Investigator: Stefan Winkler.

We analyse emotions from facial expressions and other cues. We provide valence/arousal representation of emotions (continuous, 2D). The system was trained & tested with dataset of 100,000+ images, created from psychophysical validation studies achieving error: 7-8% (RMSE) and >90% accuracy. On the One-Minute Gradual (OMG) Emotion Behavior Challenge 2018, we ranked 1st for vision-only arousal/ valence prediction, and 2nd for overall valence prediction.

For commercialization opportunities, please contact Opsis.

References:

• Peng, L. Zhang, Y. Ban, M. Fang, S. Winkler. A deep network for arousal-valence emotion prediction with acoustic-visual cues. eprint arXiv:1805.00638, May 2018.
• Peng, L. Zhang, S. Winkler, M. Winslett. Give me one portrait image, I will tell you your personality. Proc. ACM Multimedia Conference, Seoul, Korea, Oct. 22-26, 2018.