How to Participate

How to Participate:

1. Complete the Registration Form:
   All participants must officially register for the competition by completing the sign-up form.
   Google Form: [Click here]

2. Download the Dataset:
   Once registered, participants can access the CASED dataset hosted on Zenodo. This repository contains the training videos, test videos, and accompanying metadata.
   Zenodo Link: [Click here]

3. Register on Codabench:
   To submit your model predictions and view the official leaderboard, you must create an account and join the CASED competition on the Codabench.
   Codabench Link:  [Click here]

Tracks:

Participants may compete in one or both tracks.

• Regression Track: Predict the continuous mean engagement score.

• Classification Track: Predict binary engagement labels derived from thresholded engagement scores.

Each track is evaluated independently. Teams participating in both tracks must submit a separate paper for each track.

Evaluation:

Scoring is performed automatically via Docker scoring programs.

• Regression Metrics:

  • Mean Squared Error (MSE)

  • Root Mean Squared Error (RMSE)

  • Mean Absolute Error (MAE)

  • R²

  • Pearson Correlation

  • Concordance Correlation Coefficient (CCC)

• Classification Metrics:

• F1 Macro

• F1 Weighted

• Precision Macro

• Recall Macro

• Matthews Correlation Coefficient (MCC)

Ranking Strategy:

Final ranking is determined using a majority-voting strategy across all metrics.

• Each submission is compared across all metrics.

• Ranking is based on the number of metrics where a method outperforms others.

• Tie-breaking:

  • Regression: RMSE + CCC

  • Classification: F1 Macro + MCC

Note: Public leader board results are provisional. Final rankings are determined using the majority-voting strategy across all metrics. The decisions of the Organizers are final and binding. Participants are encouraged to consider all reported metrics, as optimizing only for the leaderboard metric may not lead to the best final ranking.