Improving an existing action spotting pipeline in videos through multi-modal neural networks Analyzing video understanding and audio processing methods for a real-world application.

Research of deep learning algorithms for visual geo-localization, which is the task of finding the location where a given photo was taken. Development of a large-scale software to perform visual geo-localization, which integrates state-of-the-art methods from a number of computer vision and machine learning fields, such as image classification, semantic segmentation and efficient nearest neighbor search.