Question about CD-FSS

[Laobyy]

Hi, I want to ask some questions on cross-domain definition.
If you eliminate the training process on PASCAL, can this method still be called "cross-domain" FSS?
Just want to know your point of view when designing this method.

[Vision-Kek]

That's a very good point, and I also faced the issue that it does not really suit the previous cdfss definition. There is a starting point in the discussion section on the source domain conceptualization, and in some of my further work it became more evident that this old cdfss definition leads to more confusion than usefulness with the availability of general-purpose pretrained large models. You don't need to call my method cdfss in your work, you could also see this work as a non-cdfss method that solves the cdfss task. Yet, it is fair to compare with the previous cdfss work since data access in this work is even more restrictive.

[Laobyy]

Ok I got your point, thanks a lot for your answer.