Recently, Fang Zhewei, a 2020 undergraduate majoring in Data Science and Big Data Technology from the School of Artificial Intelligence (now a postgraduate student in Computer Science and Technology at Hangzhou Dianzi University), has published important academic findings under the supervision of Professor Zhang Shiqing. Based on his undergraduate graduation project, he completed the research paper entitled PerDepNet: Personality-Guided Cross-Domain Multitask Learning Network for Automated Video Depression Detection as the first author. The paper has been published online in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), an internationally authoritative journal in artificial intelligence. Professor Zhang is the sole corresponding author, and Taizhou University is listed first among the institutional affiliations in the paper. This achievement marks the university’s first paper published in IEEE TPAMI and represents important progress made by the School of Artificial Intelligence in artificial intelligence research.
IEEE TPAMI is a CAS Top journal, ranked in Tier 1 in both the major and minor categories. As the world’s authoritative flagship journal in artificial intelligence and machine vision, it is also a Category A journal, the highest tier recognized by the China Computer Federation (CCF). With a latest impact factor of 20.4, it enjoys an outstanding academic reputation and extensive influence in the international computer science community. The publication further demonstrates the scientific research strength of TU in the field of artificial intelligence.
Systematic scheme of the proposed PerDepNet model
Existing automated video depression detection methods generally suffer from limitations of the scarcity of labeled depressed data and the difficulty in obtaining effective feature representations characterizing human depression. To address this core challenge, this work fully considers the correlation between human personality traits and depression, deeply explores the shared mechanism of nonverbal behavioral cues in videos from the personality and depression domains, and proposes a Personality-guided Cross-domain Multitask Learning Network (PerDepNet) for automated video depression detection. PerDepNet contains a cross-scale shared feature extractor for capturing interactive spatial joint features at different scales, and a mamba-based temporal feature extractor for modeling long-term dynamics of depressed videos in personality and depression domains. On this basis, it integrates the task of predicting personality traits into the depression detection task and adopts an adaptive weighted balance module, enabling the personality traits to guide the estimate of depression levels. Experimental results on multiple public datasets show that the proposed method can effectively improve the performance of video depression detection, achieving leading results on AVEC2013 and LMVD and demonstrating strong competitiveness on AVEC2014.
The work is expected to be applied to remote mental health screening, intelligent health monitoring, and clinical auxiliary assessment, providing technical support for non-contact depression risk assessment under data-scarce conditions.
Link to the original paper:
https://ieeexplore.ieee.org/abstract/document/11673837
