Reproduce “WildFakeToolBench: Benchmarking AI-Generated Image Detectors with Web-Sourced Image Processing” 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

Mingjie Li, Yukun Jiang, Yicong Tan, Yang Zhang

WildFakeToolBench converts public image-editing advice into 156 validated CPU-run tools and tests them on generated images classified as fake by six detectors. Tool construction and parameter proposals use no detector feedback. On 105 images, its strongest single tool achieved 45.54% attack success rate, versus 6.93% for six selected processing baselines; these results apply to the curated test set and protocol.

Summary by alphaXiv.

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