메뉴 건너뛰기
소속 기관 / 학교 인증
인증하면 논문, 학술자료 등을  무료로 열람할 수 있어요.
한국대학교, 누리자동차, 시립도서관 등 나의 기관을 확인해보세요
(국내 대학 90% 이상 구독 중)
고객센터 ENG
주제분류

논문 기본 정보

저자정보
(Korea Advanced Institute of Science and Technology) (Korea Advanced Institute of Science and Technology)
저널정보
대한전자공학회 대한전자공학회 학술대회 2025년도 대한전자공학회 하계학술대회 논문집
오류 신고하기

피인용 0

검색

    초록·키워드

    The proliferation of sophisticated deepfakes necessitates robust detection methods, yet current deep learning models often suffer poor generalization to unseen manipulation techniques. This limitation stems from their tendency to overfit to method-specific artifacts within training datasets like FaceForensics++. While complex feature disentanglement methods exist, we propose the simple Method-Bias Suppressor (MBS) framework. Motivated by the observation that standard multitask training for deepfake and method classification surprisingly degrades generalization, MBS utilizes a Gradient Reversal Layer (GRL). This enables an adversarial learning strategy where the model is trained to detect deepfakes while simultaneously being prevented from distinguishing the generation method. By learning method-invariant features, MBS effectively suppresses method-specific bias, leading to superior generalization performance on diverse and unseen deepfake videos.

    최근 본 자료 전체보기

      UCI(KEPA) : I410-151-25-02-093756069