Reproduce “MIG-COW: Transferable Adversarial Attacks on Deepfake Detectors via Gradient Decomposition” 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
Wonjune Seo, Joonhyuk Baek, Yeseong Jung, Saerom Park
MIG-COW attacks deepfake detectors by perturbing generated fake faces so they are classified as real. It combines consensus and orthogonal directions from integrated gradients across known surrogate models, then updates images within a perturbation budget. With ViT-P excluded, its average white-box attack success rate was 99.96%, versus MIG’s 93.48%; average black-box success was 7.16% on two unseen targets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.