Reproduce “WildAIGC-Bench: Can AI Image Detectors Handle the Wild? A Benchmark of Generated and Edited Images in Online Circulation” 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

Zijian Yu, Hangwei Zhang, Jun Lan, Huijia Zhu, Weiqiang Wang, Jianfu Zhang

WildAIGC-Bench evaluates detectors on photographic-looking AI images documented with news events and online claims. Its collection process screens candidates and requires human reviewers to verify evidence before inclusion. On 407 processed images, GLM-5.3-Flash has the highest recall among 16 configurations at 60.93%, leaving over one-third undetected. The set has no real-image controls, so it cannot measure false positives or accuracy.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.