Detail of the student project

Topic:Databaze pro odhad věku z videa
Department:Katedra kybernetiky
Supervisor:Ing. Vojtěch Franc, Ph.D.
Announce as:Diplomová práce, Semestrální projekt
Description:The ultimate goal of this project is to create a large database of videos of human faces annotated by biological age of the captured subjects. The work will be split into several subtasks. The first subtask will involve programming scripts for automated collection of movie trailers from the Internet together with personal data (from which the age can be deduced) and photograps of actors staring in the movies. The second subtask will require extraction of face tracks in the dowloaded videos as well as faces from the static photographs of the associated actors. The third subtask will consist of finding a link between the extracted face tracks and the static faces of the actors that will help to connect the actors' personal data to the tracks. The last subtasks will require programming an application allowing for manual assignment of the actors to the extracted face tracks. The manual annotation will be used to estimate accuracy of the automated annotation. The outcome of the thesis will be a database along with the statistics characterizing accuracy of the automatically collected annotation.
Bibliography:[1] Melten Demirkus, Matthew Toews, James J. Clark, and Tal Arbel. Gender classification fromunconstrained video sequences. In proc of. IEEE Computer Society Conference on Computer Visionand Pattern Recognition - Workshops, 2010.
[2] Esam Ghaleb, Makarand Tapaswi, and Ziad Al-Halah. Accio: A data set for face track retrievalin movies across age. In Proc. of ICMR, 2015.
[3] V. Franc, J. Cech. Learning CNNs from Weakly Annotated Facial Images. Journal of Image and Vision Computing, 2018.
Responsible person: Petr Pošík