Reproduce “Quality-Aware Calibration for AI-Generated Image Detection in the Wild” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Fabrizio Guillaro, Vincenzo De Rosa, Davide Cozzolino, Luisa Verdoliva
Researchers from the University Federico II of Naples developed QuAD, a framework that improves AI-generated image detection by calibrating detector scores based on image quality across multiple near-duplicate versions. This approach yielded an average 8% balanced accuracy gain over simple aggregation on a controlled dataset and achieved 70.3% balanced accuracy on a real-world dataset, demonstrating robustness to online image degradation.
Key insights
- Online post-processing operations significantly obscure forensic traces, leading to inconsistent detection performance for AI-generated images when only a single, often degraded, instance is considered.
- Integrating image quality assessment (IQA) into the detection pipeline allows for intelligent calibration of detector scores, effectively weighting the reliability of each near-duplicate instance.
- Aggregating quality-aware calibrated scores from multiple near-duplicate images provides a more robust and accurate decision than analyzing single instances or naive score aggregation, even when dealing with low-quality copies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.