Regression in video transformers

[PabloMikiMarti]

Thank you so much for these resources!

I have a question though, would it be possible to fine tune a video transformer, namely TimeSformer, to do regression instead of classification?
I see several things wouldn't fit out of the box as the label folders, the trainer_factory, etc.
I assume a work around would be binning the variable I want to regress and predict to which bin each observation belongs...
I would really appreciate some insight into the steps I should follow