Reproduce “RiSE: Residual Subspace Expert for Generalizable Text-Centric Image Forgery Localization” 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

Kahim Wong, Kemou Li, Yiming Chen, Haiwei Wu, Jiantao Zhou

RiSE localizes manipulated pixels in document and scene-text images by freezing dominant components of a pretrained vision transformer, training residuals and specialized experts on synthetic DocTamper subsets, then selecting an expert’s mask by prediction stability under latent perturbations. Across four held-out cross-domain sets, RiSE-9 achieves average F1 0.611 versus ADCD-Net’s 0.450; RTM remains difficult, and inference evaluates every expert.

Summary by alphaXiv.

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