Three defenses harden face recognition
Three tricks detect poisoned identities at training and filter inputs at inference.
Stories per day · 30 days
Relatedvision-transformers2attention1backdoor1biometrics1deep-supervision1defense1detection1early-exit1embeddings1fiqa1ghostfacenet1hard-samples1morphing-attacks1mpc1poisoning1privacy1sota1
7 stories, newest first · one card per story
Stories tagged #face-recognition
SD-ViT trains every ViT block as a usable face-recognition exit
Early-exit probes keep over 96% of full-depth accuracy at half the compute without rebuilding the gallery.
ViT-HarDis conditions attention on hard samples for face recognition
ViT-HarDis applies stronger head decorrelation to ambiguous samples, outperforming uniform regularization on IJB-B and TinyFace.
GhostFaceMPC runs face recognition under multi-party computation
Authors quantize GhostFaceNet so extraction, storage, and matching stay encrypted and call the design the first end-to-end MPC system of its kind.
Face Image Quality Improves Morphing Attack Detection
DifFIQA(R) achieves 11.42% mean D-EER on FRLL morphing attacks using fixed quality scores.
Learning Steadily: Accumulating Relative Point Margin Scores for Face Image Quality Assessment
CARPM-FIQA stabilizes face-image quality scores by accumulating relative point margins across training.
Meta Sued Over Training Data for Its AI and Face-Recognition Systems
Illinois and California families sue Meta alleging Facebook and Instagram photos were used without consent to build faceprints and train AI models.
2