PAVNet: A Personality-Aware Audio-Visual Fusion Network for Depression Detection

This repository contains the code, checkpoints, experimental results, and environment configuration for the paper:

S. Liu, Y. Huang and X. Yuan, "PAVNet: A Personality-Aware Audio-Visual Fusion Network for Depression Detection," in 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Wuhan, China, 2025, pp. 5868-5875.
DOI: 10.1109/BIBM66473.2025.11356059

Overview

Depression detection is an important topic in computational mental health. This project proposes PAVNet, a Personality-Aware Audio-Visual Fusion Network for depression detection. The model integrates multimodal information, including audio-visual representations and personality-related cues, to improve robustness and prediction performance.

This repository provides:

• training and testing scripts;
• core implementation files;
• model checkpoints and logs from k-fold experiments;
• prediction results and evaluation scores for different feature combinations;
• environment configuration for reproducibility.

Repository Structure

BIBM-2025/
├── environment.yaml          # Conda environment configuration file
├── train.sh                  # Shell script for starting model training
├── test.sh                   # Shell script for running model evaluation
├── train.py                  # Main training code
├── test.py                   # Main testing code
├── kfold_checkpoints/        # Model weights and training logs for k-fold experiments
└── answer_Track2/            # Prediction results and evaluation scores on the testing platform