Detail of the student project

Topic:Learning CNNs from Weakly annotated facial images
Supervisor:Ing. Vojtěch Franc Ph.D.
Announce as:DP,PTO
Description:Learning Convolutional Neural Networks (CNNs) for face recognition requires large sets of annotated face examples. Manual annotation of faces is laborious work and results are often imprecise. The goal of this project is to implement a system that can learn CNNs from weakly annotated images created by an automated process not requiring humans. As a case study the system will be applied to learn CNN for age and gender recognition from weakly annotated IMDB database that contains 300k faces of movie celebrities. Each image in the database is annotated by the name, biological age and gender of the captured celebrity. However, the images can contain multiple faces and its is not know to which face the annotation should be linked (see example image below). The implemented system should establish the missing link between the images and the annotations.

Instruction:A detailed description of the method will be provided by the supervisor.
Bibliography:- V. Franc, J. Cech. Learning CNNs from Weakly Annotated Facial Images. Image and Vision Computing, 2018.

- V. Franc, J. Cech. Face attribute learning from weakly annotated examples. In Proc. of International Conference on Automatic Face and Gesture Recognition Workshops, Biometrics in the Wild (BWILD), 2017
Max.number of students:3

Warning: the registration to the PTO can be canceled only by supervisor.
Responsible person: Petr Pošík