Zorana Dozdor

I am a computer vision engineer focused on self-supervised learning and human activity recognition, currently conducting research at ZEMRIS laboratory, FER. My work centers on leveraging skeleton-guided methods to improve RGB-based action recognition and online gesture classification.

I have developed novel approaches including SV-data2vec, a latent skeleton-targeted video representation method, and TY-Net, a transformer-based model that predicts gesture classes and boundaries from continuous skeleton input.

My research portfolio includes publications at WACV, VISIGRAPP, and IEEE Access, and I have investigated lightweight CNN solutions and custom loss functions for facial age and gender estimation on embedded systems.

I hold a Master’s degree in Computer Science (2020) from FER, Zagreb, am pursuing doctoral studies in Computer Science (2025) with a focus on self-supervised learning and activity recognition, and I bring practical experience from a machine learning internship at Kraken.