Distilled transformers with locally enhanced global representations for face forgery detection

Published: 2025, Last Modified: 22 Jan 2026Pattern Recognit. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A distilled transformer network to capture locally enhanced global forgery traces.•A plug-and-play MAS module to increase the diversity of attention maps.•A novel self-distillation scheme that yields soft labels and combines metric learning.
Loading