Department of Computer Engineering
CS 590 SEMINAR
Recognition of Personality Traits from Five Factor Model Using Deep Convolutional LSTM Networks
Computer Engineering Department
Five Factor Model (FFM) comprises five personality dimensions as a complete description of personality and it has been influential in incorporating personality traits into artificial intelligence and agent based systems. In this work, we propose an approach to recognize these five apparent personality traits for a given video of a person facing and speaking to a camera. In our method we use a multimodal system to make use of facial, ambient, and audio features where features are learned with separate modality-specific deep neural networks and late fusion is applied to predict the apparent personality traits. We utilize Convolutional Neural Networks (CNNs) to extract facial expression and ambient features per-frame basis, and Long Short Term Memory networks (LSTMs) to integrate the temporal information. The modality-specific regressors for visual and audio features are combined to obtain the estimated personality traits. Our proposed method is evaluated on ChaLearn First Impressions V2 dataset where the personality traits are to be recognized for first impressions.
DATE: 02 April, 2018, Monday, CS590 & CS690 presentations begin at @ 15:40