Peer-Reviewed Scientific Publications
A. Kumar, A. Agarwal
An attention-guided, frequency-selective framework that keeps faces visibly intact to humans while rendering them unrecognizable to AI recognition models.
S. Gehlot, A. Kumar, A. Agarwal
Modeling identity in hyperbolic space to close the viewpoint gap between aerial and ground imagery for robust person re-identification.
A. Kumar, A. Agarwal, N. Ratha
A novel adversarial defense mechanism engineered to protect facial biometrics from unauthorized deep learning feature extraction.
M. RA*, A. Kumar*, A. Agarwal
An empirical investigation comparing synthetic vs. real distributions for robust demographic and soft-biometric classification.
R. Jain, A. Agarwal, A. Kumar
Exploring the resilience of foundational vision models when subjected to severe cross-domain shifts utilizing sketch-based modalities.
R. Jain, A. Kumar, A. Agarwal
A generative framework leveraging semantic constraints to synthesize high-fidelity, photorealistic RGB images from structural sketches.
Mohit, A. Kumar, A. Agarwal
Utilizing Generative Adversarial Networks to normalize extreme pose and illumination variations in unconstrained facial imagery.
Officially titled "Method and System for Generating Unlearnable Images Using Non-Linear Multiplicative Perturbations" — a privacy-preserving framework that generates visually faithful yet AI-resistant images to protect identity from unauthorized machine learning models.
Indian Patent No. 522247