We conducted some experiments on the preprocessed video data for the central view, focusing on mainly two scenarios:
Different Users Same Sequences: In this scenario, the model was trained on a particular set of users having all the vocabulary (20 words, 10 phases and 5 sentences). The model was tested on a different set of users having the same vocabulary.
Same Users Same Sequences: In this scenario, the model was trained on 2 samples for a user and the testing was conducted on the third sample. (There were 3 samples for each user for each word/phrase/sentence)
For more details, please refer the preprint.