Updated Datasets for Cross-cultural Multi-domain Engagement Estimation

We had to realise that the data for the Cross-cultural Multi-domain Engagement Estimation task had several issues.

• Pre-computed features were not properly synchronised with the videos for the NoXI datasets (NoXI, NoXI-J, NoXI additional languages)
• The metadata (e.g language) was incorrect in some cases.
• On the MPIIGroupInteraction, one video was missing in the test set download.

We have corrected the data and updated all downloads. If you work on the Cross-cultural Multi-domain Engagement Estimation task, please re-download the files. We made one further change: To make it easier to compute separate performance scores for NoXI and NoXI-J, we separated these two datasets into individual folder. In this way, your implementation can be harmonised with last year.