analiza aiqbench
This paper tackles a practical weakness in AI-generated image detectors: they often fail on images from generators they weren't trained on. The authors report that whitening noise residuals with a deconvolution step improves out-of-distribution detection by 12.6% on average. The result is promising, but it is an arXiv preprint, so its gains need independent replication across a wider range of generators, compression levels, and real-world image sources before anyone relies on it.
Out-of-distribution AI-generated image detection with zero-phase deconvolution for residual whitening
arXiv:2611.00043v1 Announce Type: cross Abstract: Current AI-generated image detection methods struggle in out-of-distribution testing conditions because detectors are usually not robust when evaluated on generators not included in the training set. This work explores a way to mitigate the forensic bias between sources of images -whether real or generated -by removing generator-specific fingerprints. This new training process can be approximated by pre-processing involving a deconvolution operation designed to remove correlated noise patterns inducing loss of generalization. Indeed, we show that whitening the spatial distribution of the image noise residuals can significantly improve the performance of detectors on images coming from generators unseen during training. In particular, an end-to-end CNN model trained on whitened image residuals improves detection by 12.6% on average compared to a training involving original images.
Abstract:Current AI-generated image detection methods struggle in out-of-distribution testing conditions because detectors are usually not robust when evaluated on generators not included in the training set. This work explores a way to mitigate the forensic bias between sources of images -whether real or generated -by removing generator-specific fingerprints. This new training process can be approximated by pre-processing involving a deconvolution operation designed to remove correlated noise patterns inducing loss of generalization. Indeed, we show that whitening the spatial distribution of the image noise residuals can significantly improve the performance of detectors on images coming from generators unseen during training. In particular, an end-to-end CNN model trained on whitened image residuals improves detection by 12.6% on average compared to a training involving original images.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Image and Video Processing (eess.IV)
arXiv:2611.00043 [cs.CV]
(or arXiv:2611.00043v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2611.00043 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 228 cuvinte