Applying the Model on the 2021 Dataset

[feenix-ho]

I am currently using the 2020 text based model as the baseline for the 2021 dataset. However, I do not have the accessibility to the 2020 dataset and thus have no idea about the file containing deep captions (since I cannot find the deep_captions.txt file in the folder that the Committee provided).
I want to know whether the 2020 model can run without the deep captions data.
If yes, where can I modify that?
If no, can I do anything to work around?
Thank you.

[jormalaaksonen]

Hello! deep_captions.txt was generated at Aalto University using our pre-existing model and not distributed to other task participants. As you can see reported in https://www.eurecom.fr/publication/6438/download/data-publi-6438.mediaeval20_paper_57.pdf , it did not bring any advancement to the results so you do not lose anything even though you do not have it. To me it seems that you could replace deep_captions with an array of 590 empty strings.