October/November Graduation presentations
Tuesday October 29th and Tuesday November 5th we organise the public graduation presentations of Media Technology MSc students.
In 20-25 minutes the graduates present their graduation research projects in English, followed by a 10-minute public discussion.
Everyone is invited to attend!
Tuesday October 29th
Room 1.11 (first floor), Gravensteen Building (Pieterskerkhof 6, Leiden)
14:15 – 14:50 Alessandro Pantó
14:55 – 15:30 Vera van de Seyp
15:35 – 16:15 Pieke Heijmans
16:20 – 16:55 Nadine Roos
Tuesday November 5th
Room 408, Snellius Building (Niels Bohrweg 1, Leiden)
15:00 – 15:35 Matthias König
Titles and abstracts
The titles and abstracts for October 29th can be read here (pdf).
Below are the title and abstract of Matthias König.
Cascaded fine-tuning of deep convolutional neural networks for age estimation from unconstrained facial imagery
Age estimation from facial imagery has been an active research field in the domain of computer vision for many years and various methods have been proposed to encode facial features and map them to age. Those facial feature encodings can be hand-crafted or deep-learned, where the latter relies on Convolutional Neural Networks (CNN) which are able to automatically learn image descriptors from labeled training data. In this work, we present a comparison of different CNN architectures that have previously been applied to the task of age estimation. Furthermore, we propose a multistep fine-tuning procedure during which we train the models on a large number of training examples, while overcoming the commonly faced issue of label noise in large-scale aging datasets. Using this method, we achieve competitive performance on the FG-NET-AD and Adience benchmark datasets.