Awards on ‘National Scientific Students’ Associations Conference’ - Eftimiu Nikomidisz Jorgosz June 7, 2021
What is your project about? What results did you showcase at the conference?
The aim of my project was to colourize face images captured with a depth camera using neural networks. In a nutshell, this means taking 3D "snapshots" of faces and converting them into photographs in the more classical sense.
A primary focus of my research was comparing how different architectures and techniques affected the quality and realism of the generated colourizations. In the end, my results clearly showed that relatively small and "shallow" networks can be made just as suitable for this task as the currently popular deep neural networks, with just a few "tricks". Because of this, the method presented in my research could be used to generate realistic (if a bit blurred) images in real time.
What practical applications does this have? Where could these developments be used and what for? (How could „laymen" encounter your work in everyday life?)
This method isn't really something laymen could encounter in everyday life, as its primary application would be as a part of image processing pipelines, which makes it more "behind the scenes", so to speak. That said, the possibilities are endless, as depth colourization would make it possible for any currently available face-recognition software to use depth data. In addition to helping accelerate human-machine interface development, this could also enable face recognition in complete darkness, for instance.
How long have you been working in this field? How did you find out about it, and what sparked your interest?
I found this topic while I was looking for a tutored research project during my bachelor's program. I did some research into the applications of depth cameras as a part of human-machine interface development, and eventually started working at the ELRN Research Centre for Natural Sciences to develop my idea. The greatest motivation for me was that, while image colourization is an established field with many published papers, almost nobody had tried colourizing depth images before me.
How did you choose a supervisor? How would you describe the experience of working together?
I met my supervisor, Dr. Gergely Márton, the project lead for the National Bionics Program's multimodal human-machine interfaces subprogram, during a tutored research project exhibition. I was looking for a topic where I could make full use of my C++ and OpenCL knowledge, and one of Dr. Márton's colleagues recommended us to each other. After a short interview, it became obvious that we would work very well together.
I have always found the experience of working together to be very pleasant: I am usually given full freedom to pursue whatever avenues of research I see fit, but I always have someone to turn to if I need guidance or academic advice.
Why do you think, what is the secret to a successful conference paper?
This is a difficult question for me, since there are no two identical projects, and what makes a "good" project can depend on a lot of things – even the chosen field and specific topic. Despite that, I can safely say that quality work always requires a great deal of time and effort – only the very best research can achieve true success. That said, investing a good amount of effort is often enough in and of itself. Naturally, there are many factors to consider, from scientific relevance to novelty, but a well-founded and properly developed project is sure to attract much attention.
What future plans do you have for your research? Do you plan on developing this project further, or have you chosen a different topic?
Actually, I've been working on a similar, but markedly different topic for a while now. It's been a year since I handed in my original work, and much has happened since then – much of it of my own volition. Despite this, I still work in human-machine interface development and image processing, and machine learning is a tool I use very often in my research.
Of course, this doesn't mean that I don't plan on continuing this line of research, far from it. One day, I might return to this topic, or choose a very similar one when presented with the opportunity.