Low, High, and Semantic-Level Vision-Language Learning for AI-Generated Image Detection
LHSDet is a new AI image detector that tackles high-resolution images from generative models by treating detection as a Visual Question Answering (VQA) problem. It uses a triple-branch architecture to analyze low-level textures, high-level global features, and semantic text descriptions simultaneously. This approach allows the model to capture critical details missed by standard methods, achieving state-of-the-art accuracy of 95–96% across various AI image types.