PRA-Net: Part-and-Relation Attention Network for depression recognition from facial expression

Area of Science:

• Artificial Intelligence
• Psychiatric Disorders
• Computer Vision

Background:

• Automated depression recognition often relies on facial expressions as indicators of psychiatric disorders.
• Current methods face limitations due to insufficient feature representations for depression detection.
• Facial expression analysis is a key area for objective mental health assessment.

Purpose of the Study:

• To propose a novel Part-and-Relation Attention Network (PRA-Net) for enhanced depression recognition.
• To improve the representation of depression-related features from facial expressions.
• To achieve state-of-the-art performance in depression detection using AI.

Main Methods:

• Partitioning feature maps to obtain semantically rich part features.
• Employing self-attention to weigh individual part features.
• Utilizing relation attention to refine feature weights based on global context.
• Aggregating weighted features for compact, depression-informative representations.

Main Results:

• The proposed PRA-Net significantly enhances depression representations.
• Extensive experiments validate the superiority of the PRA-Net method.
• Achieved state-of-the-art performance on AVEC2013 and AVEC2014 datasets compared to other end-to-end methods.

Conclusions:

• PRA-Net offers a superior approach to depression recognition through enhanced feature representation.
• The method effectively integrates part and relation attention for improved accuracy.
• This work advances the application of artificial intelligence in objective mental health diagnostics.