I have a question regarding the file:
CiCo/I3D_trainer/misc/phoenix2014T/class.txt
I noticed that the file contains 5383 classes, but I’m not sure how this number was derived. Could you please clarify:
- Where do the 5383 classes come from?
- Are these classes directly extracted from the PHOENIX2014T dataset annotations, or were they generated through a specific preprocessing step?
- If there was a script or preprocessing pipeline that produced this class list, could you point me to it (or share it if it isn’t included in the repo)?
Understanding the source of the class definitions would help me reproduce the preprocessing and ensure consistency with the training setup.
Thanks in advance!