Implement a gesture classification model

[gutzcha]

The suggested approach is to use a masked autoencoder (MAE) pre-trained on a facial pose dataset.

There are implementations that we can use:

• MAE
• VideoMAE
• VideoPose3D

• Implement a time dilated CNN model (TDCNN)
• Implement a MAE model with TDCNN embedding and ViT backbone

[gutzcha]

• The models are ready, waiting to be reviewed