🧑 Human Understanding

💬 ACL2026 · 1 paper notes

ForgeryTalker: Generating Attribution Reports for Manipulated Facial Images

The paper proposes the Forgery Attribution Report Generation task, constructs the MMTT dataset with 152,217 samples (the first large-scale facial forgery dataset providing both pixel-level masks and human-written text descriptions), and introduces the ForgeryTalker end-to-end baseline that jointly generates localization masks and attribution reports via a shared encoder and dual decoders (mask + language model), achieving 59.3 CIDEr and 73.67 IoU.