AI Image Detection Using an Ensemble of Efficientnet, Resnet and Vision Transformer
Abstract
Modern generative models, such as GANs and diffusion models, produce highly realistic synthetic images that are virtually indistinguishable from authentic photographs, posing significant risks in forensics, journalism, and data verification. To address the need for a fast and accessible detection tool, this paper proposes an ensemble approach combining EfficientNet-B0, ResNet-50, and ViT-B/16 architectures. Fine-tuned via transfer learning on the CIFAKE and AI-vs-Real datasets, the ensemble leverages CNNs for local texture anomaly detection and the Vision Transformer for global patch dependencies. Evaluated on a 2,000-image test set using soft voting, the model achieved an accuracy of 89.60%, precision of 92.94%, recall of 85.70%, an F1-score of 89.17%, and an AUC-ROC of 0.9641. The notably high precision minimizes false accusations of tampering. The system is deployed as a scalable web application utilizing a microservices architecture (React, Node.js, FastAPI, PostgreSQL, Docker).