Advanced search
Start date
Betweenand

Weakly supervised learning for face and person attributes detection

Grant number: 14/24918-0
Support type:Scholarships in Brazil - Doctorate
Effective date (Start): April 01, 2015
Effective date (End): March 31, 2019
Field of knowledge:Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
Cooperation agreement: Coordination of Improvement of Higher Education Personnel (CAPES)
Principal Investigator:Roberto Marcondes Cesar Junior
Grantee:Eric Keiji Tokuda
Home Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil
Associated research grant:11/50761-2 - Models and methods of e-Science for life and agricultural sciences, AP.TEM
Associated scholarship(s):16/12077-6 - Data-driven city model representation and analysis using visual and non-visual information, BE.EP.DR

Abstract

Facial detection and facial attributes detection such as beard and glasses, play a fundamental role in different applications in computer vision. Despite the advances of research in learning of construction, detectors have lower accuracies when the faces are not well positioned or when they present additional attributes or accessories. On the other hand, techniques of deep learning has recently shown exceptional results in several problems. A Convolutional Neural Network learning is a technique in deep learning that is being widely exploited in machine learning tasks. Other classifiers, such as Wavelets Networks, has recently been proposed in the literature. One of the complexities in the utilization of techniques of deep learning consists in getting large training sets labeled. In this scenario, weakly supervised methods represent a viable alternative. In the current project, we aim to develop a method of detecting faces, facial features and accessories such as beard, mustache, glasses and hat. For this, we are going to employ techniques of deep learning in an automatic fashion to create a faces database that includes densely several position, traits and accessories positions. Finally, we are going to analyze the incorporation of newer techniques, such as Wavelets Networks in our system and use our database to develop a robust system for detection of faces and facial attributes. (AU)