ADRD: Detecting diffusion-generated images via adversarial perturbation induced reconstruction discrepancy

Yi Zhou, Xiangwei Hu, Jun Tong, Changting Lin, Mei Sun, Shuangxi Chen · 2026

Read ADRD: Detecting diffusion-generated images via adversarial perturbation induced reconstruction discrepancy by Yi Zhou, Xiangwei Hu, Jun Tong, Changting Lin, Mei Sun, Shuangxi Chen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0350655.txt.

Description

Yi Zhou leads a 2026 PLOS ONE paper on spotting images made by diffusion models, the image generators behind many fake pictures. Detectors that read image meaning falter as generators copy real image structure ever more closely. Another family of detectors uses reconstruction error, how badly a model rebuilds an image, but that fixed measure is fragile.

The new method, ADRD, pokes the reconstruction process with a controlled disturbance and watches how the error responds. Real images tended to react more strongly and more variably, while generated images stayed steadier. Tests on several benchmarks suggest this response pattern is a useful detection signal, adding a fresh angle to existing detectors.

Catalog record

CREATORYi Zhou, Xiangwei Hu, Jun Tong, Changting Lin, Mei Sun, Shuangxi Chen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59332
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0350655

Browse the full library · About the Tower of Babel