Vacancy for PhD candidate in Computer Vision Description of Activity
The QUVA Lab, established as part of the Informatics Institute at theUniversity of Amsterdam, invites applications for a PhD position inComputer vision description of activity.
State-of-the-art automatic annotation of images, also from our lab,names objects (bike, road, wedding) in images with reasonable success.Activities (driving, cooking) are an active topic of research whereconsiderable progress is being made. Adjectives (blue, wet) andattributes (open roof, high-heeled) are much harder to grasp form a real challenge. The goal is to compose sharp descriptions of an image, such as ‘a bike is falling on the sand’, as opposed to ‘sand is falling on the bike’.
In this project, we focus on the description of activity from a singleimage, requiring the identification of actor-objects as well as learning from the relationships with acted-upon-objects and their embedding in context. To generate accurate descriptions (‘man stirring soup’, ‘a woman holding a steering wheel’) would be a major step ahead in the automatic description of the content of images. To do so for arbitrary activities is a very challenging part of step-by-step bridging the gap between Natural Language Proces-sing and Computer Vision. What is requested for the project is a good understanding of computer vision and/or deep neural networks both for computer vision and natural language processing. A local, national and international training program will be part of the PhD. The candidate will be part of the QUVA Lab were the target is to perform the scientific research published in the best conferences, and to apply for patents awarded in the patent program.
- MSc in Computer Science, emphasizing computer vision and machine learning;
- Proven research experience, preferably with publication in top-tierinternational conference or journal;
- Solid knowledge of programming in C/C++ for large-scale processing;
- Excellent knowledge of the English language.