Home

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.

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.

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())