Reproduce “TGIF2: Extended Text-Guided Inpainting Forgery Dataset & Benchmark” 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
Hannes Mareen, Dimitrios Karageorgiou, Paschalis Giakoumoglou, Peter Lambert, Symeon Papadopoulos, Glenn Van Wallendael
The authors extend the TGIF image-forgery benchmark as TGIF2, adding three FLUX.1 variants and random masks to test localization across fully regenerated images and non-semantic edits. They evaluate fine-tuning and generative super-resolution attacks. For TruFor, mean pixel-level F1 on fully regenerated test subsets rises from 0.14 to 0.75 after combined fine-tuning; this average does not guarantee equal performance across settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.