Séminaire de Heydi Méndez Vázquez, directrice de recherche au CENATAV

 

Date: 09/10/2024
Heure : 11h00
Lieu : IC2, Salle des conseils
Intervenante : Heydi Méndez Vázquez
 
 

Lightweight CNNs for Face Recognition Applications

 

Face recognition (FR) is an active research topic in computer vision and image understanding. It is one of the most used and extended biometric techniques. The demands of FR have been growing quickly in recent years due to its extensive applications in video surveillance, law enforcement, access control, marketing and so forth. In practice, however, unconstrained face recognition is affected by many factors such as pose variation, illumination changes, low resolution, and motion blur, resulting in low recognition accuracies.
In the last decade, deep neural networks through multiple layers and massive training data, have reshaped the research landscape of face recognition, due to their increased effectiveness, ability and generalization to learn the essential features of data by constructing powerful representations from the low-level pixels. At the same time, the computational complexity and resource consumption (e.g. large memory and powerful GPUs) of these networks continue to increase, which make them unfeasible to deploy in real-time applications or on resource-limited devices such as mobile and embedded systems. In this talk I will show you the experience ofour group in designing efficient deep networks without significantly lowering the accuracy for different face recognition applications.