Computer Vision News - April 2018

26 Women in Computer Vision Computer Vision News Mona Jalal Mona, what is your work about? I’ve been working on different projects. One is emotion recognition. The other one is using human computation for text analysis, image segmentation and image analysis. Last semester, I took image and video computing. I also took machine learning, which was half deep learning. Of all these subjects, which one do you like the most? I like emotion recognition the best. I also like combining human computation in furthering the research. How would you create a good data set? How would you decide if your predictions are good? There is a good chance I work on a whole different computer vision topic than what I work now as the first year is for figuring out what you will research later on. Why do you find emotion recognition particularly interesting? It’s a very challenging task. It’s not easy at all, and maybe that’s why it interests me. Even the best algorithms doesn’t have very good accuracy. Even in the EmotiW challenge at ICMI, emotion recognition reaches a far from perfect accuracy. When it comes to single image and single person photo, it has even lower accuracies. We work on it using recurrent neural network or neural architecture. In fact, emotion recognition in videos has a temporal sequence and you want to rely on previous information from a video, like the previous frames, to predict the new frame. Emotion is not just what you see in the face, but also what you do with your body and posture. Another thing is people with different diseases have a different spectrum of showing emotion, and the expression of emotion is different in different people. If someone has Parkinson’s, you cannot detect their emotions very well. Aren’t you afraid that this technology could be used in a way that invades people’s privacy? I think there are a lot of ethics involved in AI. For example, I was going to this deep learning class for self-driving cars at MIT over winter break. Our professor, Dr. Lex Fridman, was talking about the fact that they are actually using emotion recognition for self-driving cars. In that case, actually, it could be a good thing. If a driver is frustrated or maybe drunk and not concentrating, you can detect it and make the car drive autonomously or stop the car. You could Mona Jalal is a first-year Computer Science PhD Fellow in Computer Vision at Boston University and an incoming research intern at NVidia. Her main interests are language and vision, semantic segmentation, gesture recognition and facial analysis for affective computing. Women in Science