Search papers, labs, and topics across Lattice.
This paper introduces GenRes, a novel framework that utilizes generative residual learning through a neural tensor network to enhance the detection of AI-generated images (AIGI) by modeling fine-grained relational features between original and transformed samples. To further improve detection across various generative transformations, the authors present GenRes++, which incorporates a learnable attention mechanism to prioritize the most informative relational cues. Experimental results show that GenRes++ significantly outperforms existing detection methods on multiple benchmark datasets, highlighting its effectiveness in combating the challenges posed by advanced generative AI techniques.
GenRes++ not only detects AI-generated images but also adapts to multiple transformations, outperforming existing methods by focusing on the most informative features.
The rapid advancement of generative AI has enabled the creation of highly realistic deepfake media, posing significant threats, including misinformation, digital identity theft, fraud, and manipulation of public opinion. AI-generated image (AIGI) detection is reliably challenging due to the diversity of generative methods and the subtle artifacts they leave behind. In this work, we propose GenRes, a novel framework for generative residual learning via a neural tensor network, which models fine-grained relational features between original and transformed samples to enhance generalization. To address scenarios involving multiple generative transformations, we introduce GenRes++, which employs a learnable attention mechanism to aggregate relational features across multiple transformed samples and enables the model to focus on the most informative cues. Both models leverage PE-Core as a feature extractor, providing generalized and semantically rich embeddings that improve cross-domain performance and enable the detection of AIGI generated by unseen methods. Comprehensive experiments on multiple benchmark datasets demonstrate that the proposed GenRes++ approach outperforms existing methods.
