Awesome

depression-detection

The final coursework for AI in Mental Health @ PKU.

Prepare the Dataset

We use the D-vlog dataset, proposed in this paper.

Yoon, J., Kang, C., Kim, S., & Han, J. (2022). D-vlog: Multimodal Vlog Dataset for Depression Detection. Proceedings of the AAAI Conference on Artificial Intelligence, 36(11), 12226–12234. https://doi.org/10.1609/aaai.v36i11.21483

Fill in the form at the bottom of the dataset website, and send a request email to the author.

We thanks a lot for the author's kind help with the dataset!

Run the Experiments

Run main.py to train and test the model.

• All the packages used in this project can be installed through conda or pip.
• We implement 3 models.
  • TMeanNet: average over the temporal domain, and then feed the features into a MLP.
  • DepressionDetector: transformer-based model, with cross-modal attention.
    • Yoon, J., Kang, C., Kim, S., & Han, J. (2022). D-vlog: Multimodal Vlog Dataset for Depression Detection. Proceedings of the AAAI Conference on Artificial Intelligence, 36(11), 12226–12234. https://doi.org/10.1609/aaai.v36i11.21483
  • TAMFN: Temporal Conv1d + Temporal attention.
    • Zhou, L., Liu, Z., Shangguan, Z., Yuan, X., Li, Y., & Hu, B. (2023). TAMFN: Time-Aware Attention Multimodal Fusion Network for Depression Detection. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31, 669–679. https://doi.org/10.1109/TNSRE.2022.3224135
• Execute python main.py -h for explanation of the command line arguments.

You need to have your own wandb account. Change these lines in main.py to your own account.

wandb.init(
    project="dvlog", entity="<your-wandb-id>", config=args, name=wandb_run_name,
)

Run the Notebook

In the notebook, we use the Integrated Gradients approach to conduct input attribution.

• Sundararajan, M., Taly, A., & Yan, Q. (2017). Axiomatic Attribution for Deep Networks (arXiv:1703.01365). arXiv. https://doi.org/10.48550/arXiv.1703.01365

Remember to locate your own registered model by chaning the following line:

if not model_path.exists():
    # download models from wandb website
    wandb.init()
    model_path = Path(wandb.use_artifact("<your-model-path>").download())