Excited to share that three of our papers have been accepted to FG 2026 (International Conference on Automatic Face and Gesture Recognition) 🎉
This year, our work focuses on data generation, quality-aware learning, and interpretability in face recognition systems.
🔹 Jernej Sabadin, Darian Tomašević, Blaž Meden, Peter Peer, Vitomir Štruc: IDSync: Improving diffusion models through identity classification
📄 https://lnkd.in/d2cnEsvh
In this work, we propose a diffusion-based framework for generating synthetic face data with improved identity consistency. By introducing an auxiliary identity classification loss, IDSync produces more reliable identity-preserving samples and enables training recognition models that perform better on real-world benchmarks using purely synthetic data.
🔹 Žiga Babnik, Fadi Boutros, Naser Damer, Deepak K Jain, Peter Peer, Vitomir Štruc: FunFace: Feature Utility and Norm Estimation for Face Recognition
📄 https://lnkd.in/d4gBCqdd
Here, we introduce a new quality-aware loss function for face recognition, combining feature norms with biometric utility (certainty ratio). This allows the model to better adapt to low-quality samples, improving robustness where traditional quality proxies fall short.
🔹 Erdi Sarıtaş, Eren Onaran, Vitomir Štruc, Hazım Kemal EKENEL: Vision-Language Models for Face Image Quality Assessment (FIQA)
📄 https://lnkd.in/dD47ws3t
💻 https://lnkd.in/da84aJJA
We explore zero-shot FIQA using vision-language models, showing they can achieve competitive biometric utility while providing human-interpretable explanations, addressing the limitations of traditional black-box quality estimators.
Looking forward to presenting these at FG 2026!
#FaceRecognition #Biometrics #ComputerVision #DeepLearning #GenerativeModels #ExplainableAI #VisionLanguageModels #FIQA #FG2026
