Fourth-year PhD candidate at Quva Lab, Informatics Institute
(IvI), the University of Amsterdam (UvA). I am supervised by
Professor Arnold Smeulders
and co-supervised by Assistant Professor Efstratios Gavves.
My research interests are in computer vision using methods of machine learning and deep learning. I conduct(ed) research on action recognition, zero-shot event detection and graph-based video storytelling. I would like to help in advancing methods for understanding human activities in videos at industrial scale.
I started the PhD studies in January 2016. I obtained MSc Artificial Intelligence from the School of Electronics and Computer Science at the University of Southampton in 2015 and BSc Computer and System Engineering from the Faculty of Engineering, Ain Shams University in 2012.
A novel temporal layer for 3D CNNs with multi-scale temporal convolutions and much reduced computation. The result is a CNN for modeling minute-long complex actions, 8 times longer than best related method.
A manifold is learned using contrastive loss, in which there is a joint embedding of videos of human events and their related articles. The result is end-to-end model with best results on TRECVID MED dataset.
Summary for our method used in the annual competition for zero-shot event recognition TRECVID MED 2016.
The goal of this on-going research is to detect the surrounding boats using multi-modal sensori data from cameras and radars. This will help achieve the over-arching goal of autonomus water taxi.
Using knownlege graphs as a priori for labels associated with event images or videos. Hence, improving the accuracy of zero-shot detection.
An method to predict missing labels of restaurant images using semantic word similarities. The method makes use of dataset of Kaggle competition: "Yelp Restaurant Photo Classification".