Search papers, labs, and topics across Lattice.
This study reveals that while traditional linear probing on frozen representations from vision foundation models (VFMs) like DINOv3 has set a benchmark for AI-generated image (AIGI) detection, the Perception Encoder (PE) model offers superior potential due to its language-aligned representation that captures high-level provenance semantics. The authors introduce Semantic Prototype Calibration (SPC), a method that enhances the PE's performance by leveraging category prototypes derived from forensic semantic information, leading to improved generalization. As a result, the PE-SPC model outperforms DINOv3 across multiple challenging benchmarks, establishing a new state-of-the-art in AIGI detection.
PE-SPC leverages semantic information to achieve unprecedented generalization in AI-generated image detection, outperforming established benchmarks.
Recent work has shown that a simple linear probe on frozen representations from modern vision foundation models (VFMs) can achieve state-of-the-art AIGI detection performance, substantially outperforming specialized detectors in challenging in-the-wild scenarios. This finding has established DINOv3 as the dominant foundation-model baseline for subsequent improvements. However, we find that the vision-language model Perception Encoder (PE) holds greater potential for AIGI detection, because its language-aligned representation preserves high-level provenance semantics. Specifically, PE exhibits stronger local provenance organization than DINOv3 in its frozen feature space. However, semantic-agnostic linear probing fails to exploit this structure, as PE-Linear still underperforms DINOv3-Linear by 4.1% on In-the-Wild. Based on this observation, we propose Semantic Prototype Calibration (SPC), which constructs category prototypes from forensic semantic information and calibrates them with supervised data. We apply SPC to PE and refer to the resulting detector as PE-SPC. Our analysis shows that this simple design achieves stronger generalization. Across cross-generator, post-processing, and in-the-wild benchmarks, PE-SPC surpasses the previous DINOv3 baseline and achieves new state-of-the-art results.
