Loss during self-supervised training stage about ts-tcc HOT 5 CLOSED

And is there any reason why you haven't used positional embedding?

from ts-tcc.

I can't generalize, but for me, I used not to consider the SSL loss as a quality measure, but the linear evaluation or the fine-tuning as you mentioned .. this also applies for some other methods I experienced, not only TS-TCC.

from ts-tcc.

Ok Thanks :)

from ts-tcc.

It didn't help to improve the results (probably because the way it is implemented is not suitable with different datasets we adopted, and exploring different PE techniques was out of our scope). However, you can test it yourself with the your datasets.

from ts-tcc.

ok thanks for clarifying

from ts-tcc.