Spotting the Synthetic: Vision Transformers vs Diffusion-Based Image Generation
Abstract
As AI-generated images become more realistic and widespread particularly those produced by diffusion models like DALL·E 3 and MidJourney, there is an increasing need for effective detection methods. This study investigates the potential of Vision Transformers (ViTs) to distinguish between real and diffusion-generated images, addressing a gap in current research that has largely focused on GAN-based detection. Using the SuSy dataset, which contains 25,561 images, a ViT model was fine-tuned for binary classification. A ResNet-50 model was also trained as a CNN baseline for comparison. The ViT achieved strong performance with an accuracy of 96.2%, F1-score of 91.6%, and an AUC of 0.9898, demonstrating that transformer-based models are competitive even in limited data scenarios. While ResNet-50 slightly outperformed the ViT across most metrics, the results support the viability of ViTs in synthetic image detection and highlight their potential for future applications as generative models continue to evolve.
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.