthiness in Adversarial Conditions
The list of code repositories published during the course of the project is reported below.
Open-Set Attribution of Synthetic Images
https://github.com/lydialy8/openset_attribution_synthetic_images
Siamese-based framework for open-set attribution of synthetic images, enabling verification and source identification even in the presence of unseen generative architectures.
RSAwareATTK
https://github.com/KAI20220922/RSAwareATTK
Implementation of attacks against randomized smoothing classifiers, designed to evaluate and stress-test certifiably robust forensic detectors.
Resynthesis (Stable Diffusion Image Laundering Analysis)
https://github.com/PietroMSB/Resynthesis
Code for analyzing and detecting image laundering through Stable Diffusion autoencoders, including countermeasures for distinguishing pristine, laundered, and fully synthetic images.
Physical-Domain Adversarial Attacks Against License Plate Detection
https://github.com/NischayPurnekar/Physical-Domain-Adversarial-Attacks-Against-License-Plate-Detection
Framework for crafting adversarial examples that remain effective after print-and-capture transformations, targeting real-world license plate detection systems.
JPEG AI Counter-Forensics
https://github.com/polimi-ispl/jpeg-ai-antifor
Experimental framework to analyze the impact of JPEG AI neural compression on deepfake detection and image splicing localization.
Western Blot Synthetic Manipulation Localization
https://github.com/polimi-ispl/western-blot-synthetic-manipulation-localization
Tools for detecting and localizing synthetic manipulations in scientific images, with a focus on Western blot integrity analysis.
Mixture-of-Experts for Speech Deepfake Detection
https://github.com/polimi-ispl/moe_speech_deepfake
Implementation of a Mixture-of-Experts architecture to improve generalization and robustness in speech deepfake detection across diverse datasets.
SAR Counter-Forensics
https://github.com/polimi-ispl/sar-counter-forensics
Codebase for counter-forensic attacks on SAR imagery, simulating re-acquisition processes to conceal local manipulations.
Synthetic Image Detection (Frequency-Based Generalization)
https://github.com/polimi-ispl/synthetic-image-detection
Deepfake detection without deepfakes: training detectors using synthetic frequency pattern injection to enhance generalization to unseen generators.
CNN JPEG Translation Invariance Study
https://github.com/polimi-ispl/cnn_jpeg_tinv
Investigation of translation invariance and shiftability issues in CNN-based forensic detectors, with a focus on JPEG-related robustness.