AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

FaceShield: Defending Facial Image against Deepfake Threats

Jaehwan Jeong, Sumin In, Sieun Kim, Hannie Shin, Jongheon Jeong, Sang Ho Yoon, Jaewook Chung, Sangpil Kim

ICCV 2025 · Adversarial robustness and security › Attacks on text and graphs · approach: Diffusion-based adversarial purification · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
target cross-attention condition (key/value) path of DM deepfakes instead of the query-image noising path used by AdvDM/Mist/PhotoGuard/SDSTFaceShield · CelebA-HQ, DiffFaceISM: 0.471 -> 0.243 (vs AdvDM)Improved
add Gaussian blur and low-pass filtering to the noise updateFaceShield · CelebA-HQLPIPS: 0.4214 -> 0.2017 (vs AdvDM)Improved
concentrate noise in low frequencies (low-pass filtering)FaceShield · CelebA-HQfrequency rate (FR): 2.1077 -> 18.4651 (vs AdvDM)Improved
JPEG compress protected images at quality 75FaceShield · CelebA-HQ, IP-AdapterISM: 0.072 -> 0.112Hurt
add ArcFace identity loss L_id (ablation)FaceShield · CelebA-HQ, DiffFaceISM: 0.446 -> 0.243Improved
add face projector loss L_proj (ablation)FaceShield · CelebA-HQ, SimSwapISM: 0.544 -> 0.184 on SimSwap; DiffFace 0.241 -> 0.243Mixed
add attention disruption loss L_attn (ablation)FaceShield · CelebA-HQ, DiffFaceISM: 0.254 -> 0.243Improved
add MTCNN attack loss L_mtcnn (ablation)FaceShield · CelebA-HQ, InfoSwapISM: 0.354 -> 0.237 on InfoSwap; IP-Adapter 0.047 -> 0.072Mixed
combine all four losses instead of any single removed (ablation)FaceShield · six deepfake models, CelebA-HQISMImproved
restrict diffusion attack to UNet mid-layer cross-attention instead of all or up/down layers (ablation)FaceShield · DM deepfake (qualitative)visual protectionImproved
apply FaceShield to GAN-based deepfake modelsSimSwap, InfoSwap · CelebA-HQISM: 0.544 -> 0.184 (SimSwap); 0.431 -> 0.237 (InfoSwap)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy