DeeCLIP: A Robust and Generalizable Transformer-Based Framework for Detecting AI-Generated Images
M. Keita, W. Hamidouche, H. Bougueffa Eutamene, A. Taleb-Ahmed, A. Hadid
DeeCLIP integrates CLIP-ViT with LoRA fine-tuning and a custom fusion module (DeeFuser) to robustly detect AI-generated images across unknown generative models and common degradations.
REVEAL: A Retrieval-Augmented Generation Approach for Contextual Identification of Synthetic Visual Content
M. Keita, W. Hamidouche, H. Bougueffa Eutamene, A. Taleb-Ahmed, A. Hadid
REVEAL leverages visual RAG and majority voting to overcome the generalization limits of existing synthetic image detectors against unseen generative models and degraded conditions.
Harnessing the Power of Large Vision Language Models for Synthetic Image Detection
M. Keita, W. Hamidouche, H. Bougueffa Eutamene, A. Taleb-Ahmed
Explores LVLMs as zero-shot and few-shot detectors of synthetic images, demonstrating strong generalization across GAN and diffusion-based generators without task-specific training.
Bi-LORA: A Vision-Language Approach for Synthetic Image Detection
M. Keita, W. Hamidouche, H. Bougueffa Eutamene, A. Taleb-Ahmed
Bi-LORA combines VLMs with low-rank adaptation (LoRA) tuning to enhance precision on unseen model-generated images, addressing the generalization gap in binary synthetic image classification.
FIDAVL: Fake Image Detection and Attribution Using Vision-Language Model
M. Keita, W. Hamidouche, H. Bougueffa Eutamene, A. Taleb-Ahmed, A. Hadid
FIDAVL goes beyond detection to attribute fake images to their generative source, leveraging VLM representations for joint detection and attribution in a unified framework.
Visual Security Evaluation of Perceptually Encrypted Images Based on Multi-Task Learning
S. A. Fezza, M. Keita, W. Hamidouche, O. Déforges
VSMML proposes a no-reference visual security metric for perceptually encrypted images using multi-output deep learning, outperforming state-of-the-art on visual security and quality tasks.