DiffMI recovers face identities from embeddings using a diffusion-driven training-free pipeline with latent initialization, ranked adversarial refinement, and confidence-aware optimization, achieving 84-93% success on resilient models.
Oulu-npu: A mobile face presentation attack database with real-world variations,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
DiffMI: Breaking Face Recognition Privacy via Diffusion-Driven Training-Free Model Inversion
DiffMI recovers face identities from embeddings using a diffusion-driven training-free pipeline with latent initialization, ranked adversarial refinement, and confidence-aware optimization, achieving 84-93% success on resilient models.