第四十八期云上微表情于2024年01月31日晚上7点进行，由中国科学院心理研究所王甦菁老师团队的李婧婷博士主持。此次讲座邀请到来自马来西亚University Malaya的在读博士研究生Gen-Bing Liong同学作主题为“Spot-then-recognize Facial Micro-expressions using Deep Learning”的报告。
Gen-Bing Liong is a Ph.D. student at the Faculty of Computer Science and Information Technology, University Malaya, Kuala Lumpur, Malaysia. He received his B.CS degree from Multimedia University (MMU) Malaysia in 2021. His research interests include machine learning, image processing, and computer vision. Currently, he is actively working on facial micro-expression analysis, image understanding, and image restoration.
报告摘要：
Facial micro-expressions are subtle, involuntary movements that reveal a person’s inner feelings in a split second. To date, very limited research has been attempted on the unified “spot-then-recognize” micro-expression analysis in untrimmed video datasets. Furthermore, despite the recent emergence of multimodal micro-expression datasets, the exploration related to depth dimension has been relatively underexplored.
First, this work proposes MEAN, a micro-expression analysis network characterized by its shallow, multi-stream, and multi-output design for the micro-expression analysis task. Specifically, it consists of three modules: a shared module to extract lower-level features, a spotting module to identify the intervals, and a recognition module to predict the emotion classes. Additionally, the network learning process incorporates inductive transfer learning to preserve learned knowledge.
Second, the depth information in multimodal videos is considered in the scene flow features to estimate changes in 3D facial motion. In line with this, the SFAMNet is presented, utilizing scene flow as network input to conduct micro-expression spotting, recognition, and analysis. To enhance the network learning process, techniques such as data augmentation, attention mechanisms, and end-to-end network optimization are implemented.